An automated in-field transport and imaging chamber system for high-throughput phenotyping of potted soybean

Abstract

Background

In major soybean-growing regions worldwide, vertical (three-dimensional) planting systems are widely adopted. Achieving precise phenotyping of individual soybean plants is crucial for breeding shade-tolerant cultivars and optimizing high yields. However, canopy shading from taller crops severely restricts the acquisition of phenotypic information from the lower-growing soybeans, and conventional phenotyping platforms struggle to meet the demands of such complex planting structures. To address this challenge, this study developed a field-based high-throughput phenotyping platform specifically designed to accommodate the structural characteristics of vertical planting systems.

Results

The platform integrates the characteristics of vertical planting systems and consists of an imaging system and a rail-based transportation system.The imaging system balances the growth requirements of soybeans under natural conditions with the stability of indoor imaging, and is equipped with adjustable sensors, an automated rotating stage for image capture, and modules for image classification and storage. The transportation system includes X and Y dual-directional tracks and programmable rail carts, enabling automated movement of potted soybean plants in the field. Platform performance was validated through correlation analysis and predictive modeling. The extracted plant height and width showed high agreement with manual measurements, with coefficients of determination (R²) of 0.99 and 0.95, respectively. During the vegetative stage, the predictive accuracy (R²) for canopy fresh weight and leaf area reached 0.965 and 0.972, demonstrating strong predictive performance and robustness. In addition, the platform supports modular sensor integration and features an open-source control architecture, allowing seamless incorporation of additional sensors such as infrared cameras, LiDAR, and fluorescence imaging. This expands trait detection capacity while reducing costs for reuse and secondary development.

Conclusion

This study demonstrated the feasibility of combining natural field conditions with standardized indoor imaging for phenotypic research on soybeans under vertical planting systems. The platform provides a flexible and scalable technical solution for analyzing plant architecture and screening germplasm in complex planting environments, opening up new technological pathways for precision agriculture and crop breeding research.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13007-025-01424-2.

Background

Soybean is a globally significant crop, serving as a source of food, oil, feed, and vegetables, with trade volumes ranking first among agricultural products. According to the United States Department of Agriculture, China imports approximately 85% of its soybean supply annually. To improve land use efficiency and increase soybean self-sufficiency, vertical planting systems are considered a potentially effective approach to boosting total soybean production [1]. Surveys indicate that major soybean-producing regions worldwide, including the United States, China, Australia, and more than ten other countries, commonly adopt mixed intercropping and relay cropping systems, interplanting soybeans with cassava [2], wheat [3–5], maize [6], sugarcane [7], sorghum [8, 9], sunflower [10], and other crops. Notably, soybeans often function as a low-canopy crop in these mixed cropping systems, making them susceptible to shading from high-canopy crops, which reduces light absorption efficiency, alters phenotypic traits, and adversely impacts yield [11].

Phenotypic monitoring is a crucial tool for assessing soybean productivity. Traditional phenotypic evaluations rely heavily on experience or destructive sampling, which is time-consuming, labor-intensive, and prone to uncertainty. With the rise of precision agriculture, advances in non-destructive monitoring technologies have provided innovative means of assessing crop growth conditions, particularly in phenotyping, which holds immense potential for application. According to the International Plant Phenotyping Network (IPPN), countries such as the United States, Australia, China, and Germany have made significant investments in field phenotyping platforms. Currently, there are nearly 200 large-scale phenotyping facilities in operation worldwide, including approximately 82 indoor mechanized platforms, 81 field mechanized platforms in Europe, while similar data for Asia is yet to be compiled. These facilities have laid a solid foundation for crop phenotyping research. The field environment, with its open and realistic conditions, better simulates the complexities of natural crop growth, making it invaluable for high-throughput phenotyping studies. However, among the global high-throughput phenotyping platforms, only 25% are used for field research, and of these, only 44% are genuinely utilized for high-throughput data collection [12].

Field phenotyping platforms can be categorized based on sensor deployment methods into self-propelled, vehicle-mounted, gantry, cable-suspended, unmanned aerial vehicles (UAVs), aerial remote sensing, and satellite-based systems [13]. Self-propelled platforms are driven by powered systems, enabling flexible phenotypic data collection across different growth stages but are less resistant to interference and have lower efficiency [14]. Vehicle-mounted platforms are suitable for low-canopy crops or early growth stages but may disrupt crop growth during operation [15]. Gantry platforms combine tracks and mobile sensor systems to provide stable, high-resolution data collection around the clock, but are expensive and have limited perspectives [16]. Cable-suspended platforms offer continuous operation with good repeatability but are restricted by coverage areas and adverse weather conditions [17]. UAVs and satellite platforms are efficient and cost-effective for large-scale phenotypic monitoring, but their resolution and sensor load capacity remain limiting factors [18].

Although existing platforms are diverse and their performance continues to improve, they still face significant challenges in vertical planting systems. In such systems, taller crops (such as maize) cause substantial shading of lower crops (such as soybeans), and factors like narrow row spacing and complex plant architecture further limit the proximity of phenotyping equipment and affect imaging quality. For example, self-propelled and vehicle-mounted platforms encounter difficulties operating stably between maize rows due to limited path width and vibration interference, which hampers precise imaging of individual soybean plants. Gantry-type platforms, while stable, are limited by their fixed overhead viewing angle, making it difficult to capture the three-dimensional canopy structure of soybean. Although UAV- and satellite-based platforms offer advantages in terms of spatial coverage and monitoring efficiency, their limited resolution and canopy penetration capacity make them insufficient for capturing detailed traits of low-stature crops at the individual level. Moreover, most existing platforms lack multi-angle and multi-stage imaging capabilities, limiting their ability to conduct continuous and systematic monitoring throughout the entire crop growth cycle. This significantly constrains the analysis of dynamic spatiotemporal changes in soybean phenotypes.

To address these challenges, this study systematically analyzed the performance characteristics and application scenarios of mainstream field-based platforms, and on this basis, developed a high-throughput phenotyping platform that integrates field rail-based transportation with standardized indoor imaging, making it particularly suitable for vertical planting environments. Table 1 provides a comparative summary of the adaptability of existing phenotyping platforms in intercropping environments and highlights the technical advantages of the proposed system.

Table 1.

Comparative adaptability of representative field phenotyping platforms in intercropping environments

Platform Type	Application Scenarios and Structural Features	Limitations in Intercropping Environments	This Study’s Strategies and Innovative Advantages
Self-propelled Platform	Autonomous mobility; suitable for uniformly planted open fields	Obstructed by maize rows, limited path selection, severe occlusion	Fixed rail system with automated transport cart to avoid occlusion from tall crops
Vehicle-mounted Platform	Suitable for early-stage, low-stature crops; relatively efficient data collection	Likely to disturb root systems; vibrations affect imaging stability	Use of potted plant transport and stabilized rotational imaging stage to minimize interference
Gantry-type Platform	High-resolution, all-weather monitoring	High cost, fixed angles, limited ability to capture diverse soybean canopy structures	Rotational stage with adjustable viewing angles and standardized lighting for structural detail capture
Cable/UAV/Satellite Platforms	Wide spatial coverage; ideal for macro-scale phenotyping	Insufficient resolution and canopy penetration; difficult to resolve individual	Integration of field conditions with indoor high-precision imaging to capture fine structural traits
This Study’s Platform	Specifically designed for strip intercropping; integrates transport and imaging	—	Enables automated, non-destructive, high-reproducibility imaging of individual plants across full growth stages

In summary, existing platforms still face critical technical limitations in adapting to the complex conditions of vertical planting systems, including severe canopy occlusion, difficulty in individual plant recognition, and insufficient imaging precision. To address these challenges, and based on a systematic analysis of crop spatial distribution and growth dynamics under vertical planting models, this study innovatively developed an integrated high-throughput phenotyping system. This platform combines rail-based transportation, a standardized imaging chamber, a rotational imaging stage, and automated image acquisition and management modules. It enables non-destructive, precise, and automated data collection of individual soybean plants throughout the entire growth period, providing strong technical support for shade-tolerance evaluation and high-yield soybean breeding.

Materials and methods

Field soybean plant architecture image acquisition platform: structure and working principl

Overall structure

The field soybean phenotypic acquisition platform consists of two main components: an imaging system and a transportation system. The imaging system’s primary function is to ensure the stability of soybean image collection, while the transportation system enables fully automated transportation of field-grown soybean plants to the imaging chamber.

This study utilized NX10.0 software to design a rail-transport-based potted plant platform for efficient image acquisition of soybean plant architecture in the field. The platform centers around a fixed imaging chamber, which is equipped with a high-precision imaging system. Using the rail transportation system, the platform automatically transports potted soybean plants grown under strip intercropping conditions from the field to the imaging chamber, enabling precise acquisition of phenotypic data. This design effectively integrates the natural growth conditions of soybeans in the field with the stability requirements of indoor imaging. The three-dimensional structural model of the platform is shown in Fig. 1, illustrating its complete functional layout.

Fig. 1.

Overall structural diagram of the platform

Working principle and key technical parameters

To achieve efficient and precise research on the architecture of individual soybean plants under a strip intercropping system, this platform was designed with clear objectives. On the one hand, it ensures that soybeans can grow healthily in the field conditions of strip intercropping while maintaining their phenotypic traits under natural light; on the other hand, it enables stable and accurate phenotypic data collection in a controlled environment. This design effectively avoids data deviations caused by maize shading, light fluctuations, wind and rain interference, and mutual shading among soybean plants.

In addition, the platform emphasizes easy disassembly and portability to accommodate field operations such as land preparation while ensuring that the equipment remains structurally simple, stable, and easy to maintain for long-term use. Specifically, the transport rails consist of single rails and sleepers, with each rail section measuring 2 m in length. Sections are connected with screws and washers, and removing a single screw takes only about 2 s, allowing 2 m of rail to be disassembled in just one minute. The rail cars operate independently, making them easy to transport. The main hardware in the imaging chamber—including the computer, cameras, and light sources—can be fully dismantled in about 30 min. Images of each component and the platform’s operational performance are provided in Supplementary Material 1.

The platform’s 3D model was developed using Siemens NX software, which facilitated structural design of the equipment, sensor layout planning, and operational mode configuration. The platform underwent systematic mechanical design and structural analysis using mechanical design methods and stress-check calculations [19] to verify its operational feasibility. Further finite element analysis and optimization of critical components were conducted using Siemens NX to ensure operational stability and reliability.

The platform integrates industrial automation technology, employing a Programmable Logic Controller (PLC) for centralized control of moving components, enabling fully automated operation. Combined with visual perception technology, the platform achieves automatic plant image acquisition and information exchange, fully meeting research requirements. The main structural parameters of the finalized field-based soybean phenotyping platform are shown in Table 2.

Table 2.

Platform structure parameters

Platform part parameters	Specifications and models
Imaging room dimensions /mm	3000 × 6000 × 2700
Imaging area dimensions /mm	1500 × 1500 × 1500
Dimensions of the tie blocks /mm	1100 × 120 × 75
Camera model (Hikvision)	MV-CH250-90GC
Camera lens model (Hikvision)	MVL-KF2524M-25MP
Turntable main body dimensions /mm	600 × 500 × 3.5
Stepper motor model	SST42D2121
Stepper motor turntable model	NT02RA100
X track vehicle motor specifications	48&1000 W
Y track vehicle motor specifications	48 V&500 W
Square tube cross-sectional dimensions/mm	40 × 60
X track width /mm	600
Y track width /mm	300

Design of the field-based soybean phenotyping platform

Imaging system

The imaging system primarily consists of the imaging chamber, a computer workstation (CPU: Intel i5-10400F, GPU: ASUS GTX1060 6G, RAM: 16GB, storage: 512GB), a QR code scanner, a QR code printer (Han Yin D35, 80 mm), an industrial camera (MV-CH250-90GC), a Hikvision robotic lens (MVL-KF1624M-25MP, focal length: 16 mm, maximum aperture: F2.4, sensor size: 1.2’’, manufactured in Hangzhou, China), and a lighting system.

The camera position can be adjusted automatically, and the final installation height was set at 1500 mm, with a focal length of 2.3 mm. The white balance parameters were set to 2270 × 1024 × 1970, the exposure time was 60,000.00 EV, the aperture value was 2.4, and the frame rate was maintained at ≥ 4.5 fps. Images were stored in JPG format at a resolution of 5108 × 4604. In addition, to ensure the accuracy of image feature extraction, a 30 cm diameter white reference panel was used as a scale reference.

Imaging chamber

The external dimensions of the imaging chamber are 6000 × 3000 × 2700 mm. The design prioritizes ease of transportation and maintenance while providing ample space for internal layout. The imaging area measures 1500 × 1500 × 1500 mm, accommodating soybean plants with a maximum height of approximately 1300 mm, as determined by preliminary survey data. The imaging sensor frame is constructed using 4040 industrial aluminum profiles, which include grooves for mounting camera installation plates. This design enables adjustable camera positioning and rotation along the groove to meet phenotyping requirements for various soybean varieties throughout their growth cycle.

To address challenges posed by harsh environments and long-duration operations, the imaging sensor incorporates a high-performance industrial camera (Hikvision MV-CH250-90GC), ensuring high-quality imaging and reliable equipment performance. Additionally, the imaging chamber is equipped with an automatic sliding door system and a sensor-controlled light-blocking curtain, creating a sealed imaging environment. This is complemented by a global light source providing stable and uniform illumination, significantly enhancing imaging precision and consistency. The internal structure of the imaging chamber is shown in Fig. 2.

Fig. 2.

Schematic diagram of the imaging chamber layout

Imaging control

The primary functions of the imaging system include controlling the imaging sensors, transmitting phenotypic image data, and categorizing and storing the data. Since soybean growth and imaging occur in separate environments, the imaging system must interact with the transportation system for seamless operation. Consequently, the displacement information from the transportation system is converted into a trigger signal for the imaging system, enabling efficient and synchronized control between the two systems.

The core of the imaging system ensures the stable and efficient collection of large volumes of raw images and the automated classification and storage of data. The control program is compiled in the Visual Studio 2019 environment to ensure operational stability. The components and implementation process of the system are illustrated in Fig. 3: ① Identity Encoding: Assigns a unique identifier to each soybean plant, recording details such as variety name and treatment conditions. ② Identity Printing: Prints the identifier as a QR code label, which is then affixed to the corresponding plant. ③ Identity Recognition: When the plant is transported into the imaging area, a scanner automatically detects the QR code label. ④ Triggering Capture: Using OpenCV technology, the frame rate change in the QR code scanner’s data serves as the trigger condition. Upon successful QR code recognition, the camera is triggered via the SDK interface for imaging. ⑤ Capture Control: The host system’s embedded program allows real-time adjustments to imaging parameters, such as timing and the number of images captured. ⑥ Data Acquisition: Sensors collect raw image data. ⑦ Data Storage: Raw images are automatically renamed and transferred to designated folders, ensuring proper data categorization and storage.

Fig. 3.

Imaging system

Through this workflow, the imaging system achieves fully automated capture and classification of soybean plant images, providing efficient and precise data support for subsequent research.

Transportation system

The platform’s transportation system operates over a large area, requiring the transport routes to align with the strip intercropping model. Additionally, the transportation equipment must support route selection, automatic start and stop, and directional changes. The control system must also ensure high stability for prolonged operation under field conditions.

Structure of the transportation system

This study introduced an innovative design that incorporates intersecting X and Y tracks, enabling rapid directional transport of potted soybean plants in the field. Specifically, a movable X track is mounted on the Y track vehicle, allowing the X track to connect seamlessly with multiple X tracks as the vehicle moves (Fig. 4a). This design overcomes the limitations of traditional track systems, which rely on switch rails and curved tracks to achieve directional changes. Such traditional systems often have drawbacks, including structural complexity, large footprint, and high cost (Fig. 4b).

Fig. 4.

Schematic diagram of X-Y track intersection

Ultimately, the transportation system consists of a rail network formed by one Y track and multiple intersecting X tracks, as well as X track vehicles and Y track vehicles that operate along these tracks (Fig. 4c). To enhance durability, the track supports are made of anti-corrosion wood, effectively isolating the system from soil and grass, preventing corrosion, and providing shock absorption for stable operation. This design also facilitates the transportation, installation, and subsequent maintenance of the track system.

The platform’s transport system is driven by the motor of the railcar, with X railcars and Y railcars operating on their respective tracks. The main function of the X railcar is to push the soybean pots on the X track and precisely move them to the Y railcar. The Y railcar is responsible for transporting the soybean pots to the imaging room for shooting and returning to the initial position after the imaging is complete. During this process, the X railcar continues to push the pots, transferring the photographed soybeans to another X track for storage, thereby achieving efficient cyclical transport.

As the core of the transport system, the Y railcar is equipped with a transport control system to ensure the precise delivery of soybean pots, while also allowing for plant rotation. To ensure the stability of the soybean pots during transport and imaging, the railcar must have powerful driving capabilities and automatic start/stop functions. Compared to designing an independent braking system, using a motor with an electromagnetic braking function provides a simpler and more efficient solution. Additionally, to improve the system’s reliability and stability, the wheels are made of custom U-shaped steel that perfectly matches the rectangular tube track size, effectively preventing derailment of the railcar (Fig. 5).

Fig. 5.

Structure diagram of the intelligent railcar

Core components of the transport system

When the railcar moves, it is mainly subjected to rolling resistance, air resistance, acceleration resistance, and ramp resistance. Since it operates on a rectangular tube track, and the track construction ensures a level surface with relatively slow movement speed, air resistance, acceleration resistance, and ramp resistance can be neglected. This study only considers rolling resistance. The force ​ acting on the railcar platform is:

1

2

In Eq. (2): is the rolling friction coefficient on the track, which is 0.1. The torque of the driving motor is:

3

In Eq. (3): is the wheel radius, with a value of 0.1 m; is the overall efficiency of the power transmission system, with a value of 0.8; and​ is the reduction ratio of the drive motor, a constant of 10. The power of the drive motor is:

4

In Eq. (4): is the maximum speed of the drive motor, which is 1500 r/min. is the safety factor, taken as 2.0 due to the complexity of the field environment. The maximum torque and power of the drive motor are calculated to be , , respectively. The X motor, which needs to move a larger number of plant pots, is selected based on the same calculations, yielding, . Therefore, the X motor is selected as a 48 V, 1000 W DC motor, and the Y motor is selected as a 48 V, 500 W DC motor.

In this study, we innovatively installed an imaging turntable (used to carry and rotate the potted soybeans) on the track vehicle and fixed a movable X track on the turntable, enabling it to serve both transportation and rotating imaging functions. This design achieves seamless integration between the transportation and imaging stages, avoiding the complicated transfer of plants from the transport device to the imaging device, reducing system complexity and cost, and laying the foundation for automated imaging.

During plant imaging, the stability of the rotating turntable directly affects the quality of the image collection. Therefore, a reasonable design of the turntable structure is required, as well as stress and strain calibration under operating conditions. Considering both the material processability and stability, the turntable body is initially selected to be made of 304 stainless steel. In the NX10.0 software, material properties are set as follows: material density of 7.93 g/cm³, Young’s modulus of 1.9 × 10⁵ MPa, and Poisson’s ratio of 0.3. The maximum element size in the finite element analysis is controlled at 6.2 mm, with the assumption of linear stress-strain behavior. The simulation analysis shows that when the turntable carries the potted soybeans, the short edges of the turntable contact the track, and it bears a downward normal force. The total mass of the potted soybeans and their support is converted into pressure, and this is used as the load condition in the simulation.

As shown in Fig. 6, under full load, the maximum deformation of the turntable occurs at the middle of the short edge, with a deformation of 0.498 mm. The maximum stress occurs at the universal steel wheel supporting the turntable, at 42.064 MPa, which is well below the yield limit of 304 stainless steel. Based on the finite element analysis results and the actual size and material properties of the turntable, the stress and deformation of the turntable under full load meet the design requirements, ensuring its stable operation. Therefore, this turntable structure is fully suitable for stable rotational imaging of potted soybeans.

Fig. 6.

Stress-strain simulation of the turntable under full load

Transportation control

Transportation control refers to the automated, precise transport of soybean pots between the imaging room and the field, integrating field cultivation with indoor imaging. The transportation control system includes components such as the PLC controller, relays, stepper motors, stepper motor controllers, drive motor controllers, remote control signal transmitters and receivers, servo drivers, and proximity sensors. Given the complex and variable field environment in which the platform operates, the main controller selected is the Huaqingjun 16-input 8-output PLC integrated machine, which is resistant to external interference, has a low failure rate, and is easy to expand for external modules. It supports 4G remote connections and RS485 communication, facilitating the expansion of control modules.

The process of linking the control system components is as follows: After checking the system’s normal operation, a remote control with a fixed frequency of 2.4 GHz transmits instructions to the remote receiver. According to the RS485 protocol, the instructions are sent to the main controller (PLC). After the main controller (PLC) parses the instructions, it sends commands to the controllers of the two track vehicles, enabling automatic start-stop and direction selection. The control flow diagram for the transportation system is shown in Fig. 7. Additionally, the PLC outputs different pulse signals to control the direction, speed, and angle of the stepper motors, with the relationship given by:

5

Fig. 7.

Flowchart of the transportation system control

In Eq. (5): ​is the number of pulses; is the rotation angle of the controlled object (the turntable); is the subdivision number set by the stepper motor driver; 1 is the worm gear transmission ratio of the turntable. When =1:180, =20, and =3600, =360°. By changing the output pulse count of the PLC, the controlled object can be rotated to any desired angle.

The PLC programming implements the following functions: controlling the track vehicles to run automatically along a set route; controlling the imaging turntable to automatically rotate and carry soybean plants, matching the photography system for shooting; controlling the track vehicles to trigger the matching automatic door to open and close, allowing entry into the imaging room; controlling the track vehicles to sense obstacles and perform an emergency stop; and achieving precise transfer of the pots between the X and Y tracks.

Information interaction and output

The information interaction between the platform’s transportation system and imaging system is achieved through the QR code scanner in the imaging system. The scanner scans the QR code tags on the soybean pots, converting the movement information of the soybean plants into an electrical signal that activates the imaging sensor. Simultaneously, after recognizing the soybean plant’s variety and field management information, the system automatically creates corresponding categorized folders to store the collected phenotypic data.

The image data collected by the imaging sensor is stored in the desktop computer’s main unit and offers three data output methods:

1. The image data can be input into the data analysis system on the host computer for analysis, after which detailed phenotypic information can be exported.

2. The host computer is connected to the internet, allowing phenotypic data to be transferred to the cloud for downloading elsewhere.

3. The host computer has a USB interface, enabling data to be extracted using a USB drive or external hard drive.

Experimental site and soybean materials

This experiment was conducted at the Modern Agriculture Research and Development Base of Sichuan Agricultural University, located in Longxing Town, Chongzhou City, Chengdu, Sichuan Province. A vertical planting system was established using maize and soybean. The maize cultivar ‘Zhongyu No. 3’, a semi-compact type, was selected, along with nine soybean cultivars (E9-2, ND25, ND12, CXD3, DZD, BYH, GX12, SJZ, HHD) exhibiting distinct differences in plant architecture.

Soybean sampling time and method

Throughout the entire soybean growth period, sampling was conducted at 7-day intervals from the V1 stage (when the first trifoliate leaf is fully expanded) to the R8 stage (maturity). At each sampling time point, images were captured for three potted plants of each soybean cultivar. After imaging, plant height and width were measured using a tape measure. The entire plant was then cut at the base, and canopy biomass was weighed using an electronic balance.

To determine leaf area, all leaves from each plant were manually removed and spread flat on a black, light-absorbing background cloth, ensuring no overlap and naturally extended edges. A smartphone camera was used to take overhead images, which were then processed in Image-Pro Plus software to extract leaf area.

To obtain image-based metrics, image sequences for 3D reconstruction were extracted from the original images, and 3D models of the plants were created. After point cloud preprocessing, a 3D canopy point cloud model of each soybean plant was built. Subsequently, a self-developed parameter extraction program running on the MATLAB platform was used to calculate the following structural phenotypic traits: plant length, plant width, plant height, centroid height, projection area, concave hull volume, convex hull volume, convex hull surface area, table volume, volume of voxels, alpha shape volume, and alpha surface area.

For detailed algorithms and calculation methods, please refer to our previous work [20], The soybean 3D reconstruction workflow is provided in Supplementary Material 2.

Data analysis

A polynomial model was used to estimate soybean canopy biomass and leaf area. The calculation formula is as follows:

6

In Eq. (6): Y is the predicted variable, X is the image-based phenotypic parameter, n is the highest order of the polynomial, and , , , … are the fitting coefficients.

Results

Overall platform performance

As shown in Fig. 8, the platform achieves the following functions: the transportation system is largely automated, with soybean pots automatically transported between the field and the imaging room; the imaging system creates a stable imaging environment and automatically collects phenotypic information. The imaging room automatically forms an environment suitable for image capture, with high brightness, a pure background, and stability. Combining industrial control and computer programming, an automatic imaging control program has been developed, enabling the automatic transportation of soybean plants and the automatic collection of raw image sequences, achieving an average capture time of less than 2 s per image.

Fig. 8.

High-throughput phenotyping platform in the field

Application of the high-throughput phenotyping platform

Platform accuracy verification

12 soybean canopy image metrics were extracted using self-programming. To verify the accuracy of the platform, this study performed linear fitting analysis on the plant height and plant width data extracted from the soybean canopy model over the entire growing period, comparing it with manual measurements, and used the coefficient of determination (R²) to evaluate the precision. The results (Fig. 9) show that the R² for manually measured canopy plant height and model-extracted plant height was 0.990, and the R² for manually measured canopy plant width and model-extracted plant width was 0.952. These results indicate that the platform has high accuracy and provides a reliable foundation for structural research on soybean canopies.

Fig. 9.

Platform accuracy verification

Correlation analysis of model parameters and traditional agronomic parameters

To achieve rapid, non-destructive measurement of soybean canopy biomass and leaf area, as well as continuous monitoring of growth dynamics, this study performed a correlation analysis between the model-extracted parameters and traditional agronomic parameters such as fresh weight and leaf area manually measured from the soybean canopy. The goal was to reveal the relationship between the model parameters and traditional agronomic parameters, and select the parameters for developing predictive models.

The correlation analysis results showed that image parameters and traditional agronomic parameters had strong correlations throughout the entire growth period. Therefore, three key periods were selected for detailed analysis: V4 (vegetative growth stage), R1 (flowering stage), and R6 (reproductive growth stage).

Using a correlation coefficient threshold of ≥ 0.9, three key traits were identified at the V4 stage: α-shape surface area, voxel volume, and top-projected area. Among them, the correlation coefficients of α-shape surface area and voxel volume with canopy fresh weight were 0.93 and 0.94, respectively, while their correlations with leaf area were 0.95 and 0.94. Additionally, top-projected area exhibited a correlation coefficient of 0.91 with leaf area (Fig. 10a). In the R1 period, the correlation coefficients for α-shape surface area, voxel volume, and α-shape volume with canopy fresh weight exceeded 0.90 (0.95, 0.95, and 0.90, respectively), and the correlation coefficients with leaf area were also above 0.90. However, the correlation coefficient between top projection area and leaf area dropped from 0.91 in V4 to 0.73 in R1, likely due to increased overlap in the vertical direction as more leaves emerged (Fig. 10b). In the R6 period, the correlation coefficients for α-shape surface area and voxel volume with canopy fresh weight were 0.95 and 0.96, respectively, and the correlation coefficients with leaf area were 0.96 and 0.94, respectively. In addition, the canopy volume had a correlation coefficient of 0.91 with leaf area (Fig. 10c).

Fig. 10.

Heatmap of the correlation between model-extracted parameters and manually measured parameters at different soybean growth stages

Prediction of soybean canopy biomass and leaf area

The leaf area and biomass of the soybean canopy are closely related to its photosynthetic capacity and dry matter accumulation. Among the tested traits, voxel volume exhibited the highest correlation with canopy fresh weight, with a correlation coefficient consistently ranging from 0.94 to 0.95 throughout the entire growth period. Similarly, α-shape surface area showed the strongest correlation with leaf area, with a correlation coefficient of 0.94 to 0.96. Based on these relationships, polynomial models were developed to predict biomass and leaf area, yielding good fitting performance, with R² values of 0.858 and 0.922 and RMSE values of 0.06 kg and 0.03 m², respectively. These results demonstrate that the model enables rapid and non-destructive extraction of soybean canopy parameters (Fig. 11a, d).

Fig. 11.

Fitting analysis of soybean total volume and fresh weight, surface area and leaf area

In the maize-soybean strip intercropping system, soybean undergoes two distinct growth phases: a co-growth phase and an independent growth phase. During the co-growth phase (V1-R1), soybean is primarily in the vegetative growth stage, whereas in the independent growth phase (R2-R6), it transitions to reproductive growth. The canopy characteristics of soybean exhibit significant differences between these two phases. To improve prediction accuracy, separate biomass prediction models were developed for each phase. As shown in Fig. 11b-c, the biomass prediction model performed better during the co-growth phase (R² = 0.965, RMSE = 0.006 kg) compared to the independent growth phase (R² = 0.716, RMSE = 0.087 kg). Similarly, for leaf area prediction (Fig. 4e-f), the model achieved a higher accuracy during the co-growth phase (R² = 0.972, RMSE = 0.005 m²) than in the independent growth phase (R² = 0.878, RMSE = 0.062 m²). These results suggest that a stage-specific modeling strategy should be adopted to ensure the accuracy and reliability of soybean canopy trait predictions across different growth phases.

Discussion

Research on the plant architecture traits of individual soybean plants is fundamental for breeding high-yield plant types [21]. Conducting precise phenotyping of individual soybean plants under vertical planting systems is particularly important for selecting shade-tolerant, high-yield soybean germplasm and enabling refined field management. The development of crop phenotyping platforms has greatly improved the efficiency and accuracy of phenotypic research [22]. At present, various types of crop phenotyping platforms have been developed for different field environments. Their basic design typically equips phenotyping sensors with mobile devices to enable close-range imaging of crops [23]. However, due to the compact structure of vertical planting systems, existing platforms cannot effectively avoid shading from taller crops, which limits the precise acquisition of phenotypic data from individual soybeans.

Therefore, this study aimed to address this challenge by designing a rail-based field platform for precise phenotyping of individual soybean plants. This platform comprehensively considers factors such as planting mode, soybean architectural traits, precision, efficiency, and cost. It consists of a fixed imaging chamber built in the field, a rail system connecting the chamber with potted soybeans arranged according to the strip intercropping pattern, and a self-developed intelligent rail car. This configuration enables automatic transportation and precise imaging of soybean plants, effectively minimizing the interference of natural environmental factors on data collection.

In terms of structural design, the imaging chamber is a rectangular space measuring 6 m in length, 3 m in width, and 2.7 m in height. This provides sufficient space for phenotypic data collection from the vegetative to reproductive stages of soybean growth, while matching the dimensions of standard shipping containers to facilitate standardized modification and cost control. The rail system draws on railway sleeper design to enhance stability and adopts an innovative spatially staggered layout at rail junctions, overcoming the limitations of traditional junctions that are complex and space-intensive (see Fig. 4). Additionally, the platform uses an electric rail car to transport the potted plants and integrates a rotation device on the rail car to prevent positional shifts during imaging. This reduces construction costs and shortens the time required for phenotypic data collection, resulting in a simple, stable, and cost-effective field phenotyping platform for soybeans.

Regarding functional optimization, this study complements the method developed by Ma et al. [24], who used a Kinect sensor to construct a 3D point cloud phenotyping approach for soybeans. By combining color and depth images with a bounding box algorithm, Ma et al. successfully extracted plant height and leaf area index (LAI) with R² values exceeding 0.94, demonstrating that low-cost equipment combined with point cloud analysis can also achieve high phenotyping precision in outdoor environments. In comparison, the platform developed in this study automatically transports plants via rails to a fixed imaging chamber, effectively mitigating the impact of strong light, shadows, and wind, thereby improving data consistency while ensuring operational efficiency and system stability. This provides a new approach for precise phenotyping of individual soybeans in complex planting structures.

For operational control, the platform uses position sensors and a programmable logic controller (PLC) to achieve automatic start-stop functions and overall system control [25], ensuring stable operation under preset modes. An automated imaging control program was developed that uses QR code scanning to trigger the start of image acquisition, ensuring consistent image capture under defined parameters. Currently, each image can be captured in less than 2 s, enabling the system to acquire over 12,000 soybean images per day, demonstrating high imaging efficiency.

In terms of data analysis, the study found that the prediction accuracy for canopy fresh weight and leaf area was significantly higher during the vegetative growth stage than during the reproductive stage. This is likely due to nutrient translocation to the pods and leaf occlusion affecting pod visibility in the later stages, which reduces prediction accuracy. Compared with existing models based on hyperspectral or RGB cameras for early growth stage phenotyping (R² ranging from 0.442 to 0.848) [26], the linear model developed here based on surface area demonstrated superior performance, with a prediction R² of 0.972 and a root mean square error (RMSE) of 0.005 m² during the vegetative stage. This indicates that the method is feasible for rapid and precise estimation of biomass and leaf area at the individual plant level.

Despite successfully integrating field-based intercropping with stable indoor imaging, the platform still has room for improvement in data collection efficiency. Operational efficiency is closely tied to the level of automation. A higher degree of automation can not only accelerate the movement of plants and sensors but also automatically integrate environmental information with phenotypic data, thereby enhancing data interpretability. Various mature automation solutions are available for reference, such as UAV-based monitoring, conveyor-based platforms for automated transport of plants to the imaging chamber, and gantry platforms for scheduled monitoring. These technologies could inform further optimization of this platform.

Field trials showed that time costs are mainly concentrated in the transportation of potted soybeans. The current system uses two rail cars operating in perpendicular directions (see Fig. 8), but the transport speed is limited during the seedling stage due to the plants’ fragility, affecting overall efficiency. To enhance performance, future upgrades could include optimizing automated transport strategies, deploying multiple rail cars, and establishing rapid transfer of plants to rails near the imaging chamber, thereby improving data collection efficiency and enhancing automation.

At present, the imaging system mainly uses RGB cameras, which meet basic morphological data collection needs but are limited in representing comprehensive crop traits in complex environments. Li et al. [27] proposed a soybean wilting index based on fractal dimension calculations of multispectral images, using RGB, NIR, and other channels to build three index systems (LBC, LDBC, LDB). Their results, with R² values exceeding 0.85, demonstrated the advantages of multispectral imaging for non-destructive stress response monitoring. This provides a theoretical basis and methodological reference for integrating multi-source sensors (e.g., thermal infrared, multispectral, LiDAR) into the platform’s imaging system in the future.

Given that crop stress screening and genetic studies often require multi-dimensional phenotypic information, integrating data from multiple sensors and modalities will significantly improve the accuracy and systematization of trait analysis [28], and will help deepen the understanding of genotype–environment interactions. With sufficient space available in the imaging chamber, the platform has the potential to expand its imaging capabilities by incorporating LiDAR, fluorescence imaging, thermal infrared, and other sensors. Combined with data fusion and intelligent control technologies, this would enable the construction of a high-throughput, multi-dimensional, high-precision smart phenotyping system, extending its applications in complex trait analysis and precision breeding.

Finally, the platform adopts a modular and open-source design concept, ensuring excellent flexibility and scalability. The control system is developed on a universal PLC platform with standardized communication interfaces, allowing for the flexible integration of various sensors to capture multi-dimensional phenotypic traits. The high standardization of the rail system and imaging chamber facilitates rapid deployment and repeated use across different test sites. The transport and image acquisition programs support secondary development, which can significantly lower technical barriers and modification costs for cross-region and cross-crop applications. However, the platform still requires further verification and optimization in terms of long-term operational stability and adaptability to diverse environments.

Conclusion

This study developed a phenotyping platform designed for individual soybean plants under vertical planting conditions. Using a rail-based transportation system, potted soybeans are automatically conveyed to a fixed imaging chamber, where industrial control and machine vision technologies enable automated image capture and classification. This system effectively integrates the natural field growth environment with standardized indoor imaging conditions, providing robust technical support for high-precision acquisition of individual soybean phenotypic data and facilitating systematic evaluation and precise identification of germplasm resources.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Author contributions

Xiuni Li: Writing - review &editing, Writing-original draft, Visualization, Validation, Investigation, Formal analysis. Menggen Chen, Shuyuan He, Mei Xu, Yao Zhao: Conceptualization. Weiguo Liu: Resources, Project administration, Conceptualization. Funding acquisition.

Funding

This work was supported by the Biological Breeding-National Science and Technology Major Project (2023ZD0403405), the National Natural Science Foundation of China (32172122), the Key Research and Development Project of the Guizhou Branch of China National Tobacco Corporation(2023XM18), the National Modern Agricultural Industry Technology System, Sichuan Soybea Innovation Team (SC-CXTD-2024-21).

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

All authors agreed to publish this manuscript. Applicable for both human and/ or animal studies. Ethical committees, Internal Review Boards and guidelines followed must be named. When applicable, additional headings with statements on consent to participate and consent to publish are also required.

Consent for publication

Consent and approval for publication was obtained from all authors.

Competing interests

The authors declare no competing interests.