A high-throughput intelligent detection method and system for germinating wheat seeds

Abstract

The present invention provides a high-throughput intelligent detection method and system for germinated wheat seeds, which belongs to the field of image detection technology, and includes a feeding component, a discharging component and an identification component; the feeding component includes a feeding table, and a feeding bin is arranged on the upper side of the feeding table; the discharging component includes a base, a loading table is arranged on the upper side of the base, a loading box is arranged on the upper side of the loading table, and a discharging baffle is arranged on one side of the loading box; the identification component includes a light source, a camera is arranged on the upper side of the light source, and the lens of the camera is located directly above the light source; the feeding bin is connected to a feeding controller, and the feeding table, the loading table and the discharging baffle are all connected to an IO controller. The present invention adopts an imaging system for automatically dispersing seeds to make the collection of seed images faster and more efficient, and the optimized YOLOv7‑CFocal model significantly improves the accuracy of germinating seed detection.

Description

Technical Field

[0001] The present invention relates to the technical field of image detection, and in particular to a high-throughput intelligent detection method and system for germinating wheat seeds. Background Art

[0002] Wheat is widely grown in many countries, and its yield and quality are crucial to ensuring global food supply. Ear sprouting is the phenomenon of seeds sprouting on the ear before harvest, which is closely related to seed dormancy and germination traits. Ear sprouting has a serious negative impact on wheat grain quality and yield, and it is of great significance to breed wheat varieties with high ear sprouting resistance. By comparing the proportion of germinated seeds to all seeds, varieties with excellent ear sprouting resistance can be screened from cultivated varieties. Since counting wheat germinated seeds is difficult and very time-consuming, there is an urgent need to develop a fast and accurate detection system for counting normal and germinated wheat seeds.

[0003] In recent years, with the rapid development of machine vision and machine learning technology, various target detection methods have been proposed, and the automatic detection of plant seeds has made great progress. However, in terms of detection speed, the lack of automatic image capture devices has seriously affected the detection efficiency, and the detection accuracy also needs to be further improved. Summary of the invention

[0004] The purpose of the present invention is to provide a high-throughput intelligent detection method and system for germinated wheat seeds. The imaging system of automatically dispersed seeds is adopted to make the collection of seed images faster and more efficient. The optimized YOLOv7-CFocal model based on global-local attention and focusing mechanism significantly improves the accuracy of germinated seed detection.

[0005] To achieve the above object, the present invention provides a high-throughput intelligent detection method for germinating wheat seeds, comprising the following steps:

[0006] Step S1, initialize the intelligent detection system, set the feeding time t1, the dispersion time t2 and the discharging time t3, and the cumulative time for a single detection t=t1+t2+t3;

[0007] Step S2, putting the seeds to be tested into the feed bin, the vibration motor on the feed table starts to vibrate, stops vibrating after t1, and the seeds fall into the loading box;

[0008] Step S3, the vibration motor on the loading platform starts to vibrate, and stops vibrating after t2, and the seeds are evenly distributed in the loading box;

[0009] Step S4, turning on the light source through the light source controller, and taking pictures of the seeds in the material loading box by the camera to obtain a seed image;

[0010] Step S5, the IO controller is used to control the discharging baffle to open, and the loading platform performs the discharging operation. After t3, the seeds in the loading box are removed;

[0011] Step S6, repeating steps S2 to S5 until there are no seeds in the feed bin;

[0012] Step S7, loading the acquired seed image, and detecting normal seeds and germinated seeds in the seed image through a germinated seed detection model;

[0013] Step S8, counting the number of normal seeds and germinated seeds, and calculating the germination rate;

[0014] Step S9: output the detection result.

[0015] Preferably, the germinating seed detection model is a YOLOv7-CFocal model, which adds a convolutional block attention module CBAM to the YOLOv7 model and replaces the ELAN module in YOLOv7 with a FocalNext module.

[0016] Preferably, the convolutional block attention module CBAM includes a channel attention module and a spatial attention module.

[0017] Preferably, the FocalNext module includes multiple focus blocks. Through the different operations combined by the focus blocks, FocalNext can perform deep feature extraction and transformation in the network, further enhancing the expressiveness and performance of the model.

[0018] The present invention also provides a system for implementing the above-mentioned high-throughput intelligent detection method for germinated wheat seeds, comprising a feeding component, a discharging component and an identification component;

[0019] The feeding assembly comprises a feeding platform, and a feeding bin is arranged on the upper side of the feeding platform;

[0020] The discharging assembly comprises a base, a loading platform is arranged on the upper side of the base, a loading box is arranged on the upper side of the loading platform, and one side of the loading box is a discharging baffle;

[0021] The identification component includes a light source, a camera is arranged on the upper side of the light source, and the lens of the camera is located directly above the light source;

[0022] The feed bin is connected to the feed controller, and the feed platform, loading platform and discharge baffle are all connected to the IO controller.

[0023] Preferably, the light source is a ring-shaped LED lamp, and the light source is connected to a light source controller.

[0024] Preferably, vibration motors are provided at the four corners of the feeding platform and the loading platform.

[0025] Preferably, the outlet of the feed bin is located directly above the loading box.

[0026] Therefore, the present invention adopts the above-mentioned high-throughput intelligent detection method and system for germinated wheat seeds, and the beneficial technical effects are as follows: the use of an imaging system that automatically disperses seeds makes the collection of seed images faster and more efficient, and the optimized YOLOv7-CFocal model based on the global-local attention and focusing mechanism significantly improves the accuracy of germinated seed detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 This is a structural schematic diagram of a high-throughput intelligent detection system for germinating wheat seeds according to the present invention;

[0028] Figure 2 Schematic diagram of the material loading box;

[0029] Figure 3 This is a flow chart of a high-throughput intelligent detection method for germinating wheat seeds of the present invention;

[0030] Figure 4 The appearance of normal and germinated wheat seeds; Figure 4 (a) in the figure is a normal seed; Figure 4 (b) in the figure is germinated seeds;

[0031] Figure 5 It is a germinating seed detection solution based on the YOLOv7-CFocal model;

[0032] Figure 6 is the structure of CBAM;

[0033] Figure 7 The structure of FocalNext;

[0034] Figure 8 is the Loss and mAP of the four candidate models; among them, Figure 8 (a) in the equation is Loss; Figure 8 (b) in the figure is mAP;

[0035] Fig. 9 is the training result of the YOLOv7-CFocal model; Fig. 9 (a) in the figure is Loss and mAP; Fig. 9 (b) in the figure is the precision-recall (PR) curve;

[0036] Fig.10 The results of wheat seed testing using different models; Fig.10 (a) is the original image; Fig.10(b) in the figure is Faster-RCNN; Fig.10 (c) in the figure is SSD; Fig.10 (d) in the figure is YOLOv5; Fig.10 (e) in the figure is YOLOv7; Fig.10 (f) in the figure is YOLOv7-CFocal;

[0037] Fig.11 are the training and testing results of YOLOv7-CBAM; among them, Fig.11 (a) is the PR curve; Fig.11 (b) in the figure is the mAP of two types of seeds;

[0038] Fig.12 are the training and testing results of YOLOv7-Focal; Fig.12 (a) is the PR curve; Fig.12 (b) in the figure is the mAP of the two types of seeds.

[0039] Reference numerals

[0040] 1. Feeding platform; 2. Feeding bin; 3. Base; 4. Loading platform; 5. Loading box; 6. Discharging baffle; 7. Light source; 8. Lens; 9. Camera. DETAILED DESCRIPTION

[0041] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0042] Unless otherwise defined, technical or scientific terms used in the present invention shall have the common meanings understood by one having ordinary skills in the field to which the present invention belongs.

[0043] Embodiment 1

[0044] For high-throughput detection of germinating seeds, the number of seeds of a variety is usually more than 300. The accumulation of seeds has a great impact on the accuracy, and manual placement is obviously boring and laborious, and inefficient. Therefore, a new device for uniformly dispersing piled seeds was constructed using the principle of electromagnetic vibration. The device can capture a complete image of more than 100 seeds at a time without manual placement.

[0045] like Figure 1-Figure 2 As shown, the present invention provides an intelligent detection system for germinating wheat seeds based on high-throughput imaging and deep learning, including a feeding component, a discharging component and an identification component;

[0046] The feeding assembly comprises a feeding platform 1 , and a feeding bin 2 is arranged on the upper side of the feeding platform 1 .

[0047] The outlet of the feed bin 2 is located directly above the material loading box 5 .

[0048] The discharging assembly comprises a base 3 , a loading platform 4 is arranged on the upper side of the base 3 , a loading box 5 is arranged on the upper side of the loading platform 4 , and a discharging baffle 6 is provided on one side of the loading box 5 .

[0049] The material carrying box 5 is a container for imaging seeds and is located below the lens 8. The material carrying box 5 evenly disperses the seeds through the vibration of the material carrying platform 4.

[0050] The size of the loading box 5 is 205mm*160mm*26mm (length*width*height), and the material is white melamine, with micro-stripes on the bottom to absorb light reflection.

[0051] Vibration motors are provided at the four corners of the feeding platform 1 and the loading platform 4 .

[0052] The identification component includes a light source 7, which is a ring-shaped LED light. The light source 7 is connected to a light source controller, and the light source controller adjusts the intensity of the light to make the imaging of the seeds clear and uniform.

[0053] A camera 9 is arranged on the upper side of the light source 7 , and a lens 8 of the camera 9 is located directly above the light source 7 .

[0054] Camera 9 is a Hikvision MV-CS200-10GC camera with a focal length of 16 mm and a maximum resolution of 5472*3648.

[0055] The feed bin 2 is connected to the feed controller, and the feed speed and feed amount are controlled by the feed controller. The feed table 1, the loading table 4 and the discharge baffle 6 are all connected to the IO controller to complete the automatic feeding, dispersing and discharging operations.

[0056] Embodiment 2

[0057] like Figure 3 As shown, the present invention provides an intelligent detection method for germinated wheat seeds based on high-throughput imaging and deep learning.

[0058] First, build the dataset.

[0059] The data samples were obtained by the designed seed imaging system. The sample image resolution is 5472*3648 and is saved in the image folder. A total of 300 images were obtained from 100 varieties. These images contain different seed skin colors, germination degrees, smoothness, and postures, which increases the diversity of sample data and improves the generalization ability of the detection model.

[0060] The accuracy of data annotation is the key to building a detection model. Use the "labelImg" labeling tool to label the dataset. The shortcut keys of this software can greatly simplify the annotation work. The dataset is divided into two categories, normal seeds are marked as "1" and germinated seeds are marked as "2". The appearance of normal and germinated wheat seeds is as follows Figure 4As shown. In order to accurately extract the characteristics of wheat seeds, the marking box should surround the entire seed as much as possible without including other seeds. For germinated seeds, the marking box should include the entire seed and part of the root. The labeling results should be strictly checked to ensure their correctness.

[0061] The dataset is divided into training set, validation set and test set in a ratio of 8:1:1. The total number of images is 300, including 240 for training, 30 for validation and 30 for testing.

[0062] Then, the germination seeds were tested.

[0063] Step S1, initializing the intelligent detection system, setting the feeding time t1, the dispersion time t2 and the discharging time t3;

[0064] Step S2, putting the seeds to be tested into the feed bin, the vibration motor on the feed table starts to vibrate, stops vibrating after t1, and the seeds fall into the loading box;

[0065] Step S3, the vibration motor on the loading platform starts to vibrate, and stops vibrating after t2, and the seeds are evenly distributed in the loading box;

[0066] Step S4, turning on the light source through the light source controller, and taking pictures of the seeds in the material loading box by the camera to obtain a seed image;

[0067] Step S5, the IO controller is used to control the discharging baffle to open, and the loading platform performs the discharging operation. After t3, the seeds in the loading box are removed;

[0068] Step S6, repeating steps S2 to S5 until there are no seeds in the feed bin;

[0069] Step S7, loading the acquired seed image, and detecting normal seeds and germinated seeds in the seed image through a germinated seed detection model;

[0070] Step S8, counting the number of normal seeds and germinated seeds, and calculating the germination rate;

[0071] Step S9: output the detection result.

[0072] The germinating seed detection model is the YOLOv7-CFocal model, such as Figure 5As shown in the figure, the model uses YOLOv7 as the basic structure, adds a convolutional block attention module CBAM (Convolutional Block Attention Module) to the backbone network, and replaces all ELAN modules (Efficient Layer Aggregation Networks) of YOLOv7 with FocalNext modules. YOLOv7-CFocal has great advantages in global and local feature extraction and complex object expression.

[0073] By adding CBAM to the backbone network, channel attention and spatial attention are enhanced to learn the dependencies between different channels. In the face of the complexity and multi-scale changes in the seed detection task, the key features of seed shape and texture can be captured. In addition, the occlusion problem can be better handled to reduce false or missed detections.

[0074] By using the FocalNext module instead of the ELAN module, the fine-grained local features and coarse-grained global features in the image can be captured, achieving stronger feature expression and semantic understanding, effectively solving the problem of feature expression of small particles and seeds of various shapes, and improving the accuracy and robustness of germinating seed detection.

[0075] like Figure 6 As shown in the figure, the convolutional block attention module CBAM includes a channel attention module (CAM) and a spatial attention module (SAM). The input features are first subjected to maximum pooling and average pooling to obtain two 1×1×C feature maps, which are then sent to the MLP. The features output by the MLP are summed at the pixel level and then activated by Sigmoid to obtain the channel attention feature (CA). CA is then used as input and sent to the SAM, where maximum pooling and average pooling are performed along the channel dimension to obtain two H×W×1 feature maps. The feature map channels are then cascaded, and then reduced to a single channel through a 7×7 convolutional layer. Finally, the final spatial attention feature (SA) is obtained through Sigmoid activation.

[0076] CAM focuses on channels, which extracts the most critical features for the task by assigning different weights to each channel. SAM focuses on the spatial dimension, which helps the model extract key information from different locations in complex spatial structures. CBAM combines these two sub-modules to focus on key features and ignore irrelevant information when processing complex data such as images and videos.

[0077] In YOLOv7, the ELAN module has weak feature extraction capabilities and a limited receptive field. To address this problem, a focal block is introduced to expand the receptive field of the network. The focal block consists of an extended depthwise convolution and two skip connections to strengthen the receptive field of neurons to capture local features and global interactions. In practice, multiple focal blocks are combined into a FocalNext module to enhance the expressiveness and performance of the model. The structure of FocalNext is shown in Figure 2. Figure 7 As shown in the figure, the left figure is the focal block structure, including an expanded 7×7 depth convolution and two 1×1 jump connections, where n is the number of channels of the output feature, r is the expansion rate, and GELU is an activation unit. The right figure is the FocalNext module. First, the data of spatial size H×W is input into a Stem and N consecutive Blocks to extract high-resolution features of spatial size H / 4×W / 4. These features are then sent to the M cascade stage to extract multi-scale features. The output of each stage has P3, P4, and P5, with strides of 8, 16, and 32 respectively, but only P3 is sent to the later stage. Finally, the output fusion features P3, P4, and P5 are used for dense prediction tasks.

[0078] Second, the model performance is evaluated.

[0079] The precision, accuracy, recall, accuracy and mAP (mean average precision) are used as shown in formulas (1)-(4).

[0080]

[0081] Wherein, TP is true positive, FP is false positive, TN is true negative, and FN is false negative. N is the number of categories. In this embodiment, N is 2, i.e., normal seeds and germinated seeds. i represents the i-th category.

[0082] In this example, four commonly used target detection models, Faster-RCNN, SSD, YOLOv5, and YOLOv7, are used as candidate models for germinating seed detection. Each model is trained using a self-built wheat seed dataset. The loss and mAP of each model are shown in Figure 2. Figure 8 shown.

[0083] In the Loss graph, the convergence speed of YOLOv5 and YOLOv7 is significantly faster than that of other graphs, and the final Loss value of YOLOv7 is the smallest. The mAP graph shows that YOLOv5 and YOLOv7 increase rapidly before 50 iterations and then gradually tend to a high level; on the other hand, Faster-RCNN and SSD increase slowly before 150 iterations and then gradually converge to a lower level.

[0084] The recall, precision, and mAP of the candidate models are shown in Table 1. YOLOv7 has a recall value of 0.911, a precision value of 0.878, and a mAP value of 0.926, which performs best among the four models.

[0085] Detection speed is another important aspect to judge the detection model. The average detection time for each image is shown in Table 1. The average detection time of YOLOv7 is at least 1.082s, which is the fastest among the four models.

[0086] Table 1 shows the performance of candidate detection models

[0087] Model Iterations Recall Accuracy McKenna Average detection time(s) Faster-RCNN 300 0.839 0.726 0.691 2.515 SSD 300 0.842 0.764 0.802 1.943 YOLOv5 300 0.887 0.854 0.893 1.348 YOLOv7 300 0.911 0.878 0.926 1.082

[0088] From the results of the four candidate models, YOLOv7 has the best performance on the training set and the test set. Therefore, YOLOv7 is used as the basic seed detection model for subsequent optimization to further improve the detection performance.

[0089] Results for the YOLOv7-CFocal model.

[0090] The YOLOv7-CFocal model adds the CBAM module to YOLOv7 and replaces the original ELAN module with the FocalNext module. The YOLOv7-CFocal model is trained with the same training hyperparameters as YOLOv7. The training process is as follows: Fig. 9 shown.

[0091] from Fig. 9 It can be seen that YOLOv7-CFocal has certain rules in the training process. In the first 100 iterations, the mAP fluctuates significantly, but shows an upward trend. With the increase of iterations, especially after 150 iterations, the fluctuations gradually disappear, and the mAP converges and stabilizes at around 94%. At the same time, there are 4 small fluctuations before 100 iterations. Then the fluctuations gradually disappear, and the Loss steadily decreases and gradually converges. Finally, the mAP of YOLOv7-CFocal reaches 94%, and the Loss converges to 0.0889. The results show that the model has good convergence and stability during training.

[0092] The precision-recall relationship is as follows Fig. 9 As shown in (b), the average accuracy of the two seed types is as high as 94%. Among them, the accuracy of normal seeds is 94.3%, and the accuracy of germinated seeds is 93.8%. This result proves the excellent performance of YOLOv7-CFocal in the seed detection task.

[0093] Seed detection results using different models are as follows Fig.10 shown.

[0094] Fig.10 In the figure, the blue border indicates normal seeds, and the red border indicates germinated seeds. Fig.10 As can be seen in the figure, for scattered seeds (first row), most models except Faster-RCNN get correct results. For dense seeds with few germinations (second row), Faster-RCNN and SSD sometimes miss some seeds, YOLOv5 and YOLOv7 sometimes regard germinated seeds as normal seeds, and YOLOv7-CFocal can correctly detect the category. For seeds with more germinations (third row), Faster-RCNN and SSD often get wrong results because the roots of germinated seeds are longer, while the YOLO series is more accurate. In the case of more germinated seeds and sticking together (fourth row), Faster-RCNN is prone to wrong results, SSD is prone to judge multiple seeds as one seed, YOLOv5 can accurately detect seeds but easily misjudge the category, YOLOv7 tends to judge germinated seeds as ungerminated, and YOLOv7-CFocal effectively overcomes the influence of adhesion and can accurately obtain the number and category of seeds.

[0095] Ablation experiments;

[0096] In order to objectively evaluate the performance of the two optimization methods, CBAM and FocalNext, ablation experiments were conducted. Method 1 is to add the CBAM module to the backbone network of YOLOv7, called YOLOv7-CBAM; Method 2 is to replace the ELAN module in YOLOv7 with the FocalNext module, called YOLOv7-Focal. The training hyperparameters are the same as those of YOLOv7.

[0097] The training and testing results of YOLOv7-CBAM are as follows Fig.11 As shown. On the training set, the average accuracy of YOLOv7-CBAM is 93.0%, of which the normal seeds are 94.5% and the germinated seeds are 91.4%; on the test set, the average accuracy drops to 92.12%, of which the normal seeds and germinated seeds drop by 0.6% and 0.28% respectively. This result shows that the generalization degree of the YOLOv7-CBAM model on the test set has decreased, which requires further optimization and improvement.

[0098] The training and testing results of YOLOv7-Focal are as follows Fig.12As shown. On the training set, the average accuracy of YOLOv7-Focal is 93.6%, of which the average accuracy of normal seeds is 94.3% and the average accuracy of germinated seeds is 93%. This result proves the excellent performance of the YOLOv7-Facal model. On the test set, the accuracy of normal seeds and germinated seeds only dropped by 0.09% and 0.01% respectively, indicating that the performance of the model is very stable. This result proves that YOLOv7-Facal has good performance in the seed detection task.

[0099] Through ablation experiments, the accuracy of each model is shown in Table 2. The mAP of YOLOv7 before optimization is 92.6%. After adding CBAM, the mAP increases to 92.9%, and after replacing ELAN with FocalNext, the mAP reaches 93.6%. This means that the FocalNext solution performs better than the CBAM solution. By adding CBAM and replacing ELAN with FocalNext at the same time, the mAP reaches 94.0%, indicating that the combined use of the two solutions can further improve the accuracy. From the comparative experimental results, it can be seen that the YOLOv7-CFocal model has the highest accuracy and is the optimal model for germinating seed detection.

[0100] Table 2 Ablation experiment results

[0101] Model Name CBAM FocalNext mAP YOLOv7 × × 92.6％ YOLOv7-CBAM √ × 92.9％ YOLOv7-Focal × √ 93.6％ YOLOv7-CFocal √ √ 94.0％

[0102] Calculation of germination rate

[0103] The calculation formula for germination rate is shown in Formula 5.

[0104]

[0105] Among them, P is the germination rate, T is the number of germinated seeds, and F is the normal number of seeds.

[0106] Ten images were randomly selected from the test set. In each image, the number of normal seeds and germinated seeds was manually counted, and the germination rate detected manually was calculated using formula (5). Then, the YOLOv7-CFocal model was used to detect and count normal seeds and germinated seeds, and the germination rate detected by the model was obtained. The results are shown in Table 3.

[0107] Table 3 Results of germination rate detection by manual and YOLOv7-CFocal models

[0108]

[0109] The minimum error of the germination rate between the model and manual detection is 0%, the maximum error is 3.1%, and the average error is less than 1%, which is acceptable in actual detection tasks.

[0110] It is worth noting that the contents not elaborated in detail in the present invention are all prior art and are well known to those skilled in the art.

[0111] Therefore, the present invention adopts the above-mentioned high-throughput intelligent detection method and system for germinated wheat seeds, and adopts an imaging system that automatically disperses seeds to make the collection of seed images faster and more efficient. The optimized YOLOv7-CFocal model based on the global-local attention and focusing mechanism significantly improves the accuracy of germinated seed detection.

[0112] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A high-throughput intelligent detection method for germinating wheat seeds, characterized in that: The following steps are involved: Step S1, initializing the intelligent detection system, setting the feeding time t1, the dispersion time t2 and the discharging time t3; Step S2, putting the seeds to be tested into the feed bin, the vibration motor on the feed table starts to vibrate, stops vibrating after t1, and the seeds fall into the loading box; Step S3, the vibration motor on the loading platform starts to vibrate, and stops vibrating after t2, and the seeds are distributed in the loading box; Step S4, turning on the light source through the light source controller, and taking pictures of the seeds in the material loading box by the camera to obtain a seed image; Step S5, the IO controller is used to control the discharging baffle to open, and the loading platform performs the discharging operation. After t3, the seeds in the loading box are removed; Step S6, repeating steps S2 to S5 until there are no seeds in the feed bin; Step S7, loading the acquired seed image, and detecting normal seeds and germinated seeds in the seed image through a germinated seed detection model; Step S8, counting the number of normal seeds and germinated seeds, and calculating the germination rate; Step S9, output the detection result; The germinating seed detection model is the YOLOv7-CFocal model. The YOLOv7-CFocal model adds the convolutional block attention module CBAM to the YOLOv7 model and replaces the ELAN module in YOLOv7 with the FocalNext module. The FocalNext module includes multiple focus blocks.

2. A high-throughput intelligent detection method for germinating wheat seeds according to claim 1, characterized in that: The convolutional block attention module (CBAM) consists of a channel attention module and a spatial attention module.

3. A system for implementing the high-throughput intelligent detection method for germinated wheat seeds according to any one of claims 1 to 2, characterized in that: It includes a feeding component, a discharging component and an identification component; The feeding assembly comprises a feeding platform, and a feeding bin is arranged on the upper side of the feeding platform; The discharging assembly comprises a base, a loading platform is arranged on the upper side of the base, a loading box is arranged on the upper side of the loading platform, and one side of the loading box is a discharging baffle; The identification component includes a light source, a camera is arranged on the upper side of the light source, and the lens of the camera is located directly above the light source; The feed bin is connected to the feed controller, and the feed platform, loading platform and discharge baffle are all connected to the IO controller.

4. A high-throughput intelligent detection system for germinating wheat seeds according to claim 3, characterized in that: The light source is a ring-shaped LED lamp, and the light source is connected to a light source controller.

5. A high-throughput intelligent detection system for germinating wheat seeds according to claim 4, characterized in that: Vibration motors are arranged at the four corners of the feeding platform and the loading platform.

6. A high-throughput intelligent detection system for germinating wheat seeds according to claim 5, characterized in that: The outlet of the feed bin is located just above the material loading box.

Images