Seed germination information detection method and device based on deep learning

Abstract

The invention discloses a seed germination information detection method and device based on deep learning, and belongs to the technical field of germination detection.The method comprises the steps that a soil environment is simulated to cultivate seeds; obtaining RGB images of the seeds before and after germination, and marking the obtained RGB images; dividing the marked RGB image into a training set and a verification set; the improved YOLOv10 model is trained and verified; obtaining a germination information prediction model; the seed germination information is detected based on the germination information prediction model. The invention further discloses a detection device. According to the method and the device, the improved YOLOv10 model is adopted, the seed phenotype information can be rapidly and accurately obtained, errors caused by human intervention are effectively reduced, and the detection efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of germination detection, and in particular to a seed germination information detection method and device based on deep learning. Background Art

[0002] Germination information is an important indicator for seed quality inspection. During seed storage, as the storage time increases, quality indicators such as germination rate, fatty acid value, and viscosity will change. Affected by the metabolites of microorganisms and pests, indicators such as fatty acid value, viscosity, and falling value cannot fully reflect the changes in the quality of the seeds themselves, while germination rate is not affected by these factors, so it has become the most stable and reliable indicator in seed quality inspection. The germination rate is closely related to grain processing and agricultural production, so it is very necessary to quickly and accurately detect the seed germination rate.

[0003] At present, there are the following problems in germination rate detection: 1) Traditional seed germination information acquisition requires a lot of manpower, and each measurement requires at least 2-3 people; 2) Traditional methods for evaluating seed germination mainly rely on manual work, and the estimation of germination percentage is easily affected by subjective factors, which can easily lead to human errors caused by mis-filling or incorrect filling of germination rate, resulting in unreliable data; 3) The traditional germination detection method is mainly hydroponics, which is to put the seeds in clean water to observe whether they can quickly absorb water and germinate. If the seeds do not germinate within two to three days, it may mean that the seeds have died or have been interfered by some external factors. Although this method is simple, it cannot completely simulate the actual germination environment of seeds in the soil; 4) The large amount of image data generated by existing germination detection technology requires efficient algorithms to process and analyze, which has high requirements for computing resources, resulting in long data processing time and the inability to calculate bud length; 5) In the prior art, seed germination boxes, as important equipment in agricultural scientific research and production, are widely used to promote seed germination and growth. However, traditional germination boxes have problems such as inconvenient operation and low efficiency during the seed cultivation process. Especially when large-scale seed processing is required, the movement, observation and recording of seeds become particularly cumbersome. Summary of the invention

[0004] The purpose of the present invention is to provide a seed germination information detection method and device based on deep learning to solve the problems mentioned in the above background technology.

[0005] To achieve the above object, the present invention provides a seed germination information detection method based on deep learning, the steps comprising: S1. Clean and sterilize a culture dish, place a layer of moistened filter paper in the culture dish, place the seeds in the culture dish, and place the culture dish in the detection device; S2, simulate soil environment and cultivate seeds; S3, obtaining RGB images of seeds before and after germination, and annotating the obtained RGB images; S4, dividing the annotated RGB images into a training set and a validation set, and performing data augmentation on the training set and the validation set; S5. Use the training set to train the improved YOLOv10 model, and use the validation set to verify it, to obtain a germination information prediction model; S6. Using a test set to detect seed germination information based on a germination information prediction model, the test set is the same as the verification set image.

[0006] Preferably, the simulating soil environment in step S2 is to simulate the temperature, humidity and light intensity in natural soil by controlling the temperature, humidity and light intensity in the detection device.

[0007] Preferably, the data enhancement in step S4 includes image rotation, scaling, flipping, illumination enhancement, random noise and color jittering.

[0008] Preferably, the improved YOLOv10 model in step S5 includes an input layer, a backbone feature extraction layer, an enhanced feature network, a detection head and a data acquisition module which are arranged in sequence, the backbone feature extraction layer introduces a CBAM attention mechanism, the detection head includes a one-to-one detection head and a one-to-many detection head, and the data acquisition module includes a counting module and a bud length measurement module.

[0009] Preferably, the counting module counts the number of germinated seeds and ungerminated seeds, and calculates the number of germinated seeds and the total number of seeds to obtain the seed germination rate.

[0010] Preferably, the bud length measurement module detects the starting point and the end point of the bud and calculates its length.

[0011] Preferably, the objective function of the improved YOLOv10 model training is: ; In the formula, represents the total loss, represents the classification loss, is the bounding box regression loss, Indicates that the target has a loss. and Represents the weight coefficient.

[0012] A seed germination information detection device based on deep learning comprises a box, a conveyor belt system, a monitoring system, an environment simulation module, a voice module, a control system and a display, wherein the display is arranged on one side outside the box, the display is connected to the control system and the monitoring system, the monitoring system is connected to the voice module, the voice module is arranged on one side inside the box, the conveyor belt system and the environment simulation module are both arranged inside the box, the conveyor belt system and the environment simulation module are both connected to the control system, the monitoring system comprises a high-definition camera, an RGB camera and a host, the high-definition camera and the RGB camera are located above the conveyor belt system, and the host is arranged below the conveyor belt system.

[0013] Preferably, the environment simulation module includes a humidity control module, a temperature control module and a light intensity control module.

[0014] Therefore, the present invention adopts the above-mentioned seed germination information detection method and device based on deep learning, which has the following beneficial effects: (1) Using the improved YOLOv10 model for detection makes the acquisition of seed phenotypic information faster and more accurate, saving time and eliminating the need for breeders to check and measure each seed one by one; (2) Analyze the collected RGB images to achieve non-destructive measurement; (3) The improved YOLOv10 model is equipped with a counting module and a bud length measurement module. By introducing these automated calculation modules, the measurement of germination rate and bud length becomes more reliable, greatly reducing the interference of human errors; (4) The detection equipment is automatically controlled through the controller to improve the intelligence level of the equipment and improve work efficiency.

[0015] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A method flow chart of an embodiment of a seed germination information detection method based on deep learning of the present invention; Figure 2 This is a framework diagram of an improved YOLOv10 model of an embodiment of a seed germination information detection method based on deep learning of the present invention; Figure 3 This is a structural block diagram of an embodiment of a seed germination information detection device based on deep learning of the present invention; Reference numerals 1. Display; 2. Light; 3. Voice module; 4. Light sensor; 5. RGB camera; 6. Humidity sensor; 7. Control system; 8. Spray device; 9. Conveyor system; 10. Host; 11. Box. DETAILED DESCRIPTION

[0017] Example In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments.

[0018] Reference Figure 1 The present invention provides a seed germination information detection method based on deep learning, the steps comprising: S1. Clean and sterilize the culture dish to ensure a sterile environment for seed cultivation. Place a layer of moist filter paper in the culture dish to cover the entire bottom of the culture dish to provide suitable conditions for seed germination. Then place the seeds in the culture dish and place the culture dish in the detection device.

[0019] S2. Simulate the soil environment to cultivate seeds. By controlling the temperature, humidity and light intensity in the detection device to simulate the temperature, humidity and light conditions in natural soil, it ensures that the seeds germinate in a state close to nature, overcoming the environmental differences of traditional hydroponics.

[0020] Temperature, humidity and light intensity follow the formula: ; ; In the formula, , and represent temperature, humidity and light intensity respectively. , and Indicates the environmental conditions of natural soil, , and Indicates the allowable deviation range.

[0021] S3. Obtain RGB images of seeds before and after germination, and use Labelme 3.16.7 labeling software to annotate the obtained RGB images.

[0022] S4. Divide the labeled RGB images into training set and validation set in a ratio of 9:1, and perform data enhancement on the training set and validation set. Use data enhancement techniques such as image rotation (0° to 360°), scaling (0.5 to 2.0), flipping, illumination enhancement, random noise, and color jitter to enhance the data. The formulas for the training set and validation set after data enhancement are: In the formula, represents the training set, represents the validation set, Represents an image.

[0023] Lighting enhancement: simulate image changes under different lighting conditions to avoid the decrease in detection accuracy caused by changes in ambient light in actual detection. Random noise: add random Gaussian noise or salt and pepper noise to the image to simulate interference caused by image sensor instability or ambient noise. Color jitter: make the model more robust to different light and color environments by randomly adjusting brightness, contrast, and hue.

[0024] Data augmentation will effectively improve the robustness of the model under different lighting conditions and ensure the accuracy of germination rate and bud length detection. In addition, during model training, data-augmented images can simulate a variety of possible seed growth environments, further improving the model's ability to cope with complex actual scenarios. Especially in large-scale agricultural scenarios, environmental parameters such as lighting, temperature and humidity may be inconsistent, and the diversity of data augmentation strategies will greatly reduce the false alarm rate of the detection model.

[0025] S5. Use the training set to train the improved YOLOv10 model and use the validation set to verify it to obtain the germination information prediction model.

[0026] The improved YOLOv10 model is as follows Figure 2As shown in the figure, the input layer, the backbone feature extraction layer, the enhanced feature network, the detection head and the data acquisition module are arranged in sequence. On the basis of the original network structure, that is, in the backbone feature extraction layer, the CBAM (Convolutional Block Attention Module) attention mechanism is introduced to improve the model's ability to capture germination features, especially the starting point and end point of the bud. The CBAM module can adaptively weight the space and channel to improve the model's ability to capture germination features under complex backgrounds. Specifically, CBAM can allow the model to focus more on the key areas of seed germination, such as the starting point and end point of the bud, reduce the sensitivity to environmental interference information, and improve the detection effect. The enhanced feature network is set so that seed features of different scales can be effectively detected, which is suitable for detecting seeds of different sizes and germination states at different periods. The detection head is divided into a one-to-one detection head and a one-to-many detection head. Through the setting of the "one-to-one" and "one-to-many" detection heads, the model can flexibly respond to the germination information detection needs of a single seed and multiple seeds, further improving the detection accuracy. The data acquisition module includes a counting module and a bud length measurement module. The counting module counts the number of germinated and ungerminated seeds, and calculates the number of germinated seeds and the total number of seeds to obtain the seed germination rate. Setting up the counting module can not only quickly and accurately give the overall germination situation, but also adapt to detection tasks of different scales. The bud length measurement module detects the starting point and end point of the bud and calculates its length, eliminating the inconvenience caused by manual measurement. For semi-germinated or abnormally germinated seeds, the bud length measurement module can also give accurate judgments, thereby ensuring the integrity of germination information.

[0027] The objective function of the improved YOLOv10 model training is: ; In the formula, represents the total loss, represents the classification loss, is the bounding box regression loss, Indicates that the target has a loss. and Represents the weight coefficient.

[0028] The specific process includes: the seed germination image after data enhancement enters the backbone feature extraction network through the input layer, extracts the features of the seeds, and then undergoes multiple feature samplings to obtain three effective feature layers of different sizes. Then, these feature layers pass through the enhanced feature network. The CBAM attention mechanism adaptively learns the importance of each channel and spatial position in the backbone feature extraction layer, so as to better capture the key information in the three effective feature layers of different sizes, and finally obtains the feature information after three layers of processing. The three layers of feature information are passed to the detection head for seed classification. The classified seed information is calculated by the data acquisition module to obtain the germination rate and bud length of the seeds. After adding the CBAM attention mechanism, the effect of the model is improved by about 3%, which significantly improves the ability to capture germination features.

[0029] S6. Using a test set to detect seed germination information based on a germination information prediction model, the test set is the same as the verification set image.

[0030] like Figure 3 As shown, a seed germination information detection device based on deep learning includes a box 11, a conveyor belt system 9, a monitoring system, an environment simulation module, a voice module 3, a control system 7 and a display 1. The display 1 is arranged on the outer side of the box 11, the display 1 is connected to the control system 7 and the monitoring system, the monitoring system is connected to the voice module 3, the voice module is arranged on the inner side of the box 11, the conveyor belt system 9 and the environment simulation module are both arranged inside the box 11, the conveyor belt system 9 and the environment simulation module are both connected to the control system 7, the control system 7 is a plc control system, the monitoring system includes a high-definition camera, an RGB camera 5 and a host 10, the high-definition camera and the RGB camera 5 are located above the conveyor belt system 9, and the host 10 is arranged below the conveyor belt system 9.

[0031] The environmental simulation module includes a humidity control module, a temperature control module and a light intensity control module.

[0032] The humidity adjustment module includes a spray device 8 and a humidity sensor 6, which are both arranged on the side corresponding to the voice module 3, and keep the internal moisture by releasing water vapor. The light intensity control module includes a light lamp 2 and a light sensor 4, which are both arranged above the box 11 to simulate the natural light cycle. The temperature control module adjusts the ambient temperature through a temperature sensor. The environmental simulation module 3 monitors the environmental conditions in real time through sensors and adjusts the environmental conditions through the control system 7 to ensure that the seeds in the culture dish are in the optimal environment at different germination stages.

[0033] The control system 7 controls the transmission belt system 2 and the environmental simulation module 3 to ensure the reliable operation of the equipment under different conditions. The control system 7 is connected to the Internet of Things platform to remotely adjust the growth environment of the seeds and further realize intelligent management and control.

[0034] The conveyor belt system 9 is driven by a motor and arranged horizontally or obliquely along the length of the box. Several seed trays can be installed on the conveyor belt to carry and move seeds. The conveyor belt system 9 is controlled by the control system 7 to ensure the continuity and safety of the detection process. When the culture dish moves to the detection area, the conveyor belt will stop automatically. This process is achieved through a precise photoelectric sensor or position detection device. After the culture dish is in place, the high-precision RGB camera installed above the detection area will automatically start to take pictures of the seeds in the culture dish and collect images of seed germination. RGB cameras and high-definition cameras can capture high-resolution images in real time to ensure that the details of seed germination are clearly visible.

[0035] The monitoring system includes a high-definition camera, an RGB camera 5 and a host 10, and the high-definition camera and the RGB camera 5 are located above the conveyor belt system 9. The high-definition camera monitors the seeds on the conveyor belt in real time, and the germination of the seeds can be monitored through a mobile device or a PC remote terminal. The RGB camera 5 can capture high-resolution images in real time and transmit the image information to the host 10, and the host 10 detects the germination of the seeds. After the detection is completed, the germination of the seeds is displayed through the display 1, and the host 10 uploads the detection data to the cloud, which is convenient for users to view anytime and anywhere. The voice module 4 completes the monitoring through voice prompts, which is convenient for users to operate.

[0036] 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 seed germination information detection method based on deep learning, characterized in that the steps include: S1. Clean and sterilize a culture dish, place a layer of moistened filter paper in the culture dish, place the seeds in the culture dish, and place the culture dish in the detection device; S2, simulate soil environment and cultivate seeds; S3, obtaining RGB images of seeds before and after germination, and annotating the obtained RGB images; S4, dividing the annotated RGB images into a training set and a validation set, and performing data augmentation on the training set and the validation set; S5. Use the training set to train the improved YOLOv10 model, and use the validation set to verify it, to obtain a germination information prediction model; S6. Using a test set to detect seed germination information based on a germination information prediction model, the test set is the same as the verification set image.

2. A seed germination information detection method based on deep learning according to claim 1, characterized in that: The simulated soil environment in step S2 is to simulate the temperature, humidity and light intensity in natural soil by controlling the temperature, humidity and light intensity in the detection device.

3. A seed germination information detection method based on deep learning according to claim 1, characterized in that: The data enhancement in step S4 includes image rotation, scaling, flipping, illumination enhancement, random noise and color jittering.

4. The seed germination information detection method based on deep learning according to claim 1, characterized in that: The improved YOLOv10 model in step S5 includes an input layer, a backbone feature extraction layer, an enhanced feature network, a detection head and a data acquisition module which are arranged in sequence. The CBAM attention mechanism is introduced into the backbone feature extraction layer. The detection head includes a one-to-one detection head and a one-to-many detection head. The data acquisition module includes a counting module and a bud length measurement module.

5. A seed germination information detection method based on deep learning according to claim 4, characterized in that: The counting module counts the number of germinated seeds and ungerminated seeds, and calculates the number of germinated seeds and the total number of seeds to obtain the seed germination rate.

6. A seed germination information detection method based on deep learning according to claim 4, characterized in that: The bud length measurement module detects the starting point and the end point of the bud and calculates its length.

7. A seed germination information detection method based on deep learning according to claim 4, characterized in that: The objective function of the improved YOLOv10 model training is: ； In the formula, represents the total loss, represents the classification loss, is the bounding box regression loss, Indicates that the target has a loss. and Represents the weight coefficient.

8. A detection device using a seed germination information detection method based on deep learning as described in any one of claims 1 to 7, characterized in that: It includes a box, a conveyor belt system, a monitoring system, an environment simulation module, a voice module, a control system and a display, wherein the display is arranged on a side outside the box, the display is connected to the control system and the monitoring system, the monitoring system is connected to the voice module, the voice module is arranged on a side inside the box, the conveyor belt system and the environment simulation module are both arranged inside the box, the conveyor belt system and the environment simulation module are both connected to the control system, the monitoring system includes a high-definition camera, an RGB camera and a host, the high-definition camera and the RGB camera are located above the conveyor belt system, and the host is arranged below the conveyor belt system.

9. A seed germination information detection device based on deep learning according to claim 8, characterized in that: The environment simulation module includes a humidity control module, a temperature control module and a light intensity control module.

Images