A corn ear damage detection method based on deep learning and embedded

Abstract

The present invention discloses a method for detecting corn ear damage based on deep learning and embedded systems, and relates to the technical field of agricultural machinery. The method comprises: obtaining image data in an ear collecting box in real time when a harvester is working, and processing the image data through a pre-trained deep learning model to obtain a damage classification result; pre-processing each image data in the damage classification result, and calculating the proportion of the damaged part of the pre-processed image data, and saving the calculated proportion result. The present invention can timely and accurately determine the proportion of corn damage occurring during corn harvesting through a pre-trained deep learning model. In addition, by calculating the proportion of the damaged part, a part of data support can be further provided for subsequent harvesting schemes or improvements to harvesters.

Description

Background Art

[0002] In recent years, mechanized harvesting technology has been widely used in corn picking, peeling and threshing. However, in the process of mechanical ear picking, inappropriate mechanical operating parameters will lead to grain loss. If the mechanical operating parameters are not discovered and adjusted in time, it will lead to serious food waste. At present, during mechanized corn harvesting, the driver needs to stop the car frequently to get grain, manually check the grain loss, and then adjust the mechanical operating parameters. This affects the harvesting efficiency and the grain loss cannot be observed in time. Therefore, it is very necessary to develop a system for online detection of corn grain loss.

[0003] At present, the main information sources for non-contact detection of damage or injury in the agricultural field are multispectral images, hyperspectral images and RGB images. Among them, multispectral images and hyperspectral images are famous for the diversity of information they contain, but if they are used to detect corn damage in the ear box during field harvesting operations, these two images have the following disadvantages compared to RGB images:

[0004] 1. Expensive equipment: Multispectral and hyperspectral information collection requires specialized equipment, among which hyperspectral cameras are quite expensive and are not widely used in the field of corn harvesting.

[0005] 2. Greatly affected by the external environment: Multispectral and hyperspectral images have high requirements for ambient light sources, and even the slightest change will have a significant impact on the image quality. A very stable state is required during the collection of image information. Slight shaking or other objects passing in front of the lens will form image ghosting. They are completely unsuitable for complex working conditions during field harvesting.

[0006] 3. It takes a long time to obtain information and the amount of calculation is huge: Multispectral and hyperspectral equipment takes a long time to collect images. Although they contain more information and have a higher accuracy rate when performing detection, it is precisely because of the complexity of the information that the amount of calculation is large, detection and identification take a long time, and real-time detection cannot be achieved in a short time. Summary of the invention

[0007] The technical problem to be solved by the present invention is to address the deficiencies of the prior art and specifically provide a corn ear damage detection method based on deep learning and embedded systems, as follows:

[0008] 1) In the first aspect, the present invention provides a method for detecting corn ear damage based on deep learning and embedded systems, and the specific technical scheme is as follows:

[0009] Real-time acquisition of image data in the ear collection box when the harvester is working, and processing the image data through a pre-trained deep learning model to obtain a damage classification result;

[0010] Each image data in the damage classification result is preprocessed, and the proportion of the damaged part of the preprocessed image data is calculated, and the calculation result of the proportion is saved.

[0011] The beneficial effects of the corn ear damage detection method based on deep learning and embedded system provided by the present invention are as follows:

[0012] The pre-trained deep learning model can timely and accurately determine the proportion of corn damage that occurs during the corn harvesting process. In addition, by calculating the proportion of the damaged part, it can further provide some data support for subsequent harvesting plans or improvements to harvesters.

[0013] Based on the above solution, the present invention can also be improved as follows.

[0014] Furthermore, the preprocessing includes: marking the image data to generate a marking result, generating a mask image based on the marking result, and performing visual marking and grayscale processing on the mask image.

[0015] Furthermore, the process of calculating the percentage of the damaged part of the pre-processed image data is as follows:

[0016] The area of ​​the background part, the area of ​​the intact corn ear part and the area of ​​the damaged corn ear part are determined, and the proportion of the area of ​​the damaged corn ear part in the area excluding the area of ​​the background part is determined.

[0017] Furthermore, the image data is processed by a pre-trained deep learning model to obtain the injury classification result in the following specific process:

[0018] The image data is subjected to image segmentation processing and image enhancement processing to obtain optimized image data, and the optimized image data is subjected to pixel-level classification through the pre-trained deep learning model to obtain a damage classification result, wherein the damage classification result includes at least one of an undamaged ear area, a damaged ear area, a corn husk area, and a background area.

[0019] Furthermore, the pre-trained deep learning model is constructed by calling the UNet model in segmentation_models_pytorch and using efficiennet-b3 as an encoder in combination with a deep classifiable convolutional layer.

[0020] Furthermore, it also includes:

[0021] Based on the calculation result of the proportion and combined with the manual label, the evaluation index of the image data is calculated, and the evaluation index includes: at least one of average intersection-over-union ratio, overall classification accuracy, average classification accuracy, classification accuracy of each category, false alarm rate and missed alarm rate.

[0022] 2) In the second aspect, the present invention also provides a corn ear damage detection system based on deep learning and embedded system, and the specific technical scheme is as follows:

[0023] The acquisition module is used to: acquire the image data in the ear collection box in real time when the harvester is working, and process the image data through a pre-trained deep learning model to obtain a damage classification result;

[0024] The detection module is used to: pre-process each image data in the damage classification result, calculate the proportion of the damaged part of the pre-processed image data, and save the calculation result of the proportion.

[0025] 3) In a third aspect, the present invention further provides an electronic device, comprising a processor, wherein the processor is coupled to a memory, wherein the memory stores at least one computer program, and wherein the at least one computer program is loaded and executed by the processor so that the electronic device implements any of the above methods.

[0026] 4) In a fourth aspect, the present invention further provides a computer-readable storage medium, wherein at least one computer program is stored in the computer-readable storage medium, and the at least one computer program is loaded and executed by a processor to enable a computer to implement any of the above methods.

[0027] It should be noted that the beneficial effects achieved by the technical solutions of the second to fourth aspects of the present invention and the corresponding possible implementation methods can be found in the above-mentioned technical effects of the first aspect and its corresponding possible implementation methods, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0029] Figure 1 A schematic diagram of a process of a corn ear damage detection method based on deep learning and embedded system according to an embodiment of the present invention;

[0030] Figure 2 A specific flow chart of training a pixel-level damage classification network model for a corn ear damage detection method based on deep learning and embeddedness according to an embodiment of the present invention;

[0031] Figure 3The original image of the corn ear box during harvester operation is obtained by taking a photo with a camera of a method for detecting corn ear damage based on deep learning and embedded system in accordance with an embodiment of the present invention;

[0032] Figure 4 The original image obtained after image segmentation of a corn ear damage detection method based on deep learning and embedded system according to an embodiment of the present invention;

[0033] Figure 5 A visual marking diagram of corn damage prediction of each part of a corn ear damage detection method based on deep learning and embedded system according to an embodiment of the present invention;

[0034] Figure 6 The figure is a schematic diagram of the structure of an electronic device. DETAILED DESCRIPTION

[0035] In order to make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0036] like Figure 1 As shown, a method for detecting corn ear damage based on deep learning and embedded system in an embodiment of the present invention comprises the following steps:

[0037] S1, real-time acquisition of image data in the ear collecting box when the harvester is working, and processing the image data through a pre-trained deep learning model to obtain a damage classification result;

[0038] S2, preprocessing each image data in the damage classification result, calculating the proportion of the damaged part of the preprocessed image data, and saving the calculation result of the proportion.

[0039] The beneficial effects of the corn ear damage detection method based on deep learning and embedded system provided by the present invention are as follows:

[0040] The pre-trained deep learning model can timely and accurately determine the proportion of corn damage that occurs during the corn harvesting process. In addition, by calculating the proportion of the damaged part, it can further provide some data support for subsequent harvesting plans or improvements to harvesters.

[0041] S1, real-time acquisition of image data in the collecting box when the harvester is working, and processing the image data through a pre-trained deep learning model to obtain damage classification results.

[0042] The image data in the collecting box is collected by a camera installed above the collecting box.

[0043] The pre-trained deep learning model is trained and constructed through the training set and the validation set. The training set and the validation set are both from the sample set. The sample set is made by creating samples for different light intensities, different backgrounds, and different accumulation conditions of fruit ears in the collection box. In the sample set, 80% can be set as the training set and 20% as the test set.

[0044] It should be further explained that, in the process of constructing the sample set, it is determined whether the number of samples in the sample set meets the preset requirements. If so, the data in the sample set is directly used for subsequent model training. If not, the sample set is expanded.

[0045] The specific process of expansion processing can be:

[0046] Select an expansion method, that is, select an expansion method suitable for the sample set from random flipping, random cropping, random color change, translation, rotation, scaling, etc.

[0047] The process of selecting an expansion method is as follows:

[0048] Each image data in the sample set is subjected to random combination processing of random flipping, random cropping, random color change, translation, rotation, and scaling, and the combination components of each random combination and the processing results after the combination processing are recorded, and the feature similarity between any image data in the sample set and the image data in the corresponding processing result after the combination processing is determined, and under the combination processing, the arithmetic mean of all feature similarities is calculated to obtain the score corresponding to the random combination processing, and the distribution of feature similarity is confirmed (such as within which range the feature similarity concentration exceeds the preset concentration). After determining the combination components, scores and distribution under each combination processing, according to the preset similarity range, n combination processings with the highest overlap with the preset similarity range are determined, and the scores closest to the center value of the preset similarity range among the above n combination processings are further determined, and the combination processing is used as an expansion method for subsequent expansion processing. The center value refers to the median in the preset similarity range.

[0049] The above-mentioned method of selecting expansion can improve the diversity of expansion methods through random combination processing, and can also make the expansion results meet both the training requirements and the quantity requirements. The arithmetic mean is introduced in the process of selecting the expansion method, and the final combination method is determined by the difference between the arithmetic mean and the center value in the subsequent final selection. This is because the arithmetic mean is more susceptible to the influence of extreme values. When the arithmetic mean is closer to the center value of the preset similarity range, it means that this combination processing method will not appear in the original image. The deviation is too large or completely deviates from the original image. Based on this, the arithmetic mean is simple to operate on the one hand, and on the other hand, it can be used to perform preliminary verification of whether the expansion result can meet the subsequent use conditions.

[0050] The network model structure calls the UNet model in segmentation_models_pytorch and uses efficiennet-b3 as its encoder. The efficiennet series network mainly uses a deep separable convolutional layer to reduce the amount of calculation. The model is trained with the training set after image segmentation and image enhancement, and the parameters of the network are optimized;

[0051] For the training set, in order to expand the training samples as much as possible, the step size during cutting is set to 32. The test set does not require sample expansion, so the step size is set to 155. Use cv2.imread to read the image, slide on the original image with the set step size, extract the image and label blocks, use the math.ceil() function to set the processed image width and height, and generate a new file name, use cv2.imwrite to save the image, and save the images in .png format, so that the label data can be guaranteed not to be changed.

[0052] Perform image preprocessing on image data, including image segmentation and image enhancement preprocessing, to reduce the impact of lighting.

[0053] It should be noted that the preprocessing uses Python programming language and calls the opencv library for preprocessing. The acquired image enhancement is to add data enhancement on the basis of image segmentation, use the random series functions to randomly transform brightness, random transformation contrast, random transformation saturation, use the flip series functions to perform horizontal mirroring, vertical mirroring, horizontal and vertical mirroring, and use the rotate() function to rotate 90 degrees and -90 degrees.

[0054] When any image data is input into the pre-trained deep learning model, the trained neural network is called to perform pixel-level label prediction on the test set after image segmentation.

[0055] It should be further explained that throughout the entire model code, the following contents need to be defined: define the data set class corn (data), define the data set reading function including: define prvider, define the evaluation index function including intersection ratio, overall accuracy, average accuracy, etc., define the creation, appending, and reading of cvs table functions, define the category color function, and load the model. Among them, defining the creation, appending, and reading of cvs table functions is to create and implement the appending reading and recording of a csv file of the above evaluation indicators. Define the category color function to be consistent with the previously trained label to facilitate subsequent calculations. Loading the model is to import the best kappa model obtained in the previous pre-training.

[0056] During recognition, the entire image is divided into small areas in the form of a moving window, and then the small areas are classified at the pixel level to determine whether they belong to the undamaged area, the damaged area, or the background area.

[0057] It should be further explained that the process of image segmentation by moving the window is as follows:

[0058] In order to improve the efficiency of segmentation, the pixel values ​​of the four corner points are determined to further identify which areas are background areas.

[0059] Determine whether the image in the sliding window, that is, the image in the sliding window, contains background pixels (generally filled with fixed values, such as grayscale value 0). If the pixel values ​​at the four corner points of the window are not equal to the background values, there are no background pixels; otherwise, there are background pixels; it should be noted that the pixel values ​​and background values ​​here refer to fixed values ​​set artificially.

[0060] S2, preprocessing each image data in the damage classification result, calculating the proportion of the damaged part of the preprocessed image data, and saving the calculation result of the proportion.

[0061] Preprocessing includes marking the classified images, integrating the marks to form a mask image, and visually marking and grayscale processing the mask image and saving it.

[0062] The marking process refers to marking the ears, leaves, damage and background respectively and assigning different colors, where the yellow area represents the ear area, the red area represents the damaged corn ear area, the green area represents the leaf area, and the remaining black area represents the background area. The final output is a visual marking image.

[0063] The corn damage ratio P1 can be calculated by the following formula based on the three parameters: damaged corn area S1, intact ear area S2, and leaf area S3:

[0064] .

[0065] Furthermore, the preprocessing includes: marking the image data to generate a marking result, generating a mask image based on the marking result, and performing visual marking and grayscale processing on the mask image.

[0066] Furthermore, the process of calculating the percentage of the damaged part of the pre-processed image data is as follows:

[0067] The area of ​​the background part, the area of ​​the intact corn ear part and the area of ​​the damaged corn ear part are determined, and the proportion of the area of ​​the damaged corn ear part in the area excluding the area of ​​the background part is determined.

[0068] Furthermore, the image data is processed by a pre-trained deep learning model to obtain the injury classification result in the following specific process:

[0069] The image data is subjected to image segmentation processing and image enhancement processing to obtain optimized image data, and the optimized image data is subjected to pixel-level classification through the pre-trained deep learning model to obtain a damage classification result, wherein the damage classification result includes at least one of an undamaged ear area, a damaged ear area, a corn husk area, and a background area.

[0070] Furthermore, the pre-trained deep learning model is constructed by calling the UNet model in segmentation_models_pytorch and using efficiennet-b3 as an encoder in combination with a deep classifiable convolutional layer.

[0071] Furthermore, it also includes:

[0072] Based on the calculation result of the proportion and combined with the manual label, the evaluation index of the image data is calculated, and the evaluation index includes: at least one of average intersection-over-union ratio, overall classification accuracy, average classification accuracy, classification accuracy of each category, false alarm rate and missed alarm rate.

[0073] Embodiment 1, as Figure 2 As shown, the following steps are included:

[0074] Step 1: First, collect some sample pictures and make samples for different light intensities, different backgrounds, and different stacking conditions of ears in the ear collection box, including four categories: undamaged ear area, damaged ear area, corn husk, and background.

[0075] Step 2: Perform image preprocessing on the sample images on a high-performance computer, including image segmentation and image enhancement preprocessing to reduce the impact of lighting. Image segmentation is because the images in the real harvesting scene are cropped to remove the background that should not appear, and the image size is not uniform. After comprehensive consideration, the images are cut into 256×256 images as the input of the segmentation network.

[0076] Step 3: Use the preprocessed sample images to train an appropriate neural network model.

[0077] Step 4: During the test recognition, the image is classified at the pixel level to determine whether it belongs to the undamaged area, damaged area, or background area.

[0078] Step 5: Mark the classified images, integrate the marks to form a mask image, and perform visual marking and grayscale processing on the mask image and save it.

[0079] Step 6: Calculate the area and damage ratio of each part, and compare them with the manually labeled map to calculate the evaluation indicators, including average intersection-over-union ratio, overall classification accuracy, average classification accuracy, classification accuracy of each category, false alarm rate, and missed alarm rate.

[0080] Step 7: Save the calculation results and evaluation indicators to a table.

[0081] Step 3 also requires the subsequent steps 5 to 7 to select the optimal training model based on the evaluation indicators and the visual labeling map.

[0082] After comparing various evaluation indicators, the best training model is selected and imported into the embedded system from the computer. The specific process of the corn ear damage detection system and method based on deep learning and embedded system when the corn harvester is working is as follows:

[0083] Step 1: The camera is installed on the corn harvester's ear collecting box to obtain real-time images of corn ears in the ear collecting box when the harvester is working, and transmit the images to the embedded system.

[0084] Step 2: The embedded system reads the corn ear harvest image transmitted by the camera, such as Figure 3 shown.

[0085] Step 3: The embedded system imports several corn ear images into the pre-trained and tested pixel-level corn ear damage classification network, and first segment them to obtain Figure 4 , damage prediction is performed on these images.

[0086] Step 4: Mark the predicted image and integrate the marks to form a mask image. Visualize and grayscale the mask image and save it. The color prediction visualization image is as follows: Figure 5 shown.

[0087] Step 5: Calculate the area of ​​each part of the image in pixels, including the background, leaves, intact corn ears, and damaged corn ears. Then calculate the proportion of damaged corn ears in the area other than the background.

[0088] Step 6: Save the damaged area and percentage of corn ears to an Excel spreadsheet.

[0089] In the above embodiments, although the steps are numbered S1, S2, etc., these are only specific embodiments given by the present invention. Those skilled in the art may adjust the execution order of S1, S2, etc. according to actual conditions, which is also within the protection scope of the present invention. It can be understood that in some embodiments, some or all of the above embodiments may be included.

[0090] This solution also provides a corn ear damage detection system based on deep learning and embedded system. The specific technical solution is as follows:

[0091] The acquisition module is used to: acquire the image data in the ear collection box in real time when the harvester is working, and process the image data through a pre-trained deep learning model to obtain a damage classification result;

[0092] The detection module is used to: pre-process each image data in the damage classification result, calculate the proportion of the damaged part of the pre-processed image data, and save the calculation result of the proportion.

[0093] It should be noted that the beneficial effects of the deep learning and embedded corn ear damage detection system provided in the above embodiment are the same as the beneficial effects of the deep learning and embedded corn ear damage detection method provided in the above embodiment, which will not be repeated here. In addition, when the system provided in the above embodiment realizes its functions, it only uses the division of the above functional modules as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the system can be divided into different functional modules according to actual conditions to complete all or part of the functions described above. In addition, the system and method embodiments provided in the above embodiment belong to the same concept, and the specific implementation process is detailed in the method embodiment, which will not be repeated here.

[0094] like Figure 6 As shown, an electronic device 300 according to an embodiment of the present invention includes a processor 320, the processor 320 is coupled to a memory 310, and the memory 310 stores at least one computer program 330, and the at least one computer program 330 is loaded and executed by the processor 320, so that the electronic device 300 implements any of the above methods, specifically:

[0095] The electronic device 300 may have relatively large differences due to different configurations or performances, and may include one or more processors 320 (Central Processing Units, CPU) and one or more memories 310, wherein the one or more memories 310 store at least one computer program 330, and the at least one computer program 330 is loaded and executed by the one or more processors 320, so that the electronic device 300 implements a method for detecting corn ear damage based on deep learning and embedded systems provided in the above-mentioned embodiment. Of course, the electronic device 300 may also have components such as a wired or wireless network interface, a keyboard, and an input and output interface for input and output, and the electronic device 300 may also include other components for implementing device functions, which will not be described in detail here.

[0096] A computer-readable storage medium according to an embodiment of the present invention stores at least one computer program, and the at least one computer program is loaded and executed by a processor so that a computer implements any one of the above methods.

[0097] Optionally, the computer readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc (CD-ROM), a magnetic tape, a floppy disk, an optical data storage device, etc.

[0098] In an exemplary embodiment, a computer program product or a computer program is also provided, the computer program product or the computer program comprising computer instructions, the computer instructions being stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the electronic device performs any of the above methods.

[0099] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects and represent the definition of a specific order or sequence. The order of use of similar objects can be interchanged where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than the order shown or described.

[0100] Those skilled in the art know that the present invention can be implemented as a system, method or computer program product. Therefore, the present disclosure can be specifically implemented in the following forms, namely: it can be complete hardware, it can be complete software (including firmware, resident software, microcode, etc.), or it can be a combination of hardware and software, which is generally referred to as "circuit", "module" or "system" herein. In addition, in some embodiments, the present invention can also be implemented in the form of a computer program product in one or more computer-readable media, and the computer-readable medium contains computer-readable program code.

[0101] Any combination of one or more computer-readable media may be used. A computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, device, or device.

[0102] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A corn ear damage detection method based on deep learning and embedded system, characterized in that: include: Real-time acquisition of image data in the ear collection box when the harvester is working, and processing the image data through a pre-trained deep learning model to obtain a damage classification result; Preprocessing each image data in the damage classification result, calculating the proportion of the damaged part of the preprocessed image data, and saving the calculation result of the proportion; The pre-trained deep learning model is trained and constructed through the training set and the validation set, where both the training set and the validation set come from the sample set. During the sample set construction process, it is determined whether the number of samples in the sample set meets the preset requirements. If so, the data in the sample set is directly used for subsequent model training. If not, the sample set is expanded. The specific process of expansion processing is as follows: Select an expansion method, that is, select an expansion method for the sample set from random flipping, random cropping, random color change, translation, rotation, and scaling; The process of selecting an expansion method is as follows: Perform random combination processing of random flipping, random cropping, random color change, translation, rotation, and scaling on each image data in the sample set, and record the combination components of each random combination and the processing results after the combination processing, determine the feature similarity between any image data in the sample set and the image data in the corresponding processing result after the combination processing, and calculate the arithmetic mean of all feature similarities under the combination processing to obtain the score corresponding to the random combination processing, and confirm the distribution of feature similarity. After determining the combination components, scores and distribution under each combination processing, determine the n combination processings with the highest overlap with the preset similarity range according to the preset similarity range, further determine the score closest to the center value of the preset similarity range among the above n combination processings, and use the combination processing as an expansion method for subsequent expansion processing, wherein the center value refers to the median in the preset similarity range; For the training set, the step size during cutting is set to 32. The test set does not require sample expansion, so the step size is set to 155. Use cv2.imread to read the image, slide on the original image with the set step size, extract the image and label blocks, use the math.ceil() function to set the processed image width and height, and generate a new file name. Use cv2.imwrite to save the image, and save all the images in .png format.

2. According to claim 1, a method for detecting corn ear damage based on deep learning and embedded system, characterized in that: The preprocessing includes: marking the image data to generate a marking result, generating a mask image based on the marking result, and performing visual marking and grayscale processing on the mask image.

3. According to claim 1, a method for detecting corn ear damage based on deep learning and embedded system, characterized in that: The specific process of calculating the proportion of the damaged part of the preprocessed image data is as follows: The area of ​​the background part, the area of ​​the intact corn ear part and the area of ​​the damaged corn ear part are determined, and the proportion of the area of ​​the damaged corn ear part in the area excluding the area of ​​the background part is determined.

4. The method for detecting corn ear damage based on deep learning and embedded system according to claim 1, characterized in that: The process of processing the image data through the pre-trained deep learning model to obtain the injury classification result is specifically as follows: The image data is subjected to image segmentation processing and image enhancement processing to obtain optimized image data, and the optimized image data is subjected to pixel-level classification through the pre-trained deep learning model to obtain a damage classification result, wherein the damage classification result includes at least one of an undamaged ear area, a damaged ear area, a corn husk area, and a background area.

5. The method for detecting corn ear damage based on deep learning and embedded system according to claim 1, characterized in that: The pre-trained deep learning model is constructed by calling the UNet model in segmentation_models_pytorch and using efficiennet-b3 as the encoder in combination with a deep classifiable convolutional layer.

6. The method for detecting corn ear damage based on deep learning and embedded system according to claim 1, characterized in that: Also includes: Based on the calculation result of the proportion and combined with the manual label, the evaluation index of the image data is calculated, and the evaluation index includes: at least one of average intersection-over-union ratio, overall classification accuracy, average classification accuracy, classification accuracy of each category, false alarm rate and missed alarm rate.

7. A corn ear damage detection system based on deep learning and embedded system, using the corn ear damage detection method based on deep learning and embedded system as claimed in claim 1, characterized in that: The system includes: The acquisition module is used to: acquire the image data in the ear collection box in real time when the harvester is working, and process the image data through a pre-trained deep learning model to obtain a damage classification result; The detection module is used to: pre-process each image data in the damage classification result, calculate the proportion of the damaged part of the pre-processed image data, and save the calculation result of the proportion.

8. An electronic device, characterized in that: The electronic device comprises a processor, the processor is coupled to a memory, at least one computer program is stored in the memory, and the at least one computer program is loaded and executed by the processor so that the electronic device implements the method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores at least one computer program, and the at least one computer program is loaded and executed by a processor so that a computer implements the method according to any one of claims 1 to 6.

10. A harvester, characterized in that: It comprises a deep learning and embedded corn ear damage detection system as described in claim 7.

Images