Abstract
Real-time sensors for precision irrigation schedulating are used for enhancing water efficiency and optimizing resource usage. Poor resource management can negatively impact traditional farming practices, particularly in regions limited by water shortages. Agriculture is susceptible due to its heavy reliance on water resources. Due to global warming and its potential impacts, there is a growing emphasis on developing strategies to ensure a steady water supply for food production and consumption. As a result, research on reducing water usage in irrigation systems needs to be implemented. While traditional commercial irrigation sensors are often too expensive for smaller farms to adopt, manufacturers are now producing affordable alternatives that can be integrated with network systems to provide cost-effective solutions for efficient irrigation and agricultural monitoring. To minimize a farmer’s efforts, an Internet of Things (IoT)-based drip irrigation system is proposed in this work. Initially, the required data is collected using the IoT sensors. The gathered data is fed into the Adaptive Residual Hybrid network (ARHN) that is developed by using the Spatial Autoencoder and Stacked CapsNet. Here, the Modernized Random Variable-based Frilled Lizard Optimization (MRV-FLO) is utilized to tune the ARHN parameters. Therefore, the required water from the pump for the crops is provided by the ARHN model. In addition, this model makes the work simpler and avoids the wastage of water in the agricultural environment. Finally, the performance of the developed framework is validated over the existing works to prove the efficiency of the recommended method. The main experimental findings of the developed model achieve 99.24% and 97.32% in terms of accuracy and RMSE. Moreover, the statistical findings of the developed model shows 41.9%, 34.9%, 36.0% and 37.1% better performance than LEA-ARHN, FDA-ARHN, AOA-ARHN and FLO-ARHN in terms of best measure. Based on this performance enhancement, the developed model can effectively reduces the farmer’s effort and improves the crop productivity in the agricultural sectors.
Keywords: Smart drip irrigation system, Internet of Things, Adaptive residual hybrid network, Spatial autoencoder, Stacked CapsNet, Modernized random variable-based frilled Lizard optimization
Subject terms: Engineering, Optics and photonics, Physics
Introduction
Agriculture plays a crucial role in many countries as it contributes significantly for industrial engagement, economic development, environmental sustainability, food production, and the use of renewable energy1. A large amount of freshwater usage in arid and semi-arid countries is contributed to agricultural irrigation. Thus, agricultural sectors are facing issues as a result of more water demands coupled with expanding populations. Real-time smart irrigation solutions are required to reduce agricultural water consumption2. Smart irrigation can reduce farming expenses and negative effects on the environment in addition to increase crop growth and water productivity3. Smart irrigation is an important component of recent agricultural fields in shaping the productivity and sustainability of farming systems4. However, agricultural production faces numerous challenges that could impact the productivity, quality, and growth of crops. These challenges are to be result of issues like soil degradation, pests and diseases, chemical use, climate change, and limited water supply5. Among these, irrigation is an important factor as it directly influences productivity and plant growth1. In order to improve crop yield and growth, reliable irrigation techniques are essential. Additionally, monitoring environmental factors such as soil moisture, humidity, temperature, light intensity, wind speed, and rainfall are vital for optimizing plant development and productivity6.
Smart farming is an incorporation of technological advancements with agriculture that enables farmers to monitor plant growth in real-time without frequent field visits7. Insufficient irrigation can lead to water stress in plants, with symptoms like wilting and yellowing of leaves8. Water shortages can also damage soil structure, hinder nutrient dissolution, and increase vulnerability to pests and diseases9. On the other hand, excessive irrigation increases the danger of bacterial and fungal infections, causes nutrient loss, and deprives roots of oxygen10. Smart farming and smart irrigation procedures are accomplished by the use of IoT technologies11. Further, these procedures are recognized as organic farming fields and merge aquaculture with hydroponics. Thus, it can ultimately optimize the water usage for equally distributing hydroponics through fertilizer12. Additionally, smart irrigation is a cutting-edge watering method that made possible by network sensors, gadgets as well as remote controls. Farmers can automate their processes and have real-time visibility into the irrigation process using IoT13.
Diverse studies have employed the use of smart farming systems that integrate IoT technologies. While these systems have been advanced in laboratory settings, it often fails to demonstrate improvements in water conservation or agricultural productivity14. Additionally, in systems utilizing drip irrigation under mulching films, water stress detection methods can damage the film, which limits their large-scale applicability15. However, traditional detection algorithms struggle to extract meaningful information from images due to the similarity of features across different experimental processes16. Deep learning has proven as an effective mechanism in crop phenotypic research, particularly in detecting water stress. But, many deep learning algorithms alongside visible light images to identify water stress in plants are computationally complex and require more resources17. Even though deep learning techniques are powerful in feature learning and fitting, it requires significant computational power that makes the decision-making process related to irrigation challenging18. To address these issues, an efficient deep learning mechanism to assist in smart drip irrigation is developed.
Motivation of the developed method
Food production agriculture is a major source of livelihood for the population. The irrigation process plays an important role in providing a certain amount of water to the crops using pipelines and sprinklers for maximizing food production and reducing the stress experienced by farmers. In order to capture sustainable production and a healthier rate of foods, the freshwater constraint is significant in the irrigation process. Considering the existing method, it needs additional parameters to continuously monitor and capture sufficient and high-quality data for the training process. The energy consumption is extremely high due to the continuous utilization of sensing devices19. It has a limited deployment size for monitoring stations due to the wiring connection range and greater configuration20. It needs more energy for validating the soil moisture to gradually impact on environment. It is complex for gathering data from the prescribed sensors and it needs automation to manage the land for irrigation21. It takes more training time for the training process and does not utilize the optimization performance to reduce the overall accuracy value and enhance the overfitting issues. Thus, it affects the water demand, irrigation handling, and planning performance to provide minimal adaptable solutions22. It is impossible to track every corner of large areas in the training datasets and limited to train with tiny-size datasets. In order to solve these issues, an efficient IoT-based smart drip irrigation method is developed in this research work.
The primary contributions of the implemented framework are outlined below:
To design an IoT-based adaptive irrigation system using deep learning to manage water usage in agricultural environments by integrating real-time data of IoT sensors. The developed approach accurately predicts and controls the amount of water to be delivered to plants for minimizing water waste and enhancing irrigation efficiency.
To design an ARHN model, which combines a spatial autoencoder and stacked CapsNet for advanced feature extraction and pattern recognition from the input IoT data. By identifying intricate environmental patterns and analyzing spatial correlations in the data, the proposed ARHN enhances irrigation decisions. Residual connections can improve model accuracy and adaptability by ensuring efficient training and preventing the loss of important information.
To integrate the MRV-FLO for optimizing the parameters in ARHN model. This promotes for optimally ensuring that the irrigation system adapts to various environmental conditions and optimizes water usage. This optimization can reduce the False Omission Rate (FOR) and increase the Matthews Correlation Coefficient (MCC) rate.
The designed framework is organized as follows. In Phase II, a comprehensive review of the literature works are provided. Phase III discusses the importance of IoT in precision farming, including the details of the dataset used in this work. Phase IV focuses on the basic structural description of the deep learning networks employed for the smart irrigation process. Phase V describes the development of precision farming through the integration of the hybrid deep learning network. This section elaborates the proposed ARHN model and the MRV-FLO algorithm. Phase VI showcases the outcomes of the implemented framework. Phase VII concludes the paper with a summary of the developed approach.
Literature survey
Related works
IoT in smart irrigation
The drip irrigation systems based on IoT have been adopted to increase crop productivity and water use efficiency23. The low-power sensor nodes were used by the system to control the water delivery and analyze the environmental conditions. By altering water administration and varying the irrigation times based on soil moisture levels, field tests were carried out to determine the ideal water requirements. The work24 has developed a moisture sensor and EvapoTranspiration (ET)-based approach for an IoT-based Soil Moisture Monitoring (IoT-SMM) system, especially the study has been conducted in sweet corn. Commercial production should decide on the optimal irrigation scheduling plan to maintain the soil water content. An IoT enabled Deep Learning Enabled Smart Irrigation System (IoTDL-SIS) method has been developed with intelligent irrigation strategies for efficient water use with fewer human involvements25. The suggested IoTDL-SIS technique used separate sensors, specifically humidity, air temperature and soil moisture for the data compilation process. Arduino mechanism received the sensor data and forwarded it to the cloud server for additional processing. Then the data analysis process was carried out on the cloud server. First, a regression-based technique was used to forecast soil and environmental characteristics. Second, the clustering technique was applied to these estimated results. Third, the classification process was carried out by Deep Belief Network (DBN) model. Furthermore, a LoRa-based machine learning model has been leveraged to track and schedule accurate irrigation via the IoT26. Using the data gathered from the soil moisture sensors, the authors have created a separate irrigation method that would give the eggplant and tomato and the exact water amount was needed. Further, an intelligent system that plans for fertilization and irrigation based on plant requirements and uses automatic watering to monitor and maintain the appropriate level of soil moisture. Solar fustigation was a precision irrigation system that uses photovoltaic solar energy along with IoT technology27. Different characteristics in the agricultural field were tracked by the developed system. In various sectors also, the IoT has been investigated with bio and medical informatics for comprehensively evaluating and synthesizing medical and bioinformatics issues28. Considering the findings, several articles were validated by utilizing features like scalability, latency, sensitivity, adaptability, F1-score, accuracy, and specificity.
Deep learning models in smart irrigation system
Recently, modern technologies have been employed for the smart irrigation management system29. This innovative model was capable of guaranteeing smooth data transfer, safe communication, and effective user engagement. Moreover, this model was used in avoiding over-irrigation systems. In addition to this, a model has implemented a deep learning model to forecast soil moisture and schedule irrigation using a Deep Convolution Neural Network (DCNN) for precision farming30. In order to maintain water stability and crop growth, this model optimized the level of water effectively. This system helped to forecast irrigation planning based on the crop needs by using a variety of sensor and parameter modeling approaches. Additionally, soil moisture, humidity, and temperature were some of the anticipated characteristics considered in this work. Introducing a new method has able to forecast the water levels in cotton crop fields. A dataset with 5200 images, while the sub-membrane drip irrigation used in the fields was collected. For training and testing, five deep learning networks were assessed using cross-validation. The MobilenetV3 model31 performed better than alternative deep learning architectures, according to experimental results. These findings have highlighted the efficiency of using deep learning models for precisely determining water levels in cotton fields with various drip irrigation. To improve data integrity and storage for smart decision-making across different Internet of Drones (IoDs). Also, decentralized predictive analytics was introduced for significantly utilizing and contributing to Deep Learning (DL) frameworks32. A novel method has been developed to capture the availability and dependability of long-term issues to enhance the performance of network33. The findings of the developed method showed that the better performance to greatly handle faults and minimize the risk of damage. In34, analyzed the taxonomy of DL method applications for climate change mitigation has improved. The climate changes of have been split into several customizable groups, especially in agriculture applications.
Optimization algorithm-based research works
An environmental-based user’s demands in Quality of Service (QoS) attributes to discover the preferable atomic services for constructing the required composite service in an orchestration model. This analysis suggested a service composition method with the help of MapReduce framework and Grey Wolf Optimization (GWO)35 to efficiently validate the overall performance. An algorithm to construct an optimal spanning tree by combining an artificial bee colony, genetic operators, and density correlation degree to make suitable trees36. The overall results demonstrate the improved data collection consistency has attained by the developed model than the existing methods. An investigation of nature-inspired algorithms in IoT-based healthcare to recognize gaps37 in standardized evaluation metrics for providing insights and directions for future research in this domain, expanding the diverse landscape of nature-inspired algorithms in healthcare. This analysis could supply to an inclusive considerate of nature-inspired methods in the various landscapes of IoT-based services. A novel framework has leveraged to enhance the optimization process and amplify the capabilities of the blockchain-based Industrial Internet of Things (IIoT) with the help of the Glowworm Swarm Optimization (GSO) algorithm38. It has the ability to accurately capture relevant data and ensure decision-making performance.
In39, a novel taxonomy was introduced to classify the ChatGPT components accompanied by a chatbots definition, which contains a hybrid, generative, rule, and retrieval-based chatbots. Also,40 have explored botnets and their evolution covering aspects such as life cycles, Command and Control (C&C) models, botnet communication protocols, detection methods, the unique environments botnets operate, and strategies to evade detection tools to covert channels for detection and the reinforcement. With the aim of effectively providing guidance on addressing security concerns in evasion and detection methods, this analysis was highly utilized. A pioneering approach utilizing a novel fuzzy method based on Multicriteria Decision-Making (MCDM)41 for prioritizing vehicles to select optimal neighbors in data broadcast. Also, a fuzzy MCDM-based Re-Broadcasting scheme (FMRBS) was developed to minimize broadcast storms and enhance the effectiveness of data distribution. It has the ability to reduce end-to-end latency and overhead while enhancing Packet Delivery Ratio (PDR)42. In43, the author has analyzed Alzheimer’s disease (AD) identification to validate the insightful information. This analysis could examine model stability security and data protection issues44. In45, the author have analyzed and captured cybersecurity threats, fraud in finance, diagnosing and monitoring healthcare, biometric authentication, personalized marketing, and optimizing the supply chain in systems to provide better decisions in an effective manner. In46, the author have analyzed the fourth wave of human life of a cyber-age which was connected to everyone in any place at any time. This analysis could help to minimize human intervention in the internet of things.
Problem statement
IoT-aided drip irrigation networks play a major role in the field of cultivation by reducing water utilization as well as increasing crop productivity. But, it faces certain drawbacks due to the limited internet resources, which affect the efficiency of the IoT applications, since a stable internet connection is essential for ensuring accurate data transmission. Additionally, various approaches struggle to process huge volumes of information attained from the multiple sensors within the network. In order to address these shortcomings, it is crucial to introduce a novel framework for smart drip irrigation.
Recently, numerous investigations were performed to overcome the challenges faced by the classical techniques and their research gaps are listed below.
Considering the existing LoRa framework23, it struggles to handle huge volumes of sensor information to affect the training process for minimizing the overall irrigation performance. The generalizability of the classical IoT-SM24 model is minimized due to the inability in managing the complicated dataset. These issues can be resolved by the implemented adaptive residual hybrid network to improve the better solutions in complex issues. Also, it can significantly enhance the generalization ability. In specific cases, the effectiveness of the standard DSVM25 model is influenced by to occurrence of vanishing gradients. The residual connection in the implemented method can minimize the problems of exploding or vanishing gradients. Thus, it can accurately predict the irrigation level for the crops for improving the positive outcomes. In specific cases, the conventional HTTP29 model struggles to train the network in huge and imbalanced databases. Thus, employing an optimization model efficiently mitigates this issue to prevent overfitting and provide highly generalized outcomes. The designed framework can optimally manage a wide range of datasets within a limited duration, leading to better prediction. In the existing DCNN method30, it does not utilize optimization performance for tuning the hyperparameters to maximize the inaccurate outcomes, leading to poor generalizability. However, the developed method utilizes the optimization process to efficiently tune the hidden neurons and learning rate parameters to enhance the positive outcomes. By fine-tuning these parameters, the model aims to enhance the efficiency and accuracy of the irrigation method, resulting in better and more reliable solutions. This process is crucial for attaining successful outcomes to continuously monitor and manage large land space. Further, the conventional fuzzy logic31 method does not utilize the standard database to collect the prescribed data in the agricultural field, thus it enhances the overfitting issues. An optimal database is utilized in this research work to effectively collect better quality images to enhance the overall performance for reducing the negative solutions. The strengths and challenges of classic techniques are given in Table 1.
Table 1.
Advantages and disadvantages of classical techniques.
| Author [citation] | Methodology | Advantages | Disadvantages | Solution given by the proposed model |
|---|---|---|---|---|
| Inayah et al.23 | LoRa |
It permits the communication of sensor information in the lack amount of internet By employing the sensor nodes this model guarantees the continuous function during the planting season |
It faces challenges in processing the information from various sensors and offering stable communication in different surroundings | The developed method is capable of continuously collecting different information from various sensors |
| Kumar et al.24 | IoT-SM | This model is capable of identifying the position, depth and soil moisture | The performance of the network is affected due to the complexity of data analysis | It has minimal complexity to enhance the prediction performance |
| Suresh25 | DSVM | This model presented superior accuracy in the cultivation field by reducing water consumption | This method doesn’t offer effective performance and it takes more training time for the monitoring performance | The proposed framework has minimal training time for validating the overall irrigation performance |
| Morchid et al.29 | HTTP | This model ensures effective data flow, and secure communication according to the environmental conditions | This model does not employ an advanced tuning approach that is essential to improve crop yield as well as water security | The optimization process is performed to accurately tune the hyperparameters |
| Kumar et al.30 | DCNN, GRU | This model can determine the moisture content within the soil and suggest a plan for precision agriculture | This technique doesn’t transmit the data to the recommender network | It is effectively transmitting the trained data to the next processor |
| Sujono et al.31 | fuzzy logic | This model enhances the network feasibility by determining the cotton’s water status during the process of drip irrigation | This approach cannot be applied to perform actual irrigation decisions and assist the irrigation procedures | It utilizes adaptive mechanisms to generate better decisions in different environmental changes |
| Ahmad et al.27 | decision support system | This approach enables the cultivators to manage and analyze the fertilization and irrigation techniques automatically | This model lacks the ability to evaluate the quality of irrigation rapidly and effectively | The proposed method can truthfully validate the irrigation quality with the help of prescribed performance measures |
| Lakshmi et al.26 | LSTM | This technique has the capacity to transfer the soil condition of the weather stations | This model is not capable of processing weather information including the direction of the wind, rain, and the percentage of the cloud | It can collect and validate the temperature, heat, air, soil moisture, and water from the diverse sensors to accurately provide weather information |
| Vakili et al. 35 | GWO |
This technique assists in improving the efficacy of water utilization by implementing irrigation at the right time It has ultimately utilized the available data |
This framework faces the feature extraction process issue | The residual connection of the proposed method can significantly extract the applicable features in the prediction process |
| Heidari et al.32 | DL | It reduces the utilization of water resources and boosts the productivity of the crop by monitoring the water within the soil | It has minimal resolution for weather and soil health issues | It provides better resolution outcomes on soil health and weather situations |
| Zanbouri et al.38 | GSO | It generates promising results and it can be applied to several agricultural mechanisms | It requires additional larger time yield data in the training process | The implemented approach can be significantly trained with large and tiny size datasets |
| Heidari et al.41 | fuzzy method | It is an effective framework to improve crop yield production | This technique failed to identify the water usage details | It can ultimately recognize and validate the soil moisture, water usage and water wastage information in an effective manner |
Methods
Overview and Use of IoT in Precision Farming
Precision agriculture has the development of various applications that help farmers with well-timed, real-time details regarding crop health and field conditions. In precision agriculture, the use of IoT as well as sensor nodes is remarkable. These nodes collect valuable data in making knowledgeable decisions of crop management and resource optimization.
Typically, precision agriculture contains three primary stages:
Data collection (Phase 1): The first stage involves numerous sensors and IoT nodes, which monitor physical and environmental factors of plant health, moisture levels, and soil conditions. The moisture content is optimally captured with the help of the soil moisture sensor, while a soil nutrient sensor measures fertility levels.
Data transmission and storage (Phase 2): Once data is gathered, it needs to be transmitted for further processing. The gathered data can either be sent to the cloud for remote monitoring and examination, or it may be stored locally with a nearby fog node. This stage ensures that the information is stored for further analysis to make any decisions.
Analytics and response (Phase 3): The third stage involves in analyzing the collected data to assess the condition of the crop fields. This information supports the farmers to take necessary actions, such as activating an irrigation system or applying fertilizers to increase crop growth. This process ensures that timely action is taken to prevent the crop from any damage. The integration of IoT in precision agriculture optimizes resource use and reduces the operational costs. By providing an accurate estimate of crop yield, as well as precise measurements for water, fertilizer, and other resources, this technology not only reduces costs but also enhances overall productivity. Irrigation is one of the most resource-intensive aspects of farming and it benefits greatly from precision agriculture. By analyzing factors like rainfall and soil moisture levels, farmers can determine the ideal time to irrigate, significantly improving water efficiency.
Developed IoT-Based Smart Drip Irrigation System
This work proposes an innovative IoT-based drip irrigation system designed to optimize water usage in agricultural fields for enhancing crop yield. This approach is used for farmers to make better decisions in irrigation schedules and water distribution. The environmental factors details of soil moisture levels, humidity, and temperature, are accumulated from IoT taken from the standard database. An ARHN model is developed and it is used to process the gathered data. This approach is developed by the incorporation of two models, the Stacked CapsNet and Spatial Autoencoder to improve detection precision and decision-making for smart irrigation. This model focuses on critical aspects of the input data and it adapts with varying conditions and improves crop yield through automated irrigation processes. Effective communication between capsules is made possible by stacked CapsNet’s dynamic routing procedure, which guarantees that only the most pertinent features of input data are sent to higher layers. Predictions about water requirements become more precise as a result of this selected feature dissemination. Complex spatial and spectral properties are efficiently extracted from the input data by spatial autoencoders. The complex relationships between data are efficiently learned by the residual connection in both networks and this connection is used to manage the gradient flow. Additionally, the implementation of the MRV-FLO technique fine-tunes the model’s parameters, ensuring precise irrigation requirements are met while minimizing water wastage. This model’s capacity to analyze real-time data not only saves water but also raises the general productivity of farming methods. The system is a useful tool for resource management and sustainable farming as it incorporates with machine learning algorithms, which allow it to learn with past data to forecast future irrigation requirements. In the end, the developed strategy is compared with conventional techniques to verify its sustainability and efficiency. The architectural demonstration of the implemented IoT-based smart drip irrigation model is exhibited in Fig. 1.
Fig. 1.
Architectural demonstration of the developed IoT-based smart drip irrigation model.
Code availability
The custom code and mathematical algorithms for generating the results in IoT-based smart drip irrigation system have been deposited in the GitHub repository and also it is publicly available at https://github.com/shabanaurooj08/Automated-Smart-Drip-Irrigation-System-/tree/main.
Experimental setup
The implementation of the developed automated IoT-based smart drip irrigation system was carried out using Python. The population count and iteration count were set to be 10 and 50, respectively, with a chromosome length of 4. The performance of the implemented framework was compared with the existing irrigation systems to validate the effectiveness of the integrated optimization algorithms and neural networks in precision farming. The optimization algorithm’s performance was compared with the Lotus Effect Optimization Algorithm (LEA)47, Flow Direction Algorithm (FDA)48, Artificial Orcas Algorithm (AOA)49 and FLO50. The irrigation process of the suggested model was compared with DSVM25, DCNN30, and LSTM26. Table 2 represents the developed method’s system requirements. Here, the table illustrates the software and hardware specifications in the development system. Considering software requirements, the version specifies Python version 3.11 and Anaconda version 3. The software ‘Pycharm’ indicates the required Integrated Development Environment (IDE), which likely represents the research study’s code written in it. Hardware requirements, on the other hand, the Windows operating system are widely accessible and compatible platform of the system. Also, the different libraries, such as Keras 3.11.3, TensorFlow 2.19.0, TFLearn v0.3.2, Matplotlib 3.10.5, NumPy 2.3.2 and PrettyTable 3.16.0are utilized in the implementation process. The machine learning libraries plays the crucial role for simplifying the implementation in complex models. Additionally, the utilization of Keras, TensorFlow and TFLearn is widely applied in the machine learning and deep learning frameworks, whereas it suggest the smart drip irrigation likely adopts the AI for efficient decision-making outcome. The libraries like Matplotlib and NumPy is used for the data manipulation, visualization and analysis that provides significant outcome for processing the sensor data and provides valuable insights. Then, the PrettyTable is adopted for the data needs to be well-formatted in the table, potentially displays the visualization outcomes. The overview and features of each library are listed below:
Table 2.
System requirements of the designed framework.
| Version | 3.11 and Anaconda version 3 |
|---|---|
| Software requirements | |
| Software | Pycharam |
| Hardware requirements | |
| Machine | Windows |
| ROM | 500 GB |
| Processor | Intel core i3 processor model |
| Libraries |
Keras 3.11.3, TensorFlow 2.19.0, TFLearn v0.3.2, Matplotlib 3.10.5, NumPy 2.3.2, and PrettyTable 3.16.0 |
| Version | 11 |
| RAM | 8 GB |
Keras: A high-level and user-friendly interface for efficiently training the neural networks is Keras. In order to generate complex neural network architectures, the Keras library provides the wide range of pre-built layers, loss and activation functions. It also supports seamlessly integrate with other libraries. In various platforms, the Keras supports in transfer learning and model deployment.
TensorFlow: It is an open-source deep learning approach, which is developed by Google for effectively deploying the machine learning models at various scales. It supports in handling the complex computations and distributed computing. By utilizing the TensorFlow, it is a versatile choice for diverse machine learning frameworks.
TFLearn: It is also an open-source deep learning model, which is built on the top of the TensorFlow. Thus, it enables faster and easier experimental analysis to train the models. Also, it enables user-friendly framework, in which the TFLearn model maintains full transparency and compatibility for effectively understanding the model’s architecture and training process.
Matplotlib: For data visualization, the Matplotlib is an effective library in Python. It suggests the wide range of functions and classes in order to create the various types of plots like line plots and bar plots. This can extensively evaluates and visualize the model’s performance, which aids in interpreting and exploring the data efficiently. The Matplotlib library has the ability to integrate with other libraries also for better data visualization.
NumPy: In Python programming language, the NumPy is the fundamental library, which supports large and multi-dimensional arrays as well as matrices effectively. It can be effectively applied in various applications like natural language processing, image processing and data analysis. Here, the NumPy array evaluates an image, whereas it enables operations like resizing, filtering and cropping. As a result, it have extensive community support and active development.
PrettyTable: It is one of the simple Python libraries for designing to easily and quickly represent the tabular data visually. It allows the model, which column needs to be printed.
Performance metrics
The metrics considered in the results, such as Mean Squared Error (MSE), accuracy, Mean Absolute Error (MAE), Mean Absolute Scaled Error (MASE), and Root Mean Squared Error (RMSE) are important for evaluating the performance of the developed IoT-based smart drip irrigation framework. The correctness of the irrigation prediction obtained by the proposed model is evaluated via accuracy scores. Measures like MAE and MSE determine the method’s ability to correctly predict the need for irrigation, which is crucial for water conservation. Other measures ensure that the proposed technique can effectively avoid unnecessary irrigation by reducing water wastage. These metrics together generate a comprehensive understanding performance of the method in terms of both predictive accuracy and efficiency. The calculations of those measures are given in Eqs. (1)–(5).
 |
1 |
 |
2 |
 |
3 |
 |
4 |
 |
5 |
In the above derivations, the number of observations, actual and the predicted values are termed as 
, 
and 
.
Basic structural description of deep networks used in the smart drip irrigation system
Description of smart irrigation dataset
The dataset utilized for the smart irrigation system is described below:
Greenhouse data experiment drip irrigation 2016: Using https://data.4tu.nl/articles/dataset/Greenhouse_data_experiment_drip_irrigation_2016/12708971, this dataset is accessed on 2025–01-03. This drip irrigation dataset consists of parameter details like humidity, temperature, external rainwater intake, drain water flow and water supply. The use of this datasets is to protect the plants by providing estimated drip irrigation. The collected data is mentioned by the term 
. It contains several attributes like time, global radiation, and temperature in greenhouse, relative humidity, irrigation, irrigation time, rain water and drain. For the training process, 75% (42,538) of data are used and rest 25% (14,180) of data is utilized for testing performance. The data count is 2510, 8.
Basic spatial autoencoder model
The spatial autoencoder51 consists of a convolutional layer to compress the input image into a less-dimensional representation while preserving spatial information. The architecture includes a standard 3-layer CNN with Rectified Linear Units (ReLU) as activation functions. For each pixel 
and channel 
, the activation is computed based on Eq. (6).
 |
6 |
Here, the convolutional layer’s linear output is stated as 
. The output from the last convolutional layer produces response maps that indicate the activation levels of different features. The response maps are first passed through a spatial softmax function as defined in Eq. (7).
 |
7 |
In Eq. (7), the learnt parameter is indicated as 
. This operation converts the activation values into a probability distribution across the spatial dimensions of the image. Each softmax probability distribution 
is calculated based on Eq. (8).
 |
8 |
The feature points are denoted as 
. The output 
serves as the bottleneck representation by encoding with the spatial features. The decoder is a fully connected mapping from the feature points 
and it reconstructs the input image from the learned spatial feature points. Thus, the spatial autoencoder autonomously discovers the useful features and spatial configurations. The structural view of the spatial autoencoder is presented in Fig. 2.
Fig. 2.
Structural view of the spatial autoencoder.
Stacked CapsNet model
In stacked CapsNet52, multiple layers of capsule networks are stacked on top of each other to enhance the model’s ability to recognize and understand more complex patterns. In stacked CapsNet, the activity vectors are used to identify the particular features in the input. This vector’s length represents the feature’s characteristics. The dynamic routing method is one of stacked CapsNet’s most important features. Capsules can communicate with one another through this mechanism. Each capsule in the primary layer sends its output to next capsule layer. The connections between capsules are weighted with coupling coefficients. Capsules in the next layer receive inputs from multiple capsules in the previous layer. The routing process ensures that only the capsules that have similar outputs strengthen their connections, allowing for more accurate feature representation.
Higher-level capsules can represent the more intricate features and these capsules receive the output from the primary capsules. These capsules create a more detailed representation of the input image by combining the data from lower-level capsules. If 
is the output of 
capsule, then the prediction result from 
capsule is computed using Eq. (9).
 |
9 |
The term 
denotes the coupling coefficient. In order to shrink the short and long vectors, a nonlinear function is used and it is determined using Eq. (10).
 |
10 |
In the above Eq. (10), the output vector is mentioned as 
. The log probabilities are upgraded in the routing procedure. In order to update the coupling coefficient and log probability, the agreement 
is evaluated based on Eq. (11).
 |
11 |
The pictorial view of the stacked capsNet model is shown in Fig. 3.
Fig. 3.
Pictorial view of the stacked capsNet model.
Precision farming using developed hybrid deep learning-based smart drip irrigation system
Proposed adaptive residual hybrid network
The residual branch53 is made up with inactive convolution layer after two activated convolution layers. Moreover, an identity shortcut connection is assigned to this layer. This shortcut connection can minimize the errors by guaranteeing better solutions. Bottleneck components are utilized in the residual block and the down-sampling block layers are resampled by the initial residual layers. The first convolutional unit of the down-sampling block down-samples the spatial dimensions. The forward propagation is computed via Eq. (12).
 |
12 |
The weight matrix is specified as 
. The activation unit of layer 
is given as 
. The backward propagation is determined using Eq. (13).
 |
13 |
In the above equation, the loss signal of units is indicated as 
. The learning rate is mentioned as 
.
The ARHN is developed by using spatial autoencoder and stacked capsnet, it processes the input data by first leveraging the spatial representations captured by the spatial autoencoder. Both networks are connected in a serial manner in the ARHN model. While maintaining the spatial correlations within the dataset, the autoencoder is made to learn compressed spatial aspects of the environmental parameters, temperature, and soil moisture content. This network’s residual connections are being helped in handling the vanishing gradient issue and improve the network’s ability to maintain significant features over deep layers by allowing the output of each layer to be added with input of the subsequent layer. After processing, the data is sent to a stacked CapsNet, which uses capsule networks to comprehend the spatial orientation and hierarchical relationships of the characteristics. The residual connection in this network is beneficial in identifying intricate patterns and recording multi-dimensional aspects of the data. Moreover, due to the use of capsules in the stacked capsNet, different attributes of entities, like position and orientation in the data are effectively captured.
Designed MRV-FLO
Rationale for selecting the developed MRV-FLO model
The irrigation process plays an important role in the agricultural field and sustainable production; this allows water to drip slowly to the roots of plants. It can optimally save the water level, and fertilizer to prevent soil erosion. Too many irrigation systems have been developed with various environments of the soil. Considering the existing Genetic Algorithm54, it is not suited for large-size and small-size farmland holdings; also, it needs more computational efficiency for accurately monitoring and validating the interactions among soil, plants, and the environment. It is complex for gathering suitable datasets in the training process to enhance the imbalanced outcomes. Also, it requires regular maintenance and repair for minimizing clogging and other issues. Additionally, the Particle Swarm Optimization (PSO) algorithm55, it is not suitable for protecting sensitive irrigation data to minimize the decision-making performance, which can be time-consuming. It is not capable of continuously monitoring and managing soil health and nutrient levels. With the aim of resolving these issues, the FLO50 method is adopted in this work. The conventional FLO50 is developed based on the natural behaviors of frilled lizards. The frilled lizards are useful in effectively exploring and exploiting the solution space. However, achieving the global optimum is not possible in certain scenarios and this algorithm may not be universally superior across all optimization tasks. Thus, an advanced version of this algorithm named MRV-FLO is proposed.
Novelty of the developed MRV-FLO method
The ability of the ARHN-based smart drip irrigation system is enhanced by the optimization performance with the help of MRV-FLO. The proposed framework has a minimal duration and minimal volume of data for accurately validating the system performance thus, it enhances the generalization ability. Improving the accuracy and efficiency of the implemented approach is accomplished by optimizing the network’s parameters. It becomes flexible for monitoring and validating the training data with diverse environmental conditions. Due to this, it can ultimately enhance the decision-making performance for minimizing the negative and inaccurate outcomes. By tuning the stack depth in the stacked CapsNet and the learning rate in the spatial autoencoder, the MCC of the suggested smart drip irrigation method is improved. The ARHN model can efficiently extract pertinent features from the sensor data without overfitting or underfitting by adjusting the learning rate in the Spatial Autoencoder. Likewise, stacked CapsNet’s robustness and generalization are improved by optimizing the stack depth, which increases the network’s capacity to identify hierarchical patterns in the data. The random value 
in the FLO is upgraded to achieve this and its computation is specified in Eq. (14).
 |
14 |
In the above expression, the term 
denotes the mean fitness and the current fitness are specified as 
. The term 
indicates the worst fitness. Based on these fitness values, the random value is upgraded and it enhances the MRV-FLO algorithm to explore the solution space as more effectively. Moreover, premature convergence is resolved as this upgraded version allows for an efficient search process to identify potential solutions. Thus, the optimization process using the MRV-FLO guarantees the ARHN model effectively, handles the changing circumstances in agriculture and reduces water wastage. The flowchart of the presented MRV-FLO model is showcased in Fig. 4. The pseudocode of the MRV-FLO is given in Algorithm 1.
Fig. 4.
Flowchart of the presented MRV-FLO Algorithm.
Algorithm 1.
Developed MRV-FLO.
Comparative summary of the MRV-FLO approach
The conventional LEA47 method needs extra physical efforts in the irrigation process to amplify the crop yield and it does not control the drip irrigation strategy from anywhere thus, it significantly enhances the soil erosion. It is not capable of collecting and monitoring the sensor data, leading to minimizing the amount of water consumption. Also, it requires frequent maintenance, management, and proper design for an efficient process. The crop’s growth characteristics such as height, yield, soil level, and water productivity information are not validated in this approach to minimize the irrigation process. In the FDA48 approach, the data collection and cleaning process is affected due to enhanced water demands, insufficient water resources, rainfall variability and environmental changes. The presence of soil moisture is a critical variable for agricultural productivity, as it has affected the feature extraction performance. The temperature and soil moisture level information are not collected in the AOA49 framework to make an inaccurate irrigation schedule as needed. It has a higher cost function value to reduce the convergence speed and enhance the energy consumption rates of the irrigation process. The supplied irrigation volumes and the prevailing soil water status are indirectly connected, due to their inability for capturing variable factors and weather conditions in the irrigation space. With the aim of resolving these issues, the MRV-FLO optimization algorithm is utilized in this research work to generate timely outcomes to improve crop yield production. The designed technique utilizes several informative sensor data in the training phase to efficiently improve irrigation performance. Thus, it minimizes the evaporation and water runoff to promote healthy crop yield. It has the ability to ultimately tune the prescribed parameters in the optimization process for maximizing the convergence speed and protecting the crops from aquatic ecosystems. The designed approach can detect anomalies in irrigation patterns and optimize irrigation schedules based on spatial relationships among sensor data to improve water conservation. The raw sensor data and high-dimensional data are easily managed by the developed MRV-FLO optimizaton algorithm to highlight the key features and reduced the computational complexity. The optimization phase can mitigate the vanishing gradient issues that improve the accuracy and efficiency of irrigation scheduling also, it can minimize over and under-watering performance.
Proposed smart drip irrigation system using ARHN
Rationale for selecting the developed ARHN model: The optimal goal of agriculture is to generate safe, fresh food and quality staples for the public. Numerous researchers have developed more smart drip irrigation mechanisms for optimally improving agricultural crop production. In the existing CNN method,56 generally needs massive data in the training performance and a significant time delay can occur due to the cloud server distance. Thus, it minimizes drip irrigation efficiency and fertilization accuracy. It is complex for calculating irrigation time of crop water and generating inaccurate results that minimize the convergence speed. This method is affected from temperature, humidity and topography of the surrounding environments. Considering the existing Bi-directional Long Short Term Memory (Bi-LSTM)57, it is unable to quickly process a large amount of data for optimally extracting the complex low-level features in the gathered images and it is complex for dealing with soil moisture-related issues. Also, it can provide inaccurate outcomes in the moisture content estimation. Thus, it enhances the water and energy usage to maximize the crop yield losses. This method needs several parameters to ultimately validate the system performance for maximizing the crop coefficient values. In order to solve these problems, an ARHN framework is proposed in this work for managing an IoT-based drip irrigation system.
Novelty of the developed method
The developed ARHN approach minimizes the wastage of water and nutrients for maximizing the decision-making and generalization ability. The ARHN model ensures that a suitable amount of water is dispensed from the pump for significantly reducing water wastage in the agricultural field. The optimization technique MRV-FLO achieves better performance by efficiently tuning hyperparameters optimally. It results to be enhanced MCC scores of the irrigation predictions using the proposed model. By automating the irrigation process using the developed MRV-FLO-ARHN, the management of water resources is made easier for farmers to maintain healthy crops while conserving water.
Training process of the ARHN method
The ARHN employs a combination of a spatial autoencoder and stacked CapsNet architectures. Both these networks are incorporated with residual connections. The spatial autoencoder is accountable for successfully extracting meaningful features in the input data. This process is helpful in the ARHN model to understand the spatial relationships within the agricultural environment by analyzing the input data. Meanwhile, the stacked CapsNet improves the ability of the model to recognize patterns and relationships in the data for determining the appropriate amount of water required for irrigation. It can easily optimize nutrient delivery and water efficiency to minimize water loss. Also, it neglects the noise and imbalanced dataset to enhance better generalization and minimize feature loss, leading to faster training. The objective 
of the presented MRV-FLO-ARHN-assisted smart drip irrigation system is provided in the following Eq. (15).
 |
15 |
In Eq. (15), the stack depth in stacked CapsNet is optimized in the range of 
and it is denoted as 
. Hidden neuron count optimized from spatial autoencoder is termed as 
and it is tuned in the range of 
. The spatial autoencoder’s learning rate tuned and it is represented as 
and it lies in the interval of 
. The hidden neuron count in stacked CapsNet is optimized in the range of [5–255] and it is termed as 
. The MCC and FOR is calculated using Eqs. (16) and (17).
 |
16 |
 |
17 |
where, the terms 
and 
denotes the true negative, true positive, false negative and false positive scores. The portrayal of the proposed model is given in Fig. 5.
Fig. 5.
Portrayal of the developed smart drip irrigation model.
Comparative summary of the ARHN approach
In the conventional DSVM approach25, the feature extraction performance is not done due to sensor malfunctions, data transmission errors and environmental interference changes, leading to misinformative outcomes. It fails for monitoring the sensor information from the minimum land size requirement and it reduces the effective implementation. It is not capable of generating localized and accessible outcomes, which can lead to water conservation issues. It is insufficient for addressing the multifaceted barriers to adoption, ultimately leading to more water wastage. Finding the optimal hyperparameters for the DCNN model30 is complex to impact the generalization ability. It needs some security measurements to secure the irrigation, crop, water management, nutrition levels, and farmers’ information in an effective manner. Also, it is not a budget-friendly mechanism and it requires extra steps during the pre-processing, leading to more time-consuming due to their high dependency on time and frequency for negatively affecting the adaptability in unknown environments. The LSTM26 approach is does not have inherent memory mechanisms to capture the plant health and it consumes more time to monitor the irrigation process in a large space. The timely solution and outcomes are not generated in this framework to reduce crop cultivation and production. Therefore, it minimizes the flow of water from the pumps to the plantation, resulting in unnatural crops or diseased crops produced. These issues are rectified with the help of the developed ARHN approach, leading to better generalization and crop production. The implemented IoT-based smart drip irrigation framework can automatically extract the relevant informative features in the gathered sensor data to minimize transmission errors and misinformative results. Due to this, the obtaining water level of the crops is improved by saving significant water and promoting plant health. Further, the adaptive residual hybrid network can continuously monitor soil moisture and weather conditions to adjust the irrigation water level, as it ensures sustainable irrigation practices. It takes minimal computation duration for analyzing vast amounts of datasets to reduce the risk of stress on plants, leading to healthier and more robust crops. This timely accurate outcome can lead to increased crop yields and their quality by reducing the environmental impact of irrigation. It can ensure the farmers to make more informed timely decisions in water management for minimizing the cause of disease. Also, it enhances the flow of water from the pumps to the plantation, resulting in better crops produced.
Results
Performance analysis of the developed smart drip irrigation model
The implemented IoT-based smart drip irrigation system achieves superior performance compared to optimization approaches as well as conventional methods and it is evident from the results as shown in Figs. 6 and 7. The accuracy range of the recommended framework is 2.11%, 1.48%, 0.7% and 1.1% higher than DSVM, DCNN, LSTM, and RHN model at 55th learning percentage. The ARHN’s integration of Spatial Autoencoder and Stacked CapsNet, along with the MRV-FLO’s optimal parameter optimization is the reason for this enhanced performance. By accurately capturing spatial and hierarchical information, the ARHN model minimizes waste in agricultural sectors and guarantees accurate estimation of water requirements. Thus, the proposed ARHN-assisted smart drip irrigation mechanism is capable of handling variability in environmental and soil conditions effectively. The RMSE of the proposed model is 42.14%, 53.15%, 65.12% and 25.12% less than LEA-ARHN, FDA-ARHN, AOA-ARHN and FLO-ARHN at a learning percentage of 35. Similarly, the MAE and MASE of the suggested framework achieves as 75% and 81% lesser than FLO-ARHN due to its adaptive learning capacity at a learning percentage of 75. The suggested architecture tackles the complexity of data gathered from real-world agricultural scenarios by integrating efficient feature extraction techniques. Thus, the suggested approach gained significant improvements in accuracy as well as minimized scores in other error metrics when compared to conventional approaches.
Fig. 6.
Performance comparison of developed smart drip irrigation framework with existing models with respect to (a) MAE, (b) MSE, (c) Accuracy, (d) MASE and (e) RMSE.
Fig. 7.
Performance comparison of developed smart drip irrigation method with various conventional methods in terms of (a) MAE, (b) MSE, (c) Accuracy, (d) MASE and (e) RMSE.
Numerical assessment of developed smart drip irrigation model
The results given in Table 3 demonstrate that the implemented irrigation system optimized using the MRV-FLO-ARHN model outperforms existing optimization algorithms and methods. The presented MRV-FLO-ARHN achieves the highest accuracy value of 94.24% at 50 epochs and it is 1.1%, 1.23%, 1.62% and 0.25% greater than LEA-ARHN, FDA-ARHN, AOA-ARHN and FLO-ARHN. These improvements can highlight the model’s superior ability to manage irrigation decisions and maximize resource efficiency by minimizing water wastage. The hybrid network architecture efficiently learns both temporal and spatial dependencies in the data and it leads to enhanced performance as confirmed by the results. The MRV-FLO optimization method plays a better role to improve the model’s convergence rate and robustness, with various environmental conditions and reduces water wastage. The significant reduction in metrics like MAE and MSE further proves the efficacy of the proposed framework with existing models like DSVM, DCNN, and LSTM as these models struggled to match the developed framework performance due to the inability to handle complex temporal relationships and feature extraction process.
Table 3.
Numerical assessment of developed smart drip irrigation model.
| Algorithm comparisons | |||||
|---|---|---|---|---|---|
| Epochs\algorithm | LEA-ARHN47 | FDA-ARHN48 | AOA-ARHN49 | FLO-ARHN50 | MRV-FLO-ARHN |
| MASE | |||||
| 50 | 15.51994 | 16.55523 | 19.82996 | 8.335714 | 6.282358 |
| 100 | 15.28169 | 17.13882 | 19.74441 | 10.82338 | 7.195167 |
| 150 | 12.47193 | 18.4552 | 18.28172 | 11.33947 | 8.117953 |
| 200 | 11.91451 | 18.58679 | 14.80796 | 8.583646 | 7.010611 |
| 250 | 14.62853 | 16.82839 | 19.45483 | 10.25379 | 6.928168 |
| MAE | |||||
| 50 | 34.78469 | 37.22807 | 44.48804 | 18.9362 | 14.16587 |
| 100 | 34.17384 | 38.30941 | 44.02871 | 24.30622 | 16.26156 |
| 150 | 28.13078 | 41.04147 | 41.04944 | 25.50877 | 18.29984 |
| 200 | 26.86762 | 41.70973 | 33.60447 | 19.51356 | 15.94258 |
| 250 | 33.08293 | 37.83573 | 43.77512 | 23.19458 | 15.68581 |
| RMSE | |||||
| 50 | 159.5529 | 155.0519 | 174.239 | 107.8837 | 97.32261 |
| 100 | 155.542 | 160.3088 | 179.646 | 133.7963 | 105.5907 |
| 150 | 135.6535 | 175.4579 | 167.5925 | 131.2685 | 120.0245 |
| 200 | 134.0299 | 170.6005 | 143.6357 | 112.1155 | 106.1533 |
| 250 | 152.1936 | 162.4103 | 170.8692 | 127.9703 | 101.1214 |
| MSE | |||||
| 50 | 25,457.13 | 24,041.08 | 30,359.23 | 11,638.9 | 9471.691 |
| 100 | 24,193.31 | 25,698.92 | 32,272.68 | 17,901.46 | 11,149.39 |
| 150 | 18,401.88 | 30,785.46 | 28,087.25 | 17,231.43 | 14,405.89 |
| 200 | 17,964.01 | 29,104.54 | 20,631.2 | 12,569.89 | 11,268.52 |
| 250 | 23,162.89 | 26,377.12 | 29,196.28 | 16,376.41 | 10,225.54 |
| Accuracy | |||||
| 50 | 98.15238 | 98.0226 | 97.63698 | 98.99419 | 99.24757 |
| 100 | 98.18483 | 97.96516 | 97.66138 | 98.70895 | 99.13625 |
| 150 | 98.50581 | 97.82005 | 97.81962 | 98.64508 | 99.02799 |
| 200 | 98.5729 | 97.78455 | 98.21507 | 98.96352 | 99.1532 |
| 250 | 98.24277 | 97.99032 | 97.67485 | 98.768 | 99.16683 |
| Method comparisons | |||||
|---|---|---|---|---|---|
| Epochs\methods | DSVM25 | DCNN30 | LSTM26 | RHN | MRV-FLO-ARHN |
| MASE | |||||
| 50 | 20.55609 | 17.36197 | 11.90446 | 9.29279 | 6.282358 |
| 100 | 20.56197 | 17.92241 | 13.21144 | 8.633034 | 7.195167 |
| 150 | 21.00667 | 12.85285 | 13.61102 | 9.622835 | 8.117953 |
| 200 | 18.88402 | 15.04505 | 12.03335 | 7.992213 | 7.010611 |
| 250 | 18.59908 | 17.93943 | 12.02946 | 9.344853 | 6.928168 |
| MAE | |||||
| 50 | 46.42265 | 38.95056 | 26.62839 | 21.02871 | 14.16587 |
| 100 | 46.5933 | 40.21691 | 29.49442 | 19.53429 | 16.26156 |
| 150 | 47.22967 | 29.29665 | 30.66029 | 21.66348 | 18.29984 |
| 200 | 42.04466 | 34.17225 | 27.07815 | 17.98565 | 15.94258 |
| 250 | 41.98086 | 39.8453 | 27.0925 | 20.97129 | 15.68581 |
| RMSE | |||||
| 50 | 176.4918 | 165.4983 | 134.9263 | 123.747 | 97.32261 |
| 100 | 176.3544 | 170.2583 | 137.9994 | 123.0645 | 105.5907 |
| 150 | 177.4868 | 127.9993 | 146.7473 | 130.4652 | 120.0245 |
| 200 | 170.5544 | 148.6635 | 129.9706 | 114.7266 | 106.1533 |
| 250 | 162.6492 | 166.1704 | 134.9069 | 118.8163 | 101.1214 |
| MSE | |||||
| 50 | 31,149.34 | 27,389.69 | 18,205.12 | 15,313.33 | 9471.691 |
| 100 | 31,100.88 | 28,987.89 | 19,043.84 | 15,144.88 | 11,149.39 |
| 150 | 31,501.56 | 16,383.83 | 21,534.78 | 17,021.17 | 14,405.89 |
| 200 | 29,088.79 | 22,100.83 | 16,892.34 | 13,162.19 | 11,268.52 |
| 250 | 26,454.77 | 27,612.59 | 18,199.86 | 14,117.31 | 10,225.54 |
| Accuracy | |||||
| 50 | 97.53422 | 97.93111 | 98.58561 | 98.88304 | 99.24757 |
| 100 | 97.52516 | 97.86384 | 98.43338 | 98.96242 | 99.13625 |
| 150 | 97.49135 | 98.44388 | 98.37145 | 98.84933 | 99.02799 |
| 200 | 97.76676 | 98.18491 | 98.56172 | 99.04468 | 99.1532 |
| 250 | 97.77015 | 97.88358 | 98.56096 | 98.88609 | 99.16683 |
Overview of convergence analysis
The proposed MRV-FLO-ARHN achieves the fastest convergence when compared to other algorithms as depicted in Fig. 8. Within minimal iterations, the developed model converges into the quickest and stabilizes at the lowest cost function value. Suboptimal performance is attained by the existing algorithms due to its less convergence rates and those models obtained higher cost function scores that result in high computational complexity. At 20th iteration, the cost function is 00.2%, 8%, 7.5% and 10% better than LEA-ARHN, FDA-ARHN, AOA-ARHN and FLO-ARHN. Throughout the iterations, MRV-FLO-ARHN continuously maintains the lowest cost function value, proving its effectiveness in locating the global optimum and this is achieved by the ARHN’s parameter optimization via the MRV-FLO. Faster convergence saves processing resources by reducing the number of iterations needed, so the time requirement for analyzing the IoT data is lesser than other algorithms. Due to the MRV-FLO-ARHN’s resilience, the suggested system reliably produces the best outcomes in actual agricultural scenarios.
Fig. 8.
Convergence analysis of the proposed smart drip irrigation model.
Statistical assessment
The statistical outcomes in Table 4 depicts that the MRV-FLO-ARHN model performs better than other optimization algorithms across all key metrics. The best value for MRV-FLO-ARHN is 0.8338, which is considerably minimal than the best rates of the other models. This indicates that the MRV-FLO-ARHN model analyzes the relevant irrigation parameters with greater accuracy. The worst value of MRV-FLO-ARHN is 0.9386 and it is also much lower than that of the other models. Thus, three results demonstrate the model’s robustness in making drip irrigation-related decisions. Similarly, the median and mean values are also better than traditional algorithms that indicate the consistent performance of the proposed MRV-FLO-ARHN. The reliability of the developed model in making precise irrigation predictions is enhanced by the different parameter optimization via the enhanced version of the FLO algorithm. These statistical results validate that MRV-FLO-ARHN is highly effective in making decisions for saving water in precision farming.
Table 4.
Statistical assessment of the developed smart drip irrigation framework.
| TERMS | LEA-ARHN47 | FDA-ARHN48 | AOA-ARHN49 | FLO-ARHN50 | MRV-FLO-ARHN |
|---|---|---|---|---|---|
| BEST | 1.436048 | 1.280873 | 1.304404 | 1.327498 | 0.833842 |
| WORST | 4.925823 | 4.999028 | 2.12611 | 2.810561 | 0.938631 |
| MEAN | 2.122236 | 1.741376 | 1.566072 | 1.69085 | 0.863434 |
| MEDIAN | 1.679164 | 1.471753 | 1.304404 | 1.476167 | 0.848627 |
| STD | 0.9804 | 0.858707 | 0.326324 | 0.377013 | 0.040884 |
Comparative examination of the proposed framework with traditional approaches
Table 5 displays the developed method’s comparative validation with various traditional frameworks utilizing accuracy measure. From the analysis, the existing DSVM method attains a lower accuracy value of 97% to enhance the delayed response and training process. Thus, it enhances the inaccurate outcomes, leading to overfitting issues. However, the implemented approach obtains high accuracy rate to minimize the overall energy consumption and water wastage rates. The proposed method utilizes the optimization process to improve the decision-making and generalization ability. It has the ability to prevent both overwatering and underwatering to significantly save water usage in the irrigation process. Also, the implemented smart drip irrigation method’s performance is improved than 1.7% of DSVM, 0.6% of DCNN, 0.7% of LSTM, and 0.9% of RHN method in the 250th epoch count in terms of accuracy measure. The implemented framework has a higher accuracy value than the existing methods to improve the water quality and prevent the runoff of excess water and fertilizers. Thus, the overall analysis proved that the developed method’s better performance.
Table 5.
Comparative assessment of the proposed smart drip irrigation model.
Training time validation of the implemented method with conventional frameworks
Table 6 illustrates the implemented framework’s training time validation with several existing frameworks. This analysis can easily validate the irrigation schedule and training process for minimizing water usage and improving water quality in an effective manner. Thus, it can reduce water wastage through evaporation and runoff to enhance soil structure for maximizing crop yields. From the table validation, the conventional AOA-ARHN algorithm shows the highest execution time of 18.4 (mins) to gradually reduce the decision-making performance for delaying the generation of timely irrigation recommendations. However, the developed MRV-FLO-ARHN method attains a minimal and better training time of 14 (mins) to enable quick processing of the data in diverse weather forecasts and sensors. This allows immediate and automatic adjustments of irrigation schedules. The training time of the proposed framework is minimized than 27% of DSVM, 18% of DCNN, 23% of LSTM, and 6% of RHN methods.
Table 6.
Training time assessment of the proposed smart drip irrigation model.
Computational complexity validation of the developed method with conventional approaches
Table 7 elucidates the computational complexity validation of the proposed framework with various traditional frameworks. From, the table analysis, the developed method attains a lower computational complexity rate than the existing methods to optimally improve the decision-making and generalization ability. From the table validation, the terms 
and 
represent the number of population and the maximum number of iteration values. Also, the term 
denotes the chromosome length. These parameters are used in the proposed optimization phase to enhance the irrigation performance, leading to significant water savings.
Table 7.
Computational complexity validation of the implemented method.
Comparative analysis of the implemented framework with specific metrics
Table 8 displays a developed method’s comparative analysis with various existing frameworks utilizing specific metrics such as Water Productivity (WP), Water Use Efficiency (WUE), and Irrigation Efficiency (IE). In Table 8, the proposed MRV-FLO-ARHN method has a better water use efficiency rate of 7.28 than the existing methods to minimize water consumption, runoff and evaporation for improving crop health and yields in an effective manner. This can generate plants with the appropriate water amount and nutrients for optimally reducing weed growth. Additionally, the existing AOA-ARHN method attains a minimal irrigation efficiency value of 3.65 to impact water conservation and crop yields, which reduces water flow and uniformity. It needs additional and regular maintenance for uniformly distributing water to crops, leading to the highest training duration. However, the developed method shows better and maximum irrigation efficiency of 7.50 to ultimately reduce energy consumption rates for maximizing plant health. This can directly deliver water to the plant root zone to minimize water loss. It has the ability to truthfully validate the optimal irrigation schedules to prevent overwatering and enhance plants receiving the right amount of water. The water productivity value of the designed framework is reduced than 82% of LEA-ARHN, 78% of FDA-ARHN, 82% of AOA-ARHN, and 75% of FLO-ARHN. The implemented approach achieved a lower water productivity value to maximize crop yields by neglecting overwatering and underwatering. Thus, it can enhance water management and reduce the energy needed for pumping and treating water. The water use efficiency of the implemented approach is maximized than 43% of DSVM, 2% of DCNN, 21% of LSTM, and 14% of RHN method to ensure the plants obtain the optimal quantity of water and nutrients at the same time for generating higher yields.
Table 8.
Comparative analysis of the designed smart drip irrigation method with specific metrics.
| Algorithm comparison | |||||
|---|---|---|---|---|---|
| Terms | LEA-ARHN47 | FDA-ARHN48 | AOA-ARHN49 | FLO-ARHN50 | MRV-FLO-ARHN |
| WUE | 4.47 | 4.73 | 2.01 | 1.87 | 7.28 |
| IE | 5.14 | 4.35 | 3.65 | 5.17 | 7.50 |
| WP | 1.39 | 1.16 | 1.40 | 1.03 | 0.25 |
Comparative analysis of the proposed approach with different environmental conditions and crop types
Table 9 displays the developed method’s comparative analysis with various traditional approaches in terms of accuracy measure. From Table 9, various crop types such as Rice, Wheat, Maize, Sugarcane, and Tomato are considered to validate the performance. Considering the rice crop, the existing LEA-ARHN method has an accuracy rate of 93% that minimizes the water distribution process, which affects uniform crop growth. It is complex for accurately managing soil and water conditions. However, the developed MRV-FLO-ARHN method has a higher accuracy value of 96% than the existing methods to minimize water consumption for enhancing yields and its quality in a limited duration, leading to healthier growth. Accurate water delivery can promote better nutrient uptake to ensure stronger plants and minimize fertilizer waste. The accuracy rate of the implemented approach is maximized than 2.1% of DSVM, 0.9% of DCNN, 1.9% of LSTM, and 0.5% of RHN in terms of sugarcane crop. Thus, the overall comparative analysis proved that the developed better performance to ensure optimal plant health.
Table 9.
Comparative analysis of the designed smart drip irrigation framework with different environmental conditions and crop types.
| Algorithm comparison | |||||
|---|---|---|---|---|---|
| Terms | LEA-ARHN 47 | FDA-ARHN 48 | AOA-ARHN 49 | FLO-ARHN 50 | MRV-FLO-ARHN |
| Rice | 93.723 | 94.358 | 95.537 | 94.968 | 96.371 |
| Wheat | 93.554 | 93.529 | 95.145 | 94.560 | 95.987 |
| Maize | 93.931 | 94.515 | 95.171 | 94.819 | 96.371 |
| Sugarcane | 94.229 | 94.328 | 95.567 | 95.296 | 96.441 |
| Tomato | 94.046 | 94.774 | 95.716 | 94.983 | 96.326 |
Reliability analysis of the implemented method
Figure 9 displays the reliability validation of the proposed method with the conventional method in various data sizes (MB). It can optimally estimate the system performance to generate accurate and consistent results among different conditions. Reliability analysis can easily identify the potential issues in the system to prevent downtime and damage for enhancing efficiency and responsiveness. In Fig. 9, the existing POA method has less reliability value that affect the system maintenance and irrigation schedules, leading to water waste and crop damage. It does not have the ability to significantly collect and transmit a lot of soil moisture, crop health, and irrigation patterns data. Yet, the implemented approach attains a higher reliability rate to reduce overall water consumption and enhance the water management process. Consistent water delivery promotes healthy plant growth and development, leading to stronger crops.
Fig. 9.
Reliability validation of the implemented smart drip irrigation framework.
Comparative analysis of the developed method over similar state-of-the-art approaches
The comparative evaluation of the developed classifier with various classifiers using several performance measures is provided in Table 10. The conventional DSVM approach has attained a higher error value of 20 in terms of the MASE measure to provide inaccurate prediction outcomes. Also, it enhances uneven and inconsistent water distribution across the agricultural field, leading to overfitting issues. Yet, the implemented approach achieves a minimal MASE error value of 7.4 for allowing a precise and timely irrigation process to minimize water waste. It has the ability to ultimately optimize the significant hyperparameter to provide accurate solutions in a restricted time. Also, the irrigation process of the implemented framework is maximized than 1.5% of DSVM, 0.9% of DCNN, 1.2% of LSTM, and 1.1% of RHN in terms of accuracy measure at the 20th epoch count. The developed framework attains a better accuracy rate to improve crop yield and promote efficient resource management. Thus, the comparative analysis demonstrates the developed method’s better and more reliable performance than the conventional frameworks.
Table 10.
Comparative analysis of the implemented smart drip irrigation classifier with similar state-of-the-art approaches.
| Classifier comparisons | |||||
|---|---|---|---|---|---|
| Epochs\methods | DSVM25 | DCNN30 | LSTM26 | RHN | MRV-FLO-ARHN |
| MASE | |||||
| 20 | 20.14525 | 15.43557 | 17.92721 | 16.51795 | 7.411386 |
| 40 | 19.22294 | 19.40983 | 15.7565 | 19.14387 | 7.911328 |
| 60 | 18.3279 | 19.39837 | 15.98978 | 15.15561 | 7.090757 |
| 80 | 15.28968 | 17.11034 | 12.69121 | 17.52698 | 7.031438 |
| 100 | 17.21046 | 18.02 | 15.18708 | 13.48567 | 9.722676 |
| MAE | |||||
| 20 | 45.19458 | 34.51356 | 40.26156 | 37.16427 | 16.79585 |
| 40 | 42.83413 | 43.10845 | 35.74322 | 42.76874 | 17.82616 |
| 60 | 41.15949 | 43.34928 | 36.16268 | 33.93461 | 16.11483 |
| 80 | 34.84211 | 38.17225 | 28.78309 | 38.91707 | 15.82137 |
| 100 | 38.36045 | 40.05742 | 34.24721 | 30.43222 | 21.85327 |
| RMSE | |||||
| 20 | 174.44 | 148.053 | 169.714 | 159.87 | 105.764 |
| 40 | 172.194 | 179.358 | 152.565 | 172.27 | 112.458 |
| 60 | 167.617 | 178.865 | 159.742 | 148.654 | 101.283 |
| 80 | 144.582 | 165.455 | 135.994 | 170.086 | 107.693 |
| 100 | 158.223 | 172.35 | 156.683 | 143.534 | 132.617 |
| MSE | |||||
| 20 | 30,429.3 | 21,919.6 | 28,802.8 | 25,558.5 | 11,186 |
| 40 | 29,650.9 | 32,169.2 | 23,276 | 29,676.9 | 12,646.9 |
| 60 | 28,095.3 | 31,992.7 | 25,517.5 | 22,098 | 10,258.2 |
| 80 | 20,904.1 | 27,375.5 | 18,494.5 | 28,929.1 | 11,597.7 |
| 100 | 25,034.5 | 29,704.4 | 24,549.6 | 20,602 | 17,587.3 |
| Accuracy | |||||
| 20 | 97.5995 | 98.1668 | 97.8615 | 98.026 | 99.1079 |
| 40 | 97.7248 | 97.7103 | 98.1015 | 97.7283 | 99.0531 |
| 60 | 97.8138 | 97.6975 | 98.0792 | 98.1975 | 99.144 |
| 80 | 98.1493 | 97.9724 | 98.4712 | 97.9329 | 99.1596 |
| 100 | 97.9625 | 97.8723 | 98.1809 | 98.3836 | 98.8392 |
Comparative analysis of the implemented algorithm with other models
The overall comparative validation of the implemented algorithm with various models is given in Table 11. From the table validation, the negative outcome of the proposed framework is reduced than 54% of LEA-ARHN, 50% of FDA-ARHN, 65% of AOA-ARHN, and 30% of FLO-ARHN methods in terms of MAE measure at 80th epoch count. The developed approach acquires less MAE rate to improve water efficiency and minimize evaporation issues in the irrigation process. It can significantly reduce the water consumption rate and fertilizer usage to improve the nutrient level of crops. Additionally, the conventional AOA-ARHN approach attains a lower accuracy value of 97.6%. It is not capable of handling large size datasets to maximize water consumption rates. Yet, the developed MRV-FLO-ARHN method attains a higher accuracy rate of 99.1% to reduce water consumption values, leading to significant energy savings and crop quality.
Table 11.
Comparative evaluation of the proposed smart drip irrigation algorithm with conventional models.
| Algorithm comparisons | |||||
|---|---|---|---|---|---|
| Epochs\methods | LEA-ARHN47 | FDA-ARHN48 | AOA-ARHN49 | FLO-ARHN50 | MRV-FLO-ARHN |
| MASE | |||||
| 20 | 19.0309 | 17.9039 | 20.0086 | 15.5375 | 7.41139 |
| 40 | 14.3494 | 12.2864 | 20.3455 | 14.0794 | 7.91133 |
| 60 | 15.0049 | 17.7301 | 18.1854 | 12.4384 | 7.09076 |
| 80 | 15.4576 | 14.2141 | 20.6276 | 10.0611 | 7.03144 |
| 100 | 16.473 | 15.5764 | 21.795 | 12.9685 | 9.72268 |
| MAE | |||||
| 20 | 42.2185 | 39.9601 | 44.9298 | 34.6986 | 16.7959 |
| 40 | 32.673 | 27.7081 | 45.7624 | 31.5758 | 17.8262 |
| 60 | 33.673 | 39.3987 | 41.2663 | 28.1547 | 16.1148 |
| 80 | 34.7831 | 31.7257 | 46.1818 | 22.6842 | 15.8214 |
| 100 | 36.9649 | 34.5614 | 48.9506 | 29.1388 | 21.8533 |
| RMSE | |||||
| 20 | 175.354 | 173.872 | 172 | 156.1 | 105.764 |
| 40 | 142.843 | 129.124 | 178.882 | 149.2 | 112.458 |
| 60 | 155.064 | 173.693 | 164.325 | 131.779 | 101.283 |
| 80 | 153.528 | 146.757 | 179.879 | 115.875 | 107.693 |
| 100 | 162.141 | 166.284 | 185.096 | 143.448 | 132.617 |
| MSE | |||||
| 20 | 30,749 | 30,231.4 | 29,584.1 | 24,367.1 | 11,186 |
| 40 | 20,404.1 | 16,672.9 | 31,998.9 | 22,260.5 | 12,646.9 |
| 60 | 24,044.8 | 30,169.1 | 27,002.8 | 17,365.6 | 10,258.2 |
| 80 | 23,570.8 | 21,537.7 | 32,356.4 | 13,427 | 11,597.7 |
| 100 | 26,289.8 | 27,650.2 | 34,260.6 | 20,577.2 | 17,587.3 |
| Accuracy | |||||
| 20 | 97.7575 | 97.8775 | 97.6135 | 98.157 | 99.1079 |
| 40 | 98.2645 | 98.5283 | 97.5693 | 98.3228 | 99.0531 |
| 60 | 98.2114 | 97.9073 | 97.8081 | 98.5045 | 99.144 |
| 80 | 98.1525 | 98.3149 | 97.547 | 98.7951 | 99.1596 |
| 100 | 98.0366 | 98.1642 | 97.3999 | 98.4523 | 98.8392 |
Discussion
The performance evaluation of the implemented method is validated with diverse conventional frameworks to effectively enhance water management and usage in the agricultural field. The performance validation of the introduced classifier and algorithm is displayed in Figs. 6 and 7. In Fig. 6, the conventional AOA-ARHN method shows a high MAE value of 44 at the 55th learning percentage that increases overfitting problems to enhance poor performance. However, the developed MRV-FLO-ARHN framework attains a minimal MAE value of 12, which effectively minimizes the negative outcomes for enhancing the accurate solutions with a limited duration. Further, it reduces the overfitting issues for improving the generalization ability. In Fig. 7b, the MSE of the proposed approach is reduced than 57% of DSVM, 60% of DCNN, 75% of LSTM, and 76% of RHN to enhance the smart drip irrigation performance. The proposed framework has a lower MSE value to effectively optimize the water delivery of crop yield plants and improve efficiency. In Table 3 analysis, the implemented framework’s performance is increased than 1.4% of DSVM, 0.98% of DCNN, 0.60% of LSTM, and 0.10% of RHN methods to significantly minimize the energy consumption and enhance environmental sustainability, reduce water loss and improve water savings. The convergence validation of the designed approach is given in Fig. 8. The implemented framework attains a lower convergence value to provide better generalization and gradually prevent overfitting issues, leading to better decision-making performance. In the statistical assessment of Table 4, the developed method has an effective best value to improve faster training process. It can efficiently ensure the optimization process for efficiently optimizing the hyperparameters. From Table 5, the overall performance of the developed approach is enhanced than 1.6% of DSVM, 0.9% of DCNN, 0.3% of LSTM, and 0.7% of RHN method in the 200th epoch count in terms of accuracy measure. This maximized accuracy rate of the implemented framework can minimize energy consumption to improve crop yields in an effective manner. In Table 6, the training time of the designed framework is minimized than 27% of DSVM, 18% of DCNN, 23% of LSTM, and 6% of RHN methods. Also, it takes a minimal duration to continuously monitor and validate the system performance. In Table 7, the proposed method has less computational complexity to enhance the IoT-based smart drip irrigation process without any interference. Considering Table 8 analysis, the IE of the implemented approach is maximized than 21% of DSVM, 1% of DCNN, 26% of LSTM, and 10% of RHN method to ensure the plants generate higher and healthier yields. It can optimally provide a slow and steady water supply, leading to maintaining optimal moisture levels in productive plants. The accuracy value of the developed framework (Table 9) is enhanced than 2.6% of DSVM, 1.3% of DCNN, 2.1% of LSTM, and 0.7% of RHN in terms of wheat crop. It can ultimately manage large-size datasets for reducing overfitting and underfitting issues. Figure 9 displays the reliability analysis of the developed method; it can effectively identify potential failure and optimize the system to ensure consistent and reliable irrigation delivery even under diverse environmental conditions.
Challenges occur during the implementation process
For efficient evaluation, the smart drip irrigation framework is validated by concentrating on prescribed irrigation datasets. In the development and testing phase, the collection of datasets in the irrigation prediction becomes complex for the implementation process, which fails to generate optimal outcomes. Thus, it degrades the irrigation performance accuracy while predicting the data in the network. However, the developed testing and training evaluation is considered in the validation process to examine the efficiency of the model. This testing and training validation is executed by separating the whole data into 25% and 75% to provide significant performance in the development and testing of the implemented IoT-based smart drip irrigation method.
Ethical/societal Implications of the developed method
Data security and privacy The smart drip irrigation mechanisms effectively collect and transmit vast amounts of water usage, soil conditions and crop health training data thus; it enhances unauthorized access and cyberattacks for minimizing the security level of farmers’ data.
Sustainability The smart drip irrigation process highly depends on water consumption, and environmental changes and provides inaccurate outcomes to affect the resource management. Thus, it effectively impacts on livelihoods and the environment.
Water and energy management Water usage information is crucial for irrigation performance; the water usage and wastage detail is not properly optimized to minimize the extraction performance.
Soil health: Improper irrigation practices can lead to soil degradation and evaporation that affect better crop yields. Also, it impacts with ecosystems and biodiversity, which requires careful access.
Solution addressed by the developed model An adaptive residual hybrid network is used in this research work to protect sensitive data from unauthorized access. It can effectively minimize data breaches in the training process to improve the protection of sensitive agricultural information. Also, it minimizes the environmental footprint for improving positive consequences. The developed method’s proper irrigation strategies can help to significantly analyze the soil condition for reducing soil degradation and enhancing healthier productivity.
Role of artificial intelligence (AI) in farming practices
Integrating the AI in diverse deep learning and machine learning models has been recently emerged in recent days. Because integrating the AI technology shows revolutionary in the farming practices for enhanced crop productivity and monitoring, earlier pest detection control and targeted resource application. For the precise agricultural practices, the data is collected from various sensors, drones and satellites so, that AI facilitates to improve the crop yields, other sustainable practice and cost savings approach. Also, incorporating the AI can automate and forecast the weather, optimize planting, enabling better decision-making performance and providing harvesting schedules to the farmers. Thus, it helps to enhance the profitability to the farmers. Moreover, the key roles of AI in farming practices are shown below:
Ensuring precise agriculture AI technology can effectively analyze the data to ensure real-time information based on soil condition, plant growth, which allows the precise analysis of water, fertilizers and pesticides.
Monitoring healthy crops With the help of AI, the drones and satellite captures better and high image resolution, whereas the machine learning and deep learning models helps to analyze to identify the earlier signs of disease, enabling timely interventions.
Managing resource Integrating the AI in real-time practice ensures to optimize the use of water, fertilizers, which restricts in reducing the waste and environmental impact. Based on this optimized resources, AI helps the farmers to minimize the computational burden and operational cost for effectively improves the overall profitability.
Earlier identification of pest and disease AI is the out-breaking approach that helps to quickly identify the pest infestations and plant disease, allowing the farmers to provide treatments and reduce the crop damage.
AI act as the bridging tool in the traditional and modern farming practices, whereas the AI-powered mobile applications serves to provide better guidance in real-time practice, especially in small-scale farmers who may lack access to the traditional extension services. Based on these aspects, the farmers can provide informed decision-making performance by selecting the crops based on soil conditions. Thus, it enhance the crop productivity and profitable to the farmers. Moreover, the AI-powered technologies have automate task like planting, weeding as well as harvesting, reducing the labor dependency and improves efficiency in the agricultural practices.
Conclusion
This work developed an IoT-based smart drip irrigation framework for managing water usage in agriculture to enhance crop yield. The system utilized data captured by IoT sensors that were accumulated from benchmark sources. An ARHN model, incorporating stacked CapsNet and spatial autoencoder, was developed to process the data and improve the irrigation-based decision-making process. The MRV-FLO technique was employed to fine-tune the model parameters, ensuring accurate irrigation while minimizing water wastage. Finally, the designed approach was compared with existing methods to verify its effectiveness and sustainability. When compared with DSVM, DCNN, LSTM and RHN, the suggested MRV-FLO-ARHN model’s accuracy at an epoch count of 200 was achieved higher than 1.72%, 1.32%, 0.66% and 0.36%. Overall, the MRV-FLO-ARHN model’s superior results confirm its effectiveness in precision farming and it achieves a scalable solution for efficient water management.
Advantages of the developed method
The developed IoT-based smart drip irrigation methodology is capable of accurately minimizing water wastage and usage to enhance crop yields in a prescribed duration. It can timely perform efficient water allocation to the crops to reduce evaporation and runoff, leading to significant water savings. The optimization process in the developed method can enhance the decision-making performance in an effective manner thus; it minimizes the risk of fungal and bacterial diseases. Also, it can significantly reduce water consumption, energy consumption, and chemical runoff rates to protect water quality. It can optimally reduce inaccurate and negative predictive outcomes for enhancing reliability and generalizability.
Limitations
The implemented framework has several limitations in smart drip irrigation performance. The proposed approach is restricted with training the data and it cannot manage various training datasets also, it does not utilize a feature extraction process during the training process. It needs additional prescribed performance measures to continuously validate and predict the irrigation process. Also, the preprocessing technique is not used in this developed method in the data cleaning process to minimize the irrigation performance. The proposed method does not gather agricultural data from flow sensors, pressure sensors, liquid level sensors, and rainfall sensors. In the validation process, real-time and weather forecasting data is not concentrated in this research work. Additionally, the crop yield prediction process is not to be utilized in this research work. The edge devices and cloud computing mechanisms are not used in the developed method to secure sensitive agricultural data. Impact on the generalizability of the findings: The developed method is not optimally scaled with real-time applications and also, it impacts the generalizability of the findings. Larger sample sizes lead to more generalizable findings and overfitting issues. In order to solve these issues, an efficient ensemble-based deep learning model will be considered to analyze the scalability in real-time applications.
Future work
In future, several specification sensors such as flow sensors, pressure sensors, liquid level sensors, and rainfall sensors will be combined with the developed mechanism to capture truthful solutions and ultimately provide better decisions in the irrigation process. ARHN method will be incorporated and performed with edge devices and cloud computing to optimally secure sensitive agricultural data. Moreover, in the subsequent works, the crop yield prediction process will be utilized in the upcoming works to generate optimal positive outcomes. Further, the real-time data will be utilized to precisely evaluate the system’s performance. Also, the proposed framework will be incorporated and performed with weather forecasting data to optimally predict the water delivery rates for reducing waste. The designed method will be integrated with machine learning explainability methods to improve the optimization process in the agricultural field to minimize energy consumption values. In future, the implemented framework will be extended to different crop types for generating flexibility and optimal outcomes. The preprocessing and feature selection process will be utilized in upcoming works to provide better and more reliable outcomes.
In future work, the developed irrigation approach will be integrated with real-world control of irrigation hardware such as Artificial Intelligence (AI) mechanism to efficiently save the processing duration and resources, leading to minimizing the water waste and potentially enhancing the crop yields. It can precisely control the water delivery ranges and automatically minimize the runoff for reducing the evaporation and percolation. The computer vision approaches will be utilized in upcoming works to recognize the images and videos of crops to provide insights into their health and stress levels. This enables early detection of diseases and nutrient deficiencies. Also, the edge computing mechanism will be incorporated with the proposed approach to ensure real-time data processing, reliability and decision-making of the irrigation system. Thus, it efficiently minimizes the reliance on manual labor and can contribute to agricultural operations. In future, the actuator control needs to be adopted in the developed model to enable autonomous irrigation.
Acknowledgements
Princess Nourah bint Abdulrahman University Researchers Supporting Project number (PNURSP2025R79), Princess Nourah bint Abdulrahman University, Riyadh, Saudi Arabia.
Author contributions
Ahmad Y.A. Bani Ahmad, Jafar A. Alzubi, and Chanthirasekaran K.: Conceptualization, Methodology, Software Data curation, Writing- Original draft preparation, Reviewing and Editing, Software, Validation. ShabanaUrooj, Mohammad Shahzad, and Yogapriya J.: Funding Acquistion, Writing- Reviewing and Editing, Visualization, Investigation.
Funding
This research is funded by Princess Nourah bint Abdulrahman University Researchers Supporting Project number (PNURSP2025R79), Princess Nourah bint Abdulrahman University, Riyadh, Saudi Arabia.
Data availability
The dataset utilized for the smart irrigation system is described below: Greenhouse data experiment drip irrigation 2016: Using https://data.4tu.nl/articles/dataset/Greenhouse_data_experiment_drip_irrigation_2016/12708971 this dataset is accessed on 2025–01-03. This drip irrigation dataset consists of parameter details like humidity, temperature, external rainwater intake, drain water flow and water supply.
Declarations
Competing interests
The authors declare no competing interests.
Ethical approval
Not applicable.
Footnotes
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
References
- 1.Vinod Kumar, S. et al. IoT-based smart drip irrigation scheduling and wireless monitoring of microclimate in sweet corn crop under plastic mulching. Irrigation Sci.45, 1–20 (2024). [Google Scholar]
- 2.Singh, C. D., Rao, K. V., Kumar, M. & Rajwade, Y. A. Development of a smart IoT‐based drip irrigation system for precision farming. Irrigation & Drainage 72(1) (2023).
- 3.Sharifnasab, H. et al. Evaluating the use of intelligent irrigation systems based on the IoT in grain corn irrigation. Water15(7), 1394 (2023). [Google Scholar]
- 4.Chua, W. F. D., Lim, C. L., Koh, Y. Y. & Kok, C. L. A novel IoT photovoltaic-powered water irrigation control and monitoring system for sustainable city farming. Electronics13(4), 676 (2024). [Google Scholar]
- 5.Roy, S. K., Misra, S., Raghuwanshi, N. S. & Das, S. K. AgriSens: IoT-based dynamic irrigation scheduling system for water management of irrigated crops. IEEE Internet of Things J.8(6), 5023–5030 (2020). [Google Scholar]
- 6.Singh, C. D. & Kumar, V. S. Effects on growth and yields of sweet corn using internet of things (IoT) based smart drip irrigation and ETc based drip irrigation systems. RASSA J. Sci. Soc.5(2), 87–95 (2023). [Google Scholar]
- 7.Behzadipour, F. et al. A smart IoT-based irrigation system design using AI and prediction model. Neural Comput. Appl.35(35), 24843–24857 (2023). [Google Scholar]
- 8.de Jesús, P. S. A. IoT-based Drip Irrigation Monitoring and Controlling System using NB-IOT, ESP32+ Raspberry Pi," Revista Cubana de Ciencias Informáticas (2023).
- 9.Jain, R. K. et al. Experimental performance of soil monitoring system using IoT technique for automatic drip irrigation. Int. J. Commun. Syst.36(18), e5617 (2023). [Google Scholar]
- 10.Singh, C. D., Ramana Rao, K. V. R., Rajwade, Y. A., Kumar, M. Jawaharlal, D. & Asha, K. R. IoT-based Smart Drip Irrigation Scheduling and Wireless Monitoring of Microclimate in Sweet Corn Crop under Plastic Mulching (2024).
- 11.Jani, K. A. & Chaubey, N. K. SDIPMIoT: Smart drip irrigation and preventative maintenance using IoT. SSRG Int. J. Electric. Electron. Eng.10(7), 22–30 (2023). [Google Scholar]
- 12.Kok, C. L., Kusuma, I. M. B. P., Koh, Y. Y., Tang, H. & Lim, A. B. Smart aquaponics: An automated water quality management system for sustainable urban agriculture. Electronics13, 820 (2024). [Google Scholar]
- 13.Benzaouia, M., Hajji, B., Mellit, A. & Rabhi, A. "Fuzzy-IoT smart irrigation system for precision scheduling and monitoring. Comput. Electron. Agric.215, 108407 (2023). [Google Scholar]
- 14.Akshay, D. V. S. et al. Optimizing growth and yield in aerobic rice through IoT-based drip irrigation and fertigation. Int. J. Plant Soil Sci.36(7), 190–200 (2024). [Google Scholar]
- 15.Dharshan, Y., D. Devasena, S. Kayalvizhi, & Sharmila, B. IoT-based automatic drip irrigation control using intelligent agriculture. In Drone Data Analytics in Aerial Computing 187–203 (2023).
- 16.Rehman, D. et al. Enhancing crop production and water conservation through IoT-based smart irrigation system. J. Comput. Biomed. Inf.5(01), 96–104 (2023). [Google Scholar]
- 17.Et-taibi, B. et al. Enhancing water management in smart agriculture: A cloud and IoT-Based smart irrigation system. Results Eng.22, 102283 (2024). [Google Scholar]
- 18.Santosh, D. T. et al. Development of IoT based intelligent irrigation system using particle swarm optimization and XGBoost techniques. Bullet. Electric. Eng. Inf.13(3), 1927–1934 (2024). [Google Scholar]
- 19.Vinod Kumar, S. et al. IoT-based smart drip irrigation scheduling and wireless monitoring of microclimate in sweet corn crop under plastic mulching. Irrigat. Sci.43, 1107–1126 (2024). [Google Scholar]
- 20.Anagha, C. S., Pawar, P. M. & Tamizharasan, P. S. Cost-effective IoT-based intelligent irrigation system. Int. J. Syst. Assur. Eng. Manag.14, 263–274 (2023). [Google Scholar]
- 21.Yadav, S., Singh, P., Yadav, P., Srivastav, P., Yadav, S. K. & Singh, S. Implementation of Solar Powered Based Smart Drip Irrigation System Using IoT. In 2024 10th International Conference on Electrical Energy Systems (ICEES) 1–6 (2024).
- 22.Jain, R. K. Experimental performance of smart IoT-enabled drip irrigation system using and controlled through web-based applications. Smart Agric. Technol.4, 100215 (2023). [Google Scholar]
- 23.Inayah, I., Agustirandi, B., Budiman, M., Djamal, M. & Faizal, A. Experimental Design: Implementation of IoT-based drip irrigation to enhance the productivity of Cilembu sweet potato (Ipomoea batatas) cultivation. Results Eng.25, 103600 (2024). [Google Scholar]
- 24.Kumar, V. et al. Evaluation of IoT based smart drip irrigation and ETc based system for sweet corn. Smart Agric. Technol.5, 100248 (2023). [Google Scholar]
- 25.Suresh, P. et al. IoT with evolutionary algorithm based deep learning for smart irrigation system. Comput. Mater. Continua71(1), 1713–1728 (2022). [Google Scholar]
- 26.Lakshmi, G. S. P. et al. An intelligent IOT sensor coupled precision irrigation model for agriculture. Measur. Sensors25, 100608 (2023). [Google Scholar]
- 27.Ahmad, U., Alvino, A. & Marino, S. Solar fertigation: A sustainable and smart IoT-based irrigation and fertilization system for efficient water and nutrient management. Agronomy12(5), 1012 (2022). [Google Scholar]
- 28.Amiri, Z., Heidari, A., Navimipour, N. J., Esmaeilpour, M. & Yazdani, Y. The deep learning applications in IoT-based bio- and medical informatics: A systematic literature review. Neural Comput. Appl.36, 5757–5797 (2024). [Google Scholar]
- 29.Morchid, A. et al. IoT-based smart irrigation management system to enhance agricultural water security using embedded systems, telemetry data, and cloud computing. Results Eng.23, 102829 (2024). [Google Scholar]
- 30.Kumar, P., Udayakumar, A., Kumar, A. A., Kannan, K. S. & Krishnan, N. Multiparameter optimization system with DCNN in precision agriculture for advanced irrigation planning and scheduling based on soil moisture estimation. Environ. Monitor. Assess.195(1), 13 (2023). [DOI] [PubMed] [Google Scholar]
- 31.Sujono, H. A. & Nainggolan, W. P. Drip irrigation control system based on Mamdani fuzzy logic and Internet of Things (IoT). Prz Elektrotech1, 65–69 (2024). [Google Scholar]
- 32.Heidari, A., Navimipour, N. J. & Unal, M. A secure intrusion detection platform using blockchain and radial basis function neural networks for internet of drones. IEEE Internet of Things J.10(10), 8445–8454 (2023). [Google Scholar]
- 33.Heidari, A., Amiri, Z., Jamali, M. A. J. & Jafari, N. "Assessment of reliability and availability of wireless sensor networks in industrial applications by considering permanent faults. Concurren. Computat. Pract. Exper.36(27), 8252 (2024). [Google Scholar]
- 34.Amiri, Z., Heidari, A. & Navimipour, N. J. Comprehensive survey of artificial intelligence techniques and strategies for climate change mitigation. Energy308, 132827 (2024). [Google Scholar]
- 35.Vakili, A. et al. "A new service composition method in the cloud-based Internet of things environment using a grey wolf optimization algorithm and MapReduce framework. Concurren. Comput. Pract. Exper.36(16), 8091 (2024). [Google Scholar]
- 36.Heidari, A., Shishehlou, H., Darbandi, M., Navimipour, N. J. & Yalcin, S. A reliable method for data aggregation on the industrial internet of things using a hybrid optimization algorithm and density correlation degree. Clust. Comput.27, 7521–7539 (2024). [Google Scholar]
- 37.Amiri, Z., Heidari, A., Zavvar, M., Navimipour, N. J. & Esmaeilpour, M. The applications of nature-inspired algorithms in Internet of Things-based healthcare service: A systematic literature review. Trans. Emerg. Telecommun. Technol.35(6), 4969 (2024). [Google Scholar]
- 38.Zanbouri, K. M., Darbandi, M. N., Heidari, A., Navimipour, N. J. & Yalcın, S. A GSO-based multi-objective technique for performance optimization of blockchain-based industrial Internet of things. Int. J. Commun. Syst.37(15), 5886 (2024). [Google Scholar]
- 39.Heidari, A., Navimipour, N. J., Zeadally, S. & Chamola, V. Everything you wanted to know about ChatGPT: Components, capabilities, applications, and opportunities. Internet Technol. Lett.7(6), 530 (2024). [Google Scholar]
- 40.Asadi, M., Jamali, M. A. J., Heidari, A. & Navimipour, N. J. Botnets unveiled: A comprehensive survey on evolving threats and defense strategies. Trans. Emerg. Telecommun. Technol.35(11), 5056 (2024). [Google Scholar]
- 41.Heidari, A., Jamali, M. A. J. & Navimipour, N. J. Fuzzy logic multicriteria decision-making for broadcast storm resolution in vehicular ad hoc networks. Int. J. Commun. Syst.38(5), 6034 (2024). [Google Scholar]
- 42.Heidari, A., Amiri, Z., Jamali, M. A. J. & Navimipour, N. J. Enhancing solar convection analysis with multi-core processors and GPUs. Eng. Rep.7(1), 13050 (2024). [Google Scholar]
- 43.Toumaj, S., Heidari, A., Shahhosseini, R. & Navimipour, N. J. Applications of deep learning in Alzheimer’s disease: a systematic literature review of current trends, methodologies, challenges, innovations, and future directions. Artif. Intell. Rev.58, 44 (2025). [Google Scholar]
- 44.Heidari, A. Jafari Navimipour, N. & Unal, M. The history of computing in Iran (Persia)—since the Achaemenid Empire. Technologies Vol. 10 (2022).
- 45.Amiri, Z., Heidari, A., Jafari, N. & Hosseinzadeh, M. Deep study on autonomous learning techniques for complex pattern recognition in interconnected information systems. Comput. Sci. Rev.54, 100666 (2024). [Google Scholar]
- 46.Jamali, M. A. J. Bahrami, B., Heidari, A. Allahverdizadeh, P. & Norouzi, F The IoT Landscape. Towards the Internet of Things 1–8 (2019).
- 47.Dalirinia, E., Jalali, M., Yaghoobi, M. & Tabatabaee, H. Lotus effect optimization algorithm (LEA): A lotus nature-inspired algorithm for engineering design optimization. J. Supercomput.80(1), 761–799 (2024). [Google Scholar]
- 48.Karami, H., Anaraki, M. V., Farzin, S. & Mirjalili, S. Flow direction algorithm (FDA): a novel optimization approach for solving optimization problems. Comput. Ind. Eng.156, 107224 (2021). [Google Scholar]
- 49.Drias, H., Bendimerad, L. S. & Drias, Y. A three-phase artificial orcas algorithm for continuous and discrete problems. Int. J. Appl. Metaheuristic Comput. (IJAMC)13(1), 1–20 (2022). [Google Scholar]
- 50.Falahah, I. A., Al-Baik, O. Alomari, S., Bektemyssova, G., Gochhait, S., Leonova, I., Malik, O. P., Werner, F. & Dehghani, M. Frilled Lizard Optimization: A Novel Nature-Inspired Metaheuristic Algorithm for Solving Optimization Problems. (2024).
- 51.Finn, C., Tan, X. Y. Duan, Y. Darrell, T. Levine, S & Abbeel, P. Deep spatial autoencoders for visuomotor learning. In 2016 IEEE International Conference on Robotics and Automation (ICRA) 512–519 (2016).
- 52.Opoku, M., Weyori, B. A., Adebayo, F. A. & Adu, K. SFFT-CapsNet: Stacked Fast Fourier Transform for Retina Optical Coherence Tomography Image Classification using Capsule Network. International Journal of Advanced Computer Science and Applications. 14(9) (2023).
- 53.Sarkar, R. Hazra, A., Sadhu, K. & Ghosh, P. A novel method for pneumonia diagnosis from chest X-ray images using deep residual learning with separable convolutional networks. In Computer Vision and Machine Intelligence in Medical Image Analysis: International Symposium, ISCMM 2019, 1–12. (2019).
- 54.Zhao, J., Liu, X. & Tian, M. An efficient task offloading method for drip irrigation and fertilization at edge nodes based on quantum chaotic genetic algorithm. AIP Advances 14 (2024).
- 55.Teja Santosh, D. et al. "Development of IoT based intelligent irrigation system using particle swarm optimization and XGBoost techniques. Bullet. Electric. Eng. Inf.13(3), 1927–1934 (2024). [Google Scholar]
- 56.Singh, N., Kethavath Ajaykumar, L. K. & Dhruw, B. U. C. "Optimization of irrigation timing for sprinkler irrigation system using convolutional neural network-based mobile application for sustainable agriculture. Smart Agricult. Technol.5, 100305 (2023). [Google Scholar]
- 57.Jenitha, R. & Rajesh, K. Intelligent irrigation scheduling scheme based on deep bi-directional LSTM technique. Int. J. Environ. Sci. Technol.21, 1905–1922 (2024). [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
The dataset utilized for the smart irrigation system is described below: Greenhouse data experiment drip irrigation 2016: Using https://data.4tu.nl/articles/dataset/Greenhouse_data_experiment_drip_irrigation_2016/12708971 this dataset is accessed on 2025–01-03. This drip irrigation dataset consists of parameter details like humidity, temperature, external rainwater intake, drain water flow and water supply.