teaLeafBD: A comprehensive image dataset to classify the diseased tea leaf to automate the leaf selection process in Bangladesh

Abstract

Tea is an extremely popular beverage around the world due to its exquisite taste and flavor. Unfortunately, it is prone to different types of illness, which can reduce the amount of harvest along with its standard. Among these, leaf infections are a serious concern since they negatively affect the quality of tea leaves. As a consequence, tea producers often encounter a great deal of obstacles and financial losses. Keeping this in mind, a thorough dataset has been compiled, which contains 5278 images of diseased and healthy leaves. The purpose of this dataset is to improve our knowledge of how these conditions impact cultivating tea plants and tea production. These images are collected from a variety of locations and meteorological circumstances, which provide an extensive knowledge of the disease patterns unique to tea leaves. The pictures have been captured with the help of some high-quality devices from different angles and in high resolution to ensure the standard and increase the usability of the dataset. Rigorous steps were followed when preparing the dataset that would be of great help in building a precise artificial intelligence model. The dataset carefully determined and classified six tea leaf diseases: Tea algal leaf spot, Brown Blight, Gray Blight, Helopeltis, Red spider, and Green mirid bug. There is one more class in the dataset containing images of healthy leaves. These illnesses are known for their devastating impact on tea leaves. An automated disease classification system can be made utilizing deep learning techniques that will enable estate managers to take timely action to stop the spread of the disease, and this meticulously collected dataset will immensely help to train that model.

Specifications Table

Subject	Computer Sciences
Specific subject area	Agriculture, Tea leaf disease, Disease Classification, Artificial Intelligence, Image Processing.
Type of data	Image Raw, Filtered, and Processed
Data collection	The data was collected in 2024 during the monsoon season. Eight renowned tea gardens in Bangladesh were chosen for this purpose. Firstly, all the leaves were collected from the garden, and then the images of those leaves were captured to produce the dataset. The sample leaf contains six types of leaf disease along with healthy leaves. Finally, 5278 images were selected, which were divided into seven classes. To capture the images of the leaves, four types of mobile phone cameras were used.
Data source location	The tea gardens are located in Sylhet and Sreemangal, Bangladesh. The exact locations, along with their precise latitudes and longitudes, are provided as follows: 1.Malnicherra Tea Estate in Sylhet (latitude: 24.93029, longitude: 91.86702), 2.Lakkatura Tea Garden in Sylhet (latitude: 24.91194978, longitude: 91.88657255), 3.National Tea Company Limited in Sylhet (latitude: 24.14403825, longitude: 91.4833438), 4.Zareen Tea Estate in Sreemangal (latitude: 24.28857566, longitude: 91.77168974), 5.Nurjahan Tea Estate (M. Ahmed Group) in Sreemangal (latitude: 24.29451732, longitude: 91.80275007), 6.National Tea in Sreemangal (latitude: 24.30795976, longitude: 91.73760171), 7.Finlay Tea Estate in Sreemangal (latitude: 24.30357177, longitude: 91.74245382), 8.Jungle Bari Tea Estate in Sreemangal (latitude: 24.25329433, longitude: 91.77409053)
Data accessibility	Repository name: Mendeley Data Data identification number: 10.17632/744vznw5k2.4 Direct URL to data: https://data.mendeley.com/datasets/744vznw5k2/4
Related research article	None

1. Value of the Data

• •The dataset was collected from tea harvesting areas in Bangladesh, which is one of the top tea-producing regions in the world, supplying tea globally. A total of 5278 images were captured by the camera from eight tea gardens, and they were annotated by human experts.
• •Seven deadly diseases attacked the Bangladeshi tea plantation, and the data set covers all of them. This will allow researchers to create automated systems to classify this illness instantly to assist estate managers in getting better harvests.
• •This dataset size surpasses existing tea leaf datasets. We know that tea leaf disease can vary depending on the region. That makes this dataset more valuable. Even the tea leaf dataset collected from other countries did not address as many diseases that have been covered in this dataset.
• •The image quality of this dataset is also very high; no leaf has been captured twice. Every image is of a distinct tea leaf. Training deep learning models with this dataset would make them more robust as they would be trained on more unique images. Moreover, if the researchers like to expand this dataset in the future, they can apply image augmentation techniques like- rotations, flipping, adding noise, or cutMix to repopulate the dataset. Besides, there are ways to generate synthetic data using GANs or diffusion models to create entirely new images.
• •It is taken into consideration that this leaves datasets belonging to Bangladeshi tree Gardens, but as Bangladesh has most of the kinds of tea grown here, this data can be used in other countries for disease classification as well. Especially for regions that have similar humid environments like Bangladesh, which have similar tea leaf diseases.
• •Tea is the most popular beverage in South Asia. As Bangladesh is one of the biggest tea-producing countries in Asia, many rural people work as tea laborers and perform challenging tasks to harvest tea leaves, ensuring the highest quality of tea. This intensive process is largely manual and can be very demanding. Hence, implementing an automation system to classify unhealthy tea leaves would be helpful for both the laborers and the manufacturing companies. To ensure the maximum profit with optimized manual labor, a computer vision-based model can easily process the teaLeafBD dataset and deploy such AI-controlled machines in the tea industry. [20]
• •This dataset can enable the development of robust models for automated plant disease classification by supporting core tasks such as image classification, segmentation, and anomaly detection. To provide a valuable resource for fine-tuning pre-trained models, transfer learning is a well-suited method for this dataset. The dataset can be used to evaluate model performance, serving as a standardized benchmark that fosters reproducibility and comparative analysis across different algorithms.

2. Background

Crop leaf disease classification is a vastly important aspect of the agricultural sector, and historically, it required labor-intensive manpower, and most of the time, those results were faulty. Thus, to solve this issue, the researchers are inventing various kinds of computer-based technologies for the sake of the agriculture sector. One common approach is mapping using previously developed data sets using machine learning algorithms. But for more accurate results, this dataset was collected from local tea gardens since Bangladesh has one of the richest tea plantation areas in Asia. The motivation behind this research was to create a large dataset that covers all the prominent tea leaf diseases. Fig. 1 shows the images of eight tea gardens from where the leaf samples were collected. Except for the last one [1], all the images of the tea gardens were captured by the data collection team.

Fig. 1.

Research data collection locations.

Existing datasets on tea leaf diseases found in Bangladesh are both small and do not contain all the diseases. It shows that the previously published datasets [2,3] in Bangladesh have not covered all the possible ailments of tea leaves. The TeaDiseaseNet [18] dataset contains six types of illness and has only 776 images. The tea sickness dataset [16] incorporated eight different classes in its dataset where the total number of images is 885. The Indian dataset [17] contained six classes, but more classes can be introduced. We aim to address that gap and create a comprehensive dataset comprising seven different sicknesses of tea leaves. Experienced team members collected samples, and pictures were captured with the cameras of the smartphone. This dataset can be utilized to create an automated system for classifying tea leaf diseases, which would help the estate managers to save crops from common diseases in the early stages, tea leaf plucking sessions or tea leaf grading, sorting, and screening sessions, and increase profits by increasing the production. Table 1 provides a summarized comparison between the existing datasets and the teaLeafBD dataset.

Table 1.

Comparison with existing work.

Source	Number of Classes	Name of the dataset	Location	Sample Size
				Original	Augmented
Kaggle [16]	8	Tea sickness dataset	Kenya	885	-
scientific reports [2]	5	-	Bangladesh	4000	-
ScienceDirect [17]	6	-	India	5867	-
scientific reports [3]	4	-	Bangladesh	3330	-
frontiers [18]	6	TeaDiseaseNet	China	776	7640

The prevalence of tea disease can vary from region to region. The Bangladeshi dataset [3] contains 3330 images but only for four classes, more classes have been introduced in our dataset. The dataset is more varied and robust when there are more classes. teaLeafBD dataset contains a total of seven classes with a significant quantity of original images. Among them, there are six disease classes and one class of healthy leaves. Since tea leaf diseases’ prevalence varies, creating a balanced tea leaf disease dataset is challenging. The dataset contains imbalanced classes like others [3] but the class imbalance can be addressed by applying appropriate augmentation techniques like they did for the TeaDiseaseNet [18] dataset.

3. Data Description

3.1. Dataset categories

Identifying tea diseases manually is a time-consuming process and requires expertise. To solve this problem, it is imperative to have a system that can classify tea diseases automatically. Using image processing and deep learning, researchers have created AI models that can automatically identify tea diseases. For this purpose, a quality image dataset is required. In this section, the analysis focuses on various diseases of tea leaf images from the dataset, identifying seven different categories: Algal leaf spot, Brown blight, gray blight, Helopeltis, Red spider, Green mirid bug, and Healthy leaves.

3.2. Algal leaf spot

Tea algal leaf is a very common ailment of tea leaves. Fig. 2 shows the condition of the tea leaves for this disease.

Fig. 2.

Sample of tea leaves having algal leaf spot disease.

Major symptoms of Algal Leaf Spot are:

• •When a leaf is affected by this illness, brown lesions with irregular shapes can be noticed.
• •The brown lesions can even expand up to 10mm, with brown to gray color changes in the spot centers and dark brown edges.
• •The disease can manifest as circular or irregular brown spots, initially appearing as shrinking and yellowing on the leaves [6]

3.3. Brown blight

Tea Brown blight disease is one of the most prevalent conditions affecting tea leaves. Fig. 3 shows the state of the tea leaves for that ailment.

Fig. 3.

Sample of tea leaves having brown blight disease.

Major symptoms of Brown Blight are:

• •It is caused by a severe foliar fungal infection (Camellia sinensis)
• •The primarily responsible species is the Colletotrichum genus, with Colletotrichum camelliae being the most prominent pathogen.
• •This disease is characterized by the appearance of small, brownish-to-black spots on young tea leaves; these usually expand, darken, and coalesce, leading to significant leaf necrosis.
• •With passing time, they form large necrotic areas, which causes the leaves to wither and drop prematurely.
• •Initial symptoms include small, water-soaked brown spots that enlarge and darken over time.
• •Besides Colletotrichum camelliae, other species, like C. fructicola and C. aenigma, have also been identified as responsible for this disease. These fungi thrive in warm, humid environments, making tea plantations in tropical and subtropical regions particularly vulnerable. [7,8]

3.4. Gray blight

Another very common illness is Gray Blight. The condition of the tea leaves for that disease is presented in Fig. 4.

Fig. 4.

Sample of tea leaves having gray blight disease.

Major symptoms of Gray Blight are:

• •A pathogen named Pestalotiopsis theae is to blame for this.
• •Mostly affects premature leaves; the infected leaves eventually fall off
• •This disease is recognized by grayish-brown lesions that appear on tea leaves, often starting as small spots that expand to cover larger areas, leading to leaf blight and necrosis.
• •Gray blight presents as grayish lesions with distinct brown margins on the leaf surface. As the disease progresses, these lesions coalesce, leading to large necrotic patches.
• •Pestalotiopsis theae fungus thrives in humid and warm climates. The pathogen's spores germinate on the leaf surface, penetrating the tissue and causing infection [9,10].

3.5. Helopelties

Helopeltis disease, commonly known as the tea mosquito bug infestation, is a significant pest problem affecting tea plants. It is displayed in Fig. 5.

Fig. 5.

Sample of tea leaves having heliopolis disease.

Major symptoms of Helopelties are:

• •The primary culprits are species of the Helopeltis genus, particularly Helopeltis theivora, Helopeltis antonii, and Helopeltis bradyi.
• •The nymphs and adult bugs suck sap from the tea leaves, buds, and young shoots, which leads to brownish or blackish spots on the leaves.
• •The affected areas may develop into necrotic patches, causing the leaves to curl, wither, and fall off prematurely.
• •The life cycle of these bugs involves eggs being laid on tender plant parts, from which nymphs hatch and eventually develop into adults [11,12].

3.6. Red Spider

Red spider disease in tea plants is primarily caused by the red spider mite, Oligonychus coffeae, which is a major pest in tea-growing regions worldwide. Fig. 6 illustrates the effect of the disease on tea leaves.

Fig. 6.

Sample of tea leaves having red spider disease.

Major symptoms of Red Spider are:

• •The red spider mites attack the tea leaves by lacerating the cells on the upper surface, leading to the formation of minute reddish-brown marks.
• •As the infestation intensifies, the leaves turn red and eventually dry out, resulting in premature leaf drop.
• •The mites are small, red, and barely visible to the naked eye. They tend to colonize the undersides of tea leaves, where they spin webs and feed, leading to characteristic red spots on the leaves.
• •Infestation is usually more severe during dry seasons when the mites multiply rapidly.
• •The mites have a short life cycle, allowing them to reproduce rapidly. They lay eggs on the undersides of leaves, which hatch into larvae, and within a few days, the larvae mature into adult mites, capable of further reproduction.
• •In severe infestations, up to 40-50% yield loss has been reported. The leaves become brittle and lose their flavor and aroma, impacting the overall market value of the tea [13,14].

3.7. Green mirid bugs

• Green mirid bugs, particularly species like Apolygus lucorum and Pachypeltis maesarum, are significant pests in tea plantations, leading to substantial damage to tea plants. Fig. 7 represents the leaves affected by green mirid bugs.

• Major symptoms of Green Mirid Bugs are:

• •These bugs belong to the family Miridae and are known for their piercing-sucking mouthparts, which they use to feed on plant sap.
• •The green mirid bugs pierce tea leaves, buds, and shoots, causing discoloration, deformities, and necrosis in the affected areas. This damage is often visible as brown or black spots on the leaves, which can lead to premature leaf drop.
• •The feeding activity of these bugs can severely stunt the growth of young tea shoots, reducing the overall productivity of the plant.
• •These bugs are typically more active during certain times of the year, with peak infestations occurring in warmer months.
• •They can also migrate between tea plants and nearby weed hosts, further complicating control efforts.
• •The bugs lay eggs on tea plants, from which nymphs emerge and begin feeding immediately. Both nymphs and adults cause significant damage, and their rapid reproductive cycle allows for quick population build-up [15].

Fig. 7.

Sample of tea leaves having green mirid bug disease.

3.8. Healthy

Healthy tea leaves are essential for producing high-quality tea. They exhibit strong growth, vibrant color, and no signs of pest damage or disease. Fig. 8 represents the healthy leaves.

Fig. 8.

Sample of healthy tea leaves.

Key characteristics of healthy leaves are:

• •Vibrant green in color with a smooth, uniform surface, free from spots or discoloration.
• •Firm and turgid texture, without wilting, curling, or necrotic patches.
• •Well-developed new shoots and buds, showing consistent and vigorous growth.
• •No visible insect feeding marks, such as punctures, brown/black spots, or sap leakage.
• •Leaves remain firmly attached to the plant, with no premature dropping.

Data Directory Details:

Table 2 represents the number of images for every class in this dataset. There are seven classes in the dataset, which are Algal leaf spot, Brown blight, Gray blight, Helopeltis, Red spider, Green mirid bug, and Healthy leaf. Analysis reveals that the class with the least number of images includes more than 418 photographs, which underscores the extensive variety inherent in the dataset. The overall dataset consists of 5278 images.

Table 2.

Number of images for each class.

Dataset
Class	Category / Folder	Number of images
1	Tea algal leaf spot	418
2	Brown Blight	508
3	Gray Blight	1013
4	Helopeltis	607
5	Red spider	515
6	Green mirid bug	1282
7	Healthy leaf	935
Total		5278

As shown in Fig. 9, the dataset has been systematically organized to make it easier to apply different machine-learning models. The primary directory is designated as 'teaLeafBD,' which contains all associated subfolders. Each subfolder is named after the classes available in our dataset, and pictures of each class are inside those subfolders.

Fig. 9.

Folder structure of the tea leaf dataset.

All the images in the dataset repository are raw and original. The format of the image files is .jpg format. On average, the size of each file is 235 KB, which makes the size of the repository 1.24 GB. The overall summary is mentioned in Table 3.

Table 3.

Summary of teaLeafBD dataset.

Parameter	Value
Garden	8
Class	7
Total Images	5278
Format of Images	JPG
Image Resolution	1200 px x 1600 px
Approximate Image Size	235 KB

A Readme.pdf file is attached to the root directory to provide researchers with basic information about the dataset.

4. Experimental Design, Materials and Methods

Farmers and plant specialists have historically diagnosed plant diseases by visually inspecting the plants. However, the recent flurry of technological advancements has made it feasible to deploy technologies to help humans diagnose a variety of plant diseases effectively in a comparatively shorter amount of time and at a lower cost. The field of machine learning and deep learning stand out among today's emerging AI technologies because, given enough relevant historical data, these methods may make remarkably accurate predictions. These algorithms are effectively applied in a variety of human areas, including agriculture, via the use of inexpensive computer hardware [5]. A quality dataset is essential for the effective application of machine learning.

A dataset is an integral part of any research. A diverse dataset is necessary to conduct a comprehensive study. Especially when it comes to leaf disease classification, a dataset containing high-quality images is a must. This makes data collection one of the most critical phases of the research. The overall data collection flowchart is illustrated in Fig. 10. All these steps are described in the following sub-sections.

Fig. 10.

Data collection flowchart.

Visiting Tea Gardens:

This project was initiated with an in-depth study of existing work on the diagnosis of tea leaf diseases, demonstrating the importance of having an extensive knowledge of each disease. Diseases like tea algal leaf spot and gray blight are quite common occurrences in tea plants in Bangladesh, and their symptoms are well documented. Except to provide a comprehensive analysis and precise diagnosis, we broadened the search to incorporate uncommon diseases like Helopilitis and Green mirid bug. To provide a thorough and knowledgeable data collection process, this elaborate plan includes a thorough search for precise information regarding these conditions. A key point to remember is that the intensity of disease may vary from area to area; diseases that are quite common in tea gardens in one area might not be that prominent in tea gardens from another area. The research team went to tea gardens and consulted with the estate managers and the tea pickers to gather more information to learn from their experience. Geographic mapping was among our initiatives to demonstrate the area's tea garden concentration, assuring that the sample was both concentrated and indicative of the circumstances in the area. More than 150 tea gardens, including three of the world’s largest in terms of both area and production, are located in Sreemangal and Sylhet [4]. Sreemangal known as the “Tea Capital” of Bangladesh is encircled by verdant tea terraces. These vast amounts of gardens can provide variation in diseases. Thus, these gardens were selected by the data collection team to compile an adequate dataset and obtain more varied and extensive data.

Collecting Tea Leaves:

According to findings, Helopilitis and the Green mirid bug are extremely rare in the tea gardens of Sylhet. Compared to them, there is a high prevalence of both diseases in tea gardens located in Sreemangal. This discrepancy is due to the changes in environment and soil quality along with other factors in both regions that make the tea plants of both areas prone to different kinds of illnesses. It gives us a great chance to investigate the dynamics of pathogen infection and disease dissemination in detail. The samples that were collected varied on multiple aspects such as age and environmental conditions. To maintain their freshness, the leaves were meticulously removed from the tree and placed in separate airtight containers for each class for further process.

Consulting with the Experts:

Interacting with local tea pickers provided us with fascinating insights. Their assistance allowed us to collect data efficiently. Local estate managers and tea pickers shared their experiences and the effects of diseases on their sources of income. Determining the disease of tea leaves is an extremely difficult task. We consulted with the local agriculture officers to determine the diseases as well. Based on the knowledge from previous research, photographs of affected tea leaves were captured first. Subsequently, the annotation quality was re-evaluated by presenting the annotated dataset to experts. It is anticipated that the knowledge gained from this project will greatly advance the creation of precise, focused methods for protecting crops and disease control. Eventually, it will help to improve crop conditions and farming methods in the region.

Preparing a Well-lit Environment:

Upon the identification of the diseases, we proceeded to take methodical, expert images of the infected leaves. To do that, setting up a well-lit environment was important. The process began by selecting tea leaves carefully that showed any of the diseases indicated above and setting them against a white paper screen. Ensuring images do not get oversaturated by the natural sunlight when taking photos posed quite a challenge for us. To handle this issue, we employed the required tools and carefully controlled the lighting to preserve image quality. Accurately colored, high-quality images are essential for early disease classification in tea plant leaves. As such, these photos must be taken in the best possible setting. To prevent excessive brightness and shadows, a controlled testing space with modest lighting was created. Additionally, a white background accentuated the leaves' organic surface and lessened sharp color contrasts, guaranteeing image stability.

Taking Photos of the Leaves:

This section explores the subtleties of the devices used and the process of taking images. The purpose of this arrangement was to reduce background noise so that the leaf and its state would be the only thing on display. We leverage a handful of techniques to distinguish each disease in the dataset to correctly present the healthy leaf class. The images of healthy leaves were captured alongside infected leaves to ensure consistency. To minimize biased model learning and sampling at random, keep the dataset homogeneous throughout all classes. This will help eliminate bias towards characteristics. Altogether, our combined efforts guarantee the correct representation and distinction of healthy leaf classes within the dataset. Four Android smartphones were employed for taking the photographs, each selected for its excellent camera features and capacity to generate high-quality images. Table 4 is a thorough overview of these devices' specifications, including camera resolution and other relevant aspects.

Table 4.

Device details.

Device Name	Camera Details
Oppo Reno8 Pro	50 MP, f/1.8, 23 mm (wide), 1/1.56", 1.0 µm, multi-directional PDAF
Realme 6i	48 MP, f/1.8, 26mm (wide), 1/2.0", 0.8 µm, PDAF
Redmi 12	50 MP, f/1.8, (wide), PDAF
Realme Narzo 50	50 MP, f/1.8, 26 mm (wide), 1/2.76", 0.64 µm, PDAF

In addition to enabling a wide variety of photos, this quad-device strategy guaranteed redundancy, protecting against possible data loss or quality problems. By using the Android devices overall, 5278 high-quality pictures of tea leaves were captured with a ratio of 3:4. All images were captured during daylight with a consistent camera setup. The camera was positioned approximately 0.3 m from the subject, with a 90° angle of capture (approximate). These parameters remained uniform across the entire dataset. These leaves were carefully positioned against a white background to retain image quality and make it easier for machines to interpret the images during later processing phases. This arrangement made it possible to pinpoint the features of the leaf, ensuring that clear, accurate photos had been snapped. To guarantee the consistency and flawlessness of the photographs, an advanced camera configuration was used to capture images with accurate color gradients. We gathered the leaves and captured images of them in a controlled setting with less environmental disturbance to reduce environmental influences like strong winds. Ultimately, a range of image-capturing angles was employed to guarantee comprehensive documentation and uniform quality, hence facilitating efficient image analysis.

Discarding low-quality images and rescaling to uniform dimensions:

The final step was carefully examining the photos that were taken to see which ones would work best for machine learning and deep learning applications. For those algorithms to classify the diseases accurately, it was vital to use images that accurately depicted the conditions. Following the selection procedure, we methodically categorized and arranged the photos, placing them in appropriate folders. By making it easier to find specific leaf disease photos, this organizational technique greatly improved the efficiency of data retrieval and usage. To optimize the dataset for machine learning model training, Necessary adjustments were made to the photos during the editing stage to ensure consistency and clarity. In the process of arranging the dataset folders, the quality of every image is evaluated. Any low-quality images, including those that are blurry, oversaturated, or otherwise flawed, are removed. Following the elimination of low-quality images, the dataset consists of 5278 high-quality images. Rescaling is then applied to ensure that all images have uniform dimensions.

Validating the Credibility of the Dataset:

To assess the credibility of the teaLeafBD dataset, two popular pre-trained CNN models- InceptionV3 and Densenet201 have been trained using this dataset. Fig. 11 displays the confusion matrix for the InceptionV3 and DenseNet201 models, with DenseNet201 demonstrating better performance. The actual classes are shown in each row, and predicted classes are in each column. The diagonal values indicate correctly classified instances, and the misclassifications are represented by off-diagonal.

Fig. 11.

Confusion matrix of InceptionV3 and DenseNet201.

Nonetheless, with minimal tuning, these models have achieved 84.28 % and 89.39 % accuracy with this dataset. Therefore, this suggests that this dataset contains meaningful, distinct classes and image quality, and will be able to support other machine learning or deep learning models. Table 5 illustrates the summary of the experiments on our dataset using two CNN models. Table 5. Summary of the Result of the Models.

Table 5.

Model summary.

Class	InceptionV3			DenseNet201			Support
	Precision	Recall	F1-Score	Precision	Recall	F1-Score
1. Tea algal leaf spot	0.7179	0.6364	0.6747	0.8222	0.8409	0.8315	44
2. Brown Blight	0.6500	0.5909	0.6190	0.8333	0.7955	0.8140	44
3. Gray Blight	0.8120	0.8559	0.8333	0.8909	0.8829	0.8869	111
4. Helopeltis	0.8980	0.8000	0.8462	0.8846	0.8364	0.8598	55
5. Red spider	0.8478	0.8298	0.8387	0.8864	0.8298	0.8571	47
6. Green mirid bug	0.9000	0.9000	0.9000	0.9098	0.9308	0.9202	130
7. Healthy leaf	0.8972	0.9897	0.9412	0.9412	0.9897	0.9648	97

Limitations

• •Data Collection Period: Rain is an extremely important factor for tea harvesting, and monsoon is the season when tea production is at its highest. Almost 80 % of tea is produced in the six months from June to November [19]. Month-wise tea production amount from 2018 to 2020 reveals that May–November is the period where most of the tea is produced. Since this is the period for most production, the occurrences of different diseases are more likely to happen. Considering this, we decided to collect data during May and June rather than all year round. This may result in some bias in the quality of the leaves of the captured images.
• •Bias in Sampling: The data was collected in controlled indoor conditions like consistent lighting, background, and camera settings. In the real-world scenario, such ideal conditions may not always be seen. As a result, the model trained with this dataset may face performance degradation when trying to classify images that are captured in an uncontrolled, outdoor environment.
• •Leaves having Multiple Diseases: Some of the collected leaves have multiple diseases, i.e., exhibit the traits of multiple classes. This can cause ambiguity in the ability of the learning process of the AI models.

Ethics Statement

The authors adhere to the journal's ethical guidelines and confirm that this research does not involve humans, animals, or data obtained from social media. The datasets utilized in the study are publicly accessible, and appropriate citation protocols should be followed when utilizing these datasets.

CRediT Author Statement

B. M. Shahria Alam: Data collection, Investigation, Methodology; Fahad Ahammed: Supervision, Writing – original draft, Writing – review & editing, Visualization, Validation; Golam Kibria: Data collection, Investigation, Writing – original draft; Mohammad Tahmid Noor: Investigation, Data collection, Writing – original draft, Omar Faruq Shikdar: Methodology, writing – original draft, writing – review & editing; Kazi Isat Mahzabin: Writing – original draft; Nishat Tasnim Niloy: Project administration, Validation, Supervision, writing – review & editing; Md. Nawab Yousuf Ali: Project administration, Validation, Conceptualization, Supervision.

Data Availability

• Mendeley DatateaLeafBD (Original data).