Soybean pod classification detection method, device, product, medium and equipment

Abstract

The invention discloses a soybean pod classification detection method and device, a product, a medium and equipment, and relates to the technical field of artificial intelligence machine vision seed examination, and the method comprises the steps: obtaining image data of to-be-classified soybean pods; inputting the image data into the segmentation model to obtain an image in which a single-plant pod area is segmented; and inputting the image in which the single-plant pod region is segmented into a classification model, and identifying phenotypic parameters of each single-plant pod. According to the soybean pod classification detection method and device, the product, the medium and the equipment, the efficiency and the accuracy of pod classification counting of different plants in soybean seeds can be effectively improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence machine vision seed testing technology, and in particular to a soybean pod classification detection method, device, product, medium and equipment. Background Art

[0002] Soybeans are rich in high-quality protein and a large amount of oil. They not only provide people with nutritious food resources, but also provide high-quality feed for animal husbandry. Soybean seed testing is one of the key tasks in soybean breeding. According to the "Guidelines for Specificity, Consistency and Stability Testing of New Plant Varieties - Soybeans", the soybean pod traits that need to be examined mainly include the number of pods per plant, the number of grains per plant, and the number of grains per pod. These trait parameters are important indicators reflecting soybean yield. Rapid and accurate acquisition of these trait parameters is of great significance for breeding high-yield soybean varieties. At present, soybean seed testing in many regions is still using manual phenotypic detection methods. Manual detection methods are not only costly, time-consuming and labor-intensive, and have low detection efficiency, but also prone to deviations. They are no longer suitable for the needs of modern soybean breeding. Summary of the invention

[0003] The purpose of this application is to provide a soybean pod classification and detection method, device, product, medium and equipment, which can effectively improve the efficiency and accuracy of pod classification and counting of different plants in soybean testing.

[0004] To achieve the above objectives, this application provides the following solutions:

[0005] In a first aspect, the present application provides a soybean pod classification detection method, comprising:

[0006] Acquire image data of soybean pods to be classified;

[0007] The image data is input into the segmentation model to obtain an image with a single pod area segmented out;

[0008] The images of the segmented individual pod regions are input into the classification model to identify the phenotypic parameters of each individual pod.

[0009] Optionally, the soybean pod classification detection method further comprises: obtaining QR code information of the soybean pod; the QR code information includes the variety and coding information of the soybean pod;

[0010] The QR code information and phenotypic parameters are concatenated and stored in the local database.

[0011] Optionally, the segmentation model is a Mask R-CNN model.

[0012] Optionally, the Mask R-CNN segmentation model includes a backbone network, a region of interest generation network, a region of interest alignment network and a head output network connected in sequence; wherein the backbone network includes a first feature extraction network and a feature fusion network connected in sequence; the first feature extraction network is a ResNet50 model; the feature fusion network is FPN; the region of interest generation network includes a region proposal network and a region generation network connected in sequence.

[0013] Optionally, the classification model includes a second feature extraction network and a classification output head network connected in sequence; wherein the second feature extraction network has the same structure as the first feature extraction network; and the classification output head network includes a fully connected layer and a softmax layer connected in sequence.

[0014] In a second aspect, the present application provides a soybean pod classification and detection device, comprising: a high-definition camera, a Jetson nano development board and a touch screen; the Jetson nano development board is connected to the high-definition camera and the touch screen respectively; the high-definition camera is placed directly above the soybean pods to be classified, and is used to obtain image data of the soybean pods to be classified and transmit it to the Jetson nano development board; the Jetson nano development board performs classification detection according to the image data, obtains the phenotypic parameters of each individual pod and displays them through the touch screen.

[0015] Optionally, the soybean pod classification detection device also includes a QR code scanner; the QR code scanner is connected to the Jetson nano development board to obtain the QR code information of the soybean pod.

[0016] In a third aspect, the present application provides a computer program product, including a computer program, which implements the soybean pod classification and detection method when executed by a processor.

[0017] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which implements the soybean pod classification and detection method when executed by a processor.

[0018] In a fifth aspect, the present application provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the soybean pod classification and detection method.

[0019] According to the specific embodiments provided in this application, this application has the following technical effects:

[0020] The present application provides a soybean pod classification detection method, device, product, medium and equipment. By using deep learning models such as segmentation models and classification models, the phenotypic parameters of each individual pod of the soybean pods to be classified are automatically extracted based on the acquired image data of the soybean pods to be classified, thereby alleviating the burden of manual detection in the prior art and improving the efficiency of soybean pod phenotypic parameter extraction, thereby effectively improving the efficiency of classifying and counting pods of different plants in soybean testing, and can effectively reduce the errors caused by manual detection and improve the accuracy of detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0022] Figure 1 A schematic diagram of the process of soybean pod classification and detection method provided in this application;

[0023] Figure 2 Schematic diagram of image data of soybean pods to be classified provided for this application;

[0024] Figure 3 Schematic diagram of the electronic tag of soybean pods to be classified provided for this application;

[0025] Figure 4 A schematic diagram of an image of a single plant pod region segmented for this application;

[0026] Figure 5 A schematic diagram of the classification test result display interface provided for this application;

[0027] Figure 6 Schematic diagram of the soybean pod classification and detection device provided in this application;

[0028] Figure 7 This is a detection flow chart of the soybean pod classification detection device provided in this application.

[0029] Figure numerals: overall device frame—1, QR code scanner—2, touch screen—3, device box—4, retractable slide rail—5, movable shooting background board—6, universal wheel—7, high-definition camera—8, Jetson nano development board—9. DETAILED DESCRIPTION

[0030] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0031] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0032] like Figure 1 As shown, the soybean pod classification detection method disclosed in the present application comprises:

[0033] Step 101: Obtain image data of soybean pods to be classified.

[0034] When performing soybean pod classification detection, the present application first obtains image data of the soybean pods to be classified through a high-definition camera or other equipment, such as Figure 2 In addition, as a specific embodiment, since each soybean plant of each variety is hung with a unique electronic tag to distinguish different soybean plants of different varieties, such as Figure 3 As shown, the electronic tag includes a QR code containing the variety and number information of a single soybean pod and corresponding text information. Therefore, in addition to obtaining the image data of the soybean pods to be classified, the present application also obtains the QR code information such as the variety and number information of the soybean pods by identifying the electronic tag containing the QR code on the single soybean plant.

[0035] After acquiring the image data of the soybean pods to be classified, the soybean pod classification detection method of the present application extracts the phenotypic parameters of each pod in the collected image data of the soybean pods to be classified through a pod phenotype extraction module, wherein the pod phenotype extraction module is composed of a segmentation model and a classification model, and the description of the segmentation model and the classification model are specifically shown in steps 102 and 103.

[0036] Step 102: input the image data into the segmentation model to obtain an image of a segmented single pod region.

[0037] The segmentation model of the present application segments each pod in the image data of soybean pods to be classified through the Mask R-CNN model. The Mask R-CNN segmentation model consists of a backbone network, a region of interest generation network, a region of interest alignment network and a head output network connected in sequence.

[0038] Specifically, the backbone network in the Mask R-CNN segmentation model is used to extract and fuse features from the image data of soybean pods to be classified, including a first feature extraction network and a feature fusion network connected in sequence. The first feature extraction network is composed of a ResNet50 model, which is responsible for extracting features from the input image data of soybean pods to be classified and outputting the extracted feature information of different scales; the feature fusion network is an FPN network, which is used to fuse the semantic information of high-level features and the spatial resolution of low-level features output by the first feature extraction network, so as to improve the backbone network's ability to capture features of different scales.

[0039] The region of interest generation network in the Mask R-CNN segmentation model consists of a region proposal network and a region generation network connected in sequence. First, the region proposal network generates a series of regional anchor boxes that may contain targets, each of which has a score and four coordinate information. Secondly, the region proposal network performs binary classification on each anchor box to determine whether the anchor box contains the target. In this way, the background area is filtered out and only the area that may contain the target is retained. In addition, the region proposal network also performs bounding box regression prediction on each anchor box, and then uses binary classification cross entropy loss and regression loss to supervise the output of the region proposal network. Next, the region generation network uses the bounding box regression value predicted by the region proposal network to further adjust the position and size of each anchor box to make it closer to the actual target area. Then the region generation network will filter out the anchor boxes that exceed the image boundary after adjustment or the anchor boxes with too small area to ensure that the candidate area is within the image and has a reasonable size. For the remaining candidate areas, the region generation network will sort them according to the anchor box scores. The candidate areas with higher scores are more likely to contain targets. Then the non-maximum suppression algorithm and the intersection-over-union ratio indicator are used to remove overlapping areas. Finally, the region generation network will select the top N candidate areas with the highest scores from the processed areas as the final region proposal anchor box output.

[0040] Binary Cross Entropy Loss for Region Proposal Network The formula is as follows:

[0041]

[0042] Among them, N pos is the number of valid anchor boxes, y i is the true label of the i-th valid anchor box (y i = 1 means that the i-th valid anchor frame contains the target, y i = 0 means that the i-th valid anchor frame does not contain the target), p i is the probability that the i-th sample (valid anchor box) is predicted as a positive class by the region proposal network.

[0043] Bounding Box Regression Loss for Region Proposal Network The formula is as follows:

[0044]

[0045] Among them, t j is the anchor box predicted by the jth region proposal network, The ground-truth bounding box of the jth region proposal network, smooth L1 It is the L1 smoothing loss function, which measures the difference between the predicted box and the real box. The specific formula is:

[0046]

[0047] For formula (3), when the predicted value t j and the true value When the error is small SmoothL1 loss uses a quadratic function to make the error's effect on model parameter update smoother; when the predicted value t j and the true value When the error is large Smooth L1 loss uses a linear function to alleviate the gradient explosion problem, so Smooth L1 loss is more robust to outliers during model training.

[0048] The intersection over union (IoU) is a common indicator for measuring the degree of overlap between two bounding boxes. Its formula is as follows:

[0049]

[0050] Among them, Intersection Area is the overlapping area of ​​two bounding boxes; Union Area is the joint area of ​​two bounding boxes.

[0051] The region of interest alignment network in the Mask R-CNN segmentation model is responsible for accurately extracting the features in the final region proposal anchor box output by the region generation network from the feature map. The head output network is used to output the final prediction result of the segmentation model to obtain an image of the segmented single pod region, which includes the classification result, the bounding box position result, and the mask result used for segmentation. The multi-classification cross entropy loss is used to supervise the prediction of the head output network's classification of the anchor box, the bounding box regression loss is used to supervise the prediction of the head output network's prediction of the anchor box position information, and the mask loss is used to supervise the segmentation model's prediction of the mask.

[0052] Multi-class cross entropy loss for predicting the head The formula is as follows:

[0053]

[0054] Among them, N roi represents the number of valid regions of interest, C is the number of predicted categories, and y f,c represents the true label of the fth valid region of interest of the cth predicted category, p f,c Represents the predicted probability value of the fth valid region of interest of the cth predicted category.

[0055] Predicting the bounding box regression loss of the head The formula is as follows:

[0056]

[0057] Among them, y f is the true label of the fth valid region of interest (y f = 1 means that the fth valid region of interest contains the target, y f = 0 means that the fth valid region of interest does not contain the target), b f is the predicted bounding box of the fth valid region of interest, is the ground-truth bounding box of the f-th valid region of interest.

[0058] Predicting the mask loss of the head The formula is as follows:

[0059]

[0060] Among them, m f is the prediction mask of the fth valid region of interest, is the true mask of the fth valid region of interest, is a binary cross loss function, which measures the difference between the predicted mask and the true mask. The specific formula is:

[0061]

[0062] Where H and W are the height and width of the mask, respectively. h,w is a prediction mask with height h and width w, is the true mask with height h and width w.

[0063] As a specific embodiment, the dataset used in this application to train the Mask R-CNN model is simulated data synthesized from a real dataset. The simulated dataset is divided into 5:2.5:2.5, and the model weights pre-trained by others are used for migration training. A total of 100 batches are trained. The model is tested on the real dataset and the simulated dataset. Table 1 shows the datasets used for training and testing the Mask R-CNN model, and Table 2 shows the segmentation results of the Mask R-CNN model. Figure 4 The visualization of the Mask R-CNN segmentation results is shown. As shown in Table 2, the indicators used in this application to measure the performance of Mask R-CNN segmentation are P, R, F1-score, and AP@0.5.

[0064] Table 1 Datasets used for training and testing Mask R-CNN

[0065] Dataset Number of pictures Training set 1400 Validation set 700 Simulation test set 700 Real test set 80

[0066] Table 2 Mask R-CNN model detection results

[0067]

[0068] From the detection results of the Mask R-CNN model in Table 2, it can be seen that the average accuracy of segmenting pods using the Mask R-CNN model can reach more than 98%, whether in the simulated test set or the real test set, which has a good segmentation effect.

[0069] Step 103: Input the image of the segmented individual pod region into the classification model to identify the phenotypic parameters of each individual pod.

[0070] The classification model of the present application is responsible for dividing the segmented single pods into phenotypic parameters of each single pod, such as zero pods, one pod, two pods, three pods and four pods. The classification model is composed of a second feature extraction network and a classification output head network connected in sequence, wherein the second feature extraction network has the same structure as the first feature extraction network, both are ResNet50 models, and the ResNet50 model is used to extract features. The classification output head network outputs the predicted probability for each category through a fully connected layer and a softmax layer connected in sequence. As a specific embodiment, the classification model of the present application iterated a total of 100 batches, and Table 3 shows the data sets used for training and testing the classification model. Table 4 shows the final classification results of the classification model.

[0071] Table 3 Datasets used for classification model training and testing

[0072] Category Name Training set Test Set Zero Seed Pod 300 200 A pod 300 200 Two-seed pod 300 200 Three pods 300 200 Four pods 300 200

[0073] Table 4 Final classification results of the classification model

[0074] Pod categories AP@0.5 Zero Seed Pod 99.43％ A pod 98.27％ Two-seed pod 99.57％ Three pods 99.69％ Four pods 98.30％

[0075] It can be seen from the classification result data in Table 4 that the average accuracy of each soybean pod classified and detected by the soybean pod classification and detection method of the present application can reach more than 98%, which shows that the classification model used in the present application can accurately classify each soybean pod with high accuracy.

[0076] Based on the soybean pod classification detection method, the present application also provides a soybean pod classification detection device, which is mainly a soybean pod classification detection device based on a QR code, including a data acquisition device, a pod phenotype extraction module and a pod phenotype extraction software. Among them, the data acquisition module is responsible for collecting soybean pod image data, such as Figure 6 As shown, the data acquisition module is composed of an overall equipment frame 1, a QR code scanner 2, a touch screen 3, an equipment box 4, a retractable slide rail 5, a movable background plate 6, a universal wheel 7, a high-definition camera 8, and a Jetson nano development board 9. The retractable slide rail 5 and the universal wheel 7 are connected to the overall equipment frame 1 through hexagon socket bolts, and the overall equipment frame 1 can move freely through the universal wheel 7. The movable background plate 6 is connected to the retractable slide rail 5 through a guide rail and a slide groove and can perform telescopic movement. The touch screen 3 and the QR code scanner 2 are assembled on the equipment box 4, and the Jetson nano development board 9 and the high-definition camera 8 are fixed in the equipment box 4 through hexagon socket bolts. The development board is a device developed for AI edge computing. The high-definition camera 8 and the QR code scanner 2 are connected to the Jetson nano development board 9 through a USB cable, and the touch screen 3 is connected to the Jetson nano development board 9 through an HDMI cable. The positions of the touch screen 3 and the two-dimensional code scanner 2 provided on the equipment box 4 can be set according to actual needs. The specific process of using the device for classification detection is as follows: Figure 7 As shown, a single soybean pod (soybean pod to be classified) is placed on a movable shooting background plate 6, and then the two-dimensional code (electronic two-dimensional code) on the electronic tag is scanned by a two-dimensional code scanner 2 and the soybean pod to be classified is photographed by a high-definition camera 8 to obtain the two-dimensional code information and image data of the soybean pod to be classified, and the soybean pod picture captured by the camera is processed by the deep learning model deployed on the development board, and the pod phenotypic parameters in the image are identified. Then, the measurement results of the pod phenotype are intuitively displayed on the touch screen 3, and the pod phenotypic information of the single soybean plant is printed. After scanning the two-dimensional code, the information in the two-dimensional code is spliced ​​together with various phenotypic parameter information predicted based on the image data and stored in the local database of the development board.

[0077] As a specific embodiment, the pod phenotype extraction module of this application is finally deployed on the JetsonNano development board, and the system used by the development board is based on Ubuntu 20.04. First, the segmentation model and classification model are trained in the server, and then the weight file generated by the trained model is converted into an ONNX format model, and then the TensorRT tool is used to optimize the ONNX format model to a TesnsorRT Engine, and finally the optimized TensorRT Engine is used to perform reasoning on the development board.

[0078] The pod phenotype extraction software consists of a control module, a visual graphical interface, and a data storage module. The control module is used to turn on the camera and start detection. Turning on the camera button controls the development board to turn on the camera, and turning on the detection button sends the picture captured by the camera to the inference program (i.e., the pod phenotype extraction module of this application) to detect and calculate the pod phenotypic parameters in the picture. After the detection is completed, the test results are output by the visual graphical interface. The output results include single soybean variety information, single soybean number information, number of each type of pods, total number of pods, and total number of grains. Figure 5 The data storage module is used to store the extracted result information in the SQLite database in the development board. The reason for choosing the SQLite database is that it occupies less resources and is very suitable for small development boards. The data stored in the development board are temporary data, and the data can be exported to an external database for storage.

[0079] In summary, the soybean pod classification detection method and device based on QR code single soybean pod classification detection for breeding provided in the present application solve the drawbacks of low efficiency and complicated operation of existing soybean pod testing methods. The soybean pod phenotype is automatically extracted through a deep learning model, which alleviates the burden on researchers and improves the efficiency of phenotype extraction. Moreover, the deep learning algorithm can not only replace researchers in testing work, but its recognition accuracy can even exceed that of researchers. Therefore, the soybean pod classification detection method and device of the present application can effectively improve the efficiency and accuracy of the classification and counting of pods of different plants in soybean testing.

[0080] In some embodiments, the present application also provides a computer program product, including a computer program, which implements the soybean pod classification and detection method when executed by a processor.

[0081] In some embodiments, the present application also provides a computer-readable storage medium having a computer program stored thereon, which implements the soybean pod classification and detection method when executed by a processor.

[0082] In some embodiments, the present application also provides a computer device, including a processor, a memory, an input / output interface (Input / Output, referred to as I / O), a communication interface, and a computer program stored in the memory and executable on the processor, and the processor executes the computer program to implement the soybean pod classification and detection method.

[0083] Wherein, the processor, memory and input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Wherein, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store pending transactions. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the soybean pod classification detection method is implemented.

[0084] It should be noted that the object information (including but not limited to object device information, object personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the object or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards.

[0085] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ReadOnlyMemory, ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0086] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0087] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A soybean pod classification detection method, characterized in that: include: Acquire image data of soybean pods to be classified; The image data is input into the segmentation model to obtain an image with a single pod area segmented out; The images of the segmented individual pod regions are input into the classification model to identify the phenotypic parameters of each individual pod.

2. The soybean pod classification detection method according to claim 1, characterized in that: Also includes: Obtaining QR code information of soybean pods; the QR code information includes the variety and coding information of the soybean pods; The QR code information and phenotypic parameters are concatenated and stored in the local database.

3. The soybean pod classification detection method according to claim 1, characterized in that: The segmentation model is a Mask R-CNN model.

4. The soybean pod classification detection method according to claim 3, characterized in that: The Mask R-CNN segmentation model includes a backbone network, a region of interest generation network, a region of interest alignment network and a head output network connected in sequence; wherein the backbone network includes a first feature extraction network and a feature fusion network connected in sequence; the first feature extraction network is a ResNet50 model; the feature fusion network is FPN; the region of interest generation network includes a region proposal network and a region generation network connected in sequence.

5. The soybean pod classification detection method according to claim 4, characterized in that: The classification model includes a second feature extraction network and a classification output head network connected in sequence; wherein the second feature extraction network has the same structure as the first feature extraction network; and the classification output head network includes a fully connected layer and a softmax layer connected in sequence.

6. A soybean pod classification detection device, characterized in that: include: A high-definition camera, a Jetson nano development board and a touch screen; the Jetson nano development board is connected to the high-definition camera and the touch screen respectively; the high-definition camera is placed directly above the soybean pods to be classified, and is used to obtain image data of the soybean pods to be classified and transmit it to the Jetson nano development board; the Jetson nano development board performs classification detection according to the image data, obtains the phenotypic parameters of each individual pod and displays them through the touch screen.

7. The soybean pod classification detection device according to claim 6, characterized in that: It also includes a two-dimensional code scanner; the two-dimensional code scanner is connected to the Jetson nano development board and is used to obtain the two-dimensional code information of the soybean pod.

8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the soybean pod classification and detection method according to any one of claims 1 to 7 is implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the soybean pod classification and detection method according to any one of claims 1 to 7 is implemented.

10. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the soybean pod classification and detection method according to any one of claims 1 to 7.

Images