An insect target detection method, system and device based on deep learning

Abstract

An insect target detection method, system and device based on deep learning, which belongs to the field of artificial intelligence. The present invention solves the problems of scarce image data in the existing insect target detection field and high costs for obtaining and annotating insect image data. The technical solution adopted by the present invention is as follows: obtaining a single-target insect specimen image dataset; obtaining a natural environment background image dataset; constructing an insect target detection dataset by using the single-target insect specimen image dataset and the natural environment background image dataset; training an insect target detection model by using the constructed dataset; continuously expanding the dataset according to the model feedback, and using the expanded dataset to continue training the model, repeating this process until the model performance no longer improves. The method of the present invention can be applied to the field of insect target detection.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence, and specifically relates to an insect target detection method, system and equipment based on deep learning. Background Art

[0002] At present, agriculture and forestry suffer huge economic losses every year due to insect disasters. Therefore, insect disasters are a huge threat to agriculture and forestry. Severe insect disasters may cause significant losses to agriculture and forestry and even destroy the local ecological balance. Some insect disasters occur because the signs of insect disasters are not discovered early, and it is too late after the insect disaster has formed a scale. Traditional manual detection methods do not have the ability to detect species invasions and insect disasters, and some insects are difficult to detect, so it is easy to miss detection and it is difficult to make effective early warnings for insect disasters.

[0003] Insect target detection is a task that uses computer vision technology to identify and locate different species and locations in insect images. This task has important application value in the fields of biodiversity conservation, agricultural pest control, and ecosystem monitoring. However, insect target detection also faces some challenges, the most important of which is the scarcity of image data in the field of insect target detection, the high cost of acquiring wild insect images, and the high cost of annotation. The reasons for the high cost of acquiring wild insect images are as follows: first, the wild environment is complex and changeable, which affects the quality of insect images; second, there are many species of wild insects and they are widely distributed, requiring a large amount of collection equipment and manpower; third, the behavior of wild insects is difficult to control, which may lead to occlusion, overlap, blurring, etc. in the image. The reasons for the high annotation cost are as follows: first, insect target detection requires accurate bounding box and category annotation for each insect instance, which is a time-consuming and labor-intensive task; second, insect target detection requires professional entomological knowledge in order to correctly distinguish different insect species and subspecies; third, insect target detection requires a large amount of labeled data to ensure the generalization ability and robustness of the model. Therefore, how to reduce the cost of acquiring and labeling insect target detection data is a difficult problem that needs to be solved urgently in the field of insect target detection. Summary of the invention

[0004] The purpose of the present invention is to solve the problems of scarcity of image data in the existing insect target detection field and high cost of acquiring and labeling insect image data, and to propose an insect target detection method based on deep learning.

[0005] The technical solution adopted by the present invention to solve the above technical problems is:

[0006] A method for detecting insect targets based on deep learning, the method specifically comprising the following steps:

[0007] Step 1: Obtain a single-target insect specimen image dataset, and each single-target insect specimen image has a species label;

[0008] Each single-target insect specimen image in the data set is processed separately to obtain the insect main body image in each specimen image;

[0009] Step 2: Obtain a dataset of background images of insects’ living environment in the wild;

[0010] Step 3: construct an insect target detection dataset based on the images obtained in step 1 and step 2, and divide the constructed insect target detection dataset into a training set and a test set;

[0011] The specific process of step 3 is as follows:

[0012] Step 3.1, randomly selecting N insect main body images from the insect main body images obtained in step 1, and then preprocessing the selected images to obtain preprocessed images;

[0013] Step 3.2, randomly select a background image of the insect's wild living environment from the image data set obtained in step 2;

[0014] Step 3.3, inserting the image preprocessed in step 3.1 into the selected background image of the insect's wild living environment;

[0015] Step 3.4: label the position and category of the image obtained in step 3.3;

[0016] Step 3.5, repeat the process from step 3.1 to step 3.4 until the maximum number of iterations is reached, and then obtain the insect target detection dataset;

[0017] Step 3.6, randomly divide the obtained insect target detection dataset into two parts: a training set and a test set;

[0018] Step 4: construct an insect target detection model, and use the training set to train the constructed insect target detection model;

[0019] Step 5: Use the test set to test the insect target detection model trained in step 4, and expand the training set data according to the test results. After expansion, return to step 4 for training;

[0020] Until the detection performance of the insect target detection model on the test set no longer improves, a trained insect target detection model is obtained; and insect target detection is performed on the image to be detected using the trained insect target detection model.

[0021] Furthermore, each single target insect specimen image in the data set is processed respectively to obtain the insect main body image in each specimen image; the specific process is:

[0022] Step 1.1: For any single-target insect specimen image in the data set, use the Roberts operator, Prewitt operator, Sobel operator, Canny operator and Laplacian operator to perform edge detection on the single-target insect specimen image to obtain the edge detection image corresponding to each operator;

[0023] Step 1.2: Combine the edge-detected images corresponding to the operators in step 1.1 with the original single-target insect specimen image to obtain a combined image;

[0024] Step 1.3: Input the combined image in step 1.2 into the DeepLab-v3 semantic segmentation neural network, and output the contour positioning result of the insect through the DeepLab-v3 semantic segmentation neural network;

[0025] Step 1.4: removing the background of the single target insect specimen image according to the insect outline obtained in step 1.3, and obtaining an insect main body image containing only the insect main body;

[0026] Step 1.5: Repeat the process from step 1.1 to step 1.4, and process each single target insect specimen image in the data set respectively to obtain the insect main body image in each specimen image.

[0027] Furthermore, the selected image is preprocessed, and the preprocessing method includes random rotation and random scaling.

[0028] Furthermore, the insect target detection model is YOLO, Faster R-CNN or SSD.

[0029] Furthermore, the specific process of expanding the training set data according to the test results is as follows:

[0030] For samples that are misdetected by the insect target detection model on the test set, new samples are constructed and added to the training set based on the background image used when constructing the misdetected samples and the insect subject image inserted into the background image.

[0031] That is, the insect main body image used to construct the detection error sample is randomly rotated and randomly scaled, and then the randomly rotated and randomly scaled insect main body image is inserted into the background image used to construct the detection error sample to obtain a new sample image, and then the new sample image is added to the training set.

[0032] An insect target detection system based on deep learning, the system comprises an insect specimen image acquisition module, an insect specimen image processing module, a background image acquisition module, an insect target detection data set acquisition module and an insect target detection module, wherein:

[0033] The insect specimen image acquisition module is used to acquire a single-target insect specimen image data set;

[0034] The insect specimen image processing module is used to process the images in the single-target insect specimen image data set to obtain the insect subject image in each specimen image;

[0035] The background image acquisition module is used to acquire a background image dataset of the insect's wild living environment;

[0036] The insect target detection data set acquisition module is used to construct a data set based on the images obtained by the insect specimen image processing module and the background image acquisition module;

[0037] The insect target detection module is trained based on the constructed data set. After the training is completed, the insect target detection module is used to perform insect target detection on the image to be detected.

[0038] A computer storage medium, wherein at least one instruction is stored in the storage medium, and the at least one instruction is loaded and executed by a processor to implement the insect target detection method based on deep learning.

[0039] A deep learning-based insect target detection device, the device comprising a processor and a memory, the memory storing at least one instruction, the at least one instruction being loaded and executed by the processor to implement the deep learning-based insect target detection method.

[0040] The beneficial effects of the present invention are:

[0041] The present invention can use single-target insect specimen images to construct a highly realistic high-quality insect target detection data set, solving the problem of scarcity of image data in the field of insect target detection. The insect specimen images used can be taken in an indoor environment, avoiding field operations. The insect specimen images can use existing insect collections and databases, reducing the investment in collection equipment and manpower. The insect specimen images can fix and unfold the insects, eliminating the problems of occlusion, overlap, blur and so on in the image. Therefore, compared with the field insect images, the insect specimen images are easy to obtain, reducing the cost of obtaining. In addition, the insect specimens naturally have the labels of their species, and do not need to be manually labeled again, avoiding the problem of high cost of insect image data labeling, and avoiding the problem of large amount of manpower and material resources required for obtaining and labeling field insect images. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A flowchart of an insect target detection method based on deep learning of the present invention;

[0043] Figure 2 This is a flow chart of the algorithm for obtaining insect subject images proposed by the present invention;

[0044] Figure 3 It is a schematic diagram of a single target insect specimen image;

[0045] Figure 4 is a schematic diagram of a close-up image of a natural environment;

[0046] Figure 5 The image after inserting multiple target insect images;

[0047] Figure 6 This is a block diagram of an insect target detection system based on deep learning of the present invention. DETAILED DESCRIPTION

[0048] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solution in the implementation of the present invention will be clearly and completely described below in conjunction with the drawings in the implementation of the present invention. Obviously, the described implementation is only a part of the implementation of the present invention, not all of the implementations. Based on the implementation of the present invention, all other implementations obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0049] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other. The terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the terms used in this way can be interchanged where appropriate, so that the embodiments of the present invention are described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0050] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0051] Specific implementation method 1. Combination Figure 1 and Figure 2 The present embodiment describes a method for detecting insect targets based on deep learning, and the method specifically comprises the following steps:

[0052] Step 1: Obtain a dataset of single-target insect specimen images (i.e., each image contains only one insect target), and each single-target insect specimen image has a species label;

[0053] You can collect single-target insect specimen images of different species, angles, and lighting conditions from the Internet or the laboratory, such as ladybugs and dragonflies, and assign corresponding species labels to each image. Figure 3 Shown is a schematic diagram of a single target insect specimen image.

[0054] Each single-target insect specimen image in the data set is processed separately to obtain the insect main body image in each specimen image; the specific process is as follows:

[0055] Step 1.1: For any single-target insect specimen image in the data set, use the Roberts operator, Prewitt operator, Sobel operator, Canny operator and Laplacian operator to perform edge detection on the single-target insect specimen image to obtain the edge detection image corresponding to each operator;

[0056] In actual operation, OpenCV can be used for implementation;

[0057] Step 1.2: Combine the edge-detected images corresponding to the operators in step 1.1 with the original single-target insect specimen image to obtain a combined image;

[0058] Among them, the edge detection image corresponding to each operator is used as the feature of a channel of the combined image; in actual operation, it can be implemented using programming languages ​​such as Python;

[0059] Step 1.3: Input the combined image in step 1.2 into the DeepLab-v3 semantic segmentation neural network, and output the contour positioning result of the insect through the DeepLab-v3 semantic segmentation neural network;

[0060] In the process of acquiring image-level features in DeepLab-v3, a feature is extracted for each channel of the combined image input in 1.2, i.e., the original image and the result of each operator. These features include convolution features and spatial pyramid pooling features, etc. These features are used to obtain the insect outline coordinates through DeepLab-v3;

[0061] Step 1.4: Remove the background of the single target insect specimen image according to the insect outline obtained in step 1.3, and obtain an insect main body image containing only the insect main body; in actual operation, computer vision libraries such as OpenCV can be used to implement this;

[0062] Step 1.5: Repeat the process from step 1.1 to step 1.4, and process each single target insect specimen image in the data set respectively to obtain the insect main body image in each specimen image.

[0063] Step 2: Obtain a dataset of background images of insects’ living environment in the wild;

[0064] You can take natural environment background images such as grass, flowers, leaves, etc. in different seasons, locations, and weather conditions from the Internet or on-site. Figure 4 As shown, it is a schematic diagram of a close-up image of a natural environment;

[0065] Step 3: construct an insect target detection dataset based on the images obtained in step 1 and step 2, and divide the constructed insect target detection dataset into a training set and a test set;

[0066] The specific process of step 3 is as follows:

[0067] Step 3.1, randomly select N insect main body images from the insect main body images obtained in step 1, and then preprocess the selected images to obtain preprocessed images; the preprocessing method includes random rotation and random scaling. For example, a single target insect image can be randomly rotated 0-360 degrees and randomly scaled 0.5-2 times to increase the diversity and difficulty of the image.

[0068] In the present invention, the value of N can be 1, 2, 3, 4, 5 or 6, and the N images can be from the same category or different categories;

[0069] Step 3.2, randomly select a background image of the insect's wild living environment from the image data set obtained in step 2;

[0070] Step 3.3, inserting the image preprocessed in step 3.1 into the selected background image of the insect's wild living environment;

[0071] The insertion position is random, and multiple preprocessed images do not overlap; for example, a region can be randomly selected in the background image, and the preprocessed single target insect image can be pasted into the region, and then it is checked whether there is overlap with other inserted insect images. If there is overlap, a new region is selected until there is no overlap. Repeat this process until all preprocessed single target insect images are inserted into the background image. Figure 5 As shown, it is the image obtained by inserting the preprocessed image;

[0072] Step 3.4: label the position and category of the image obtained in step 3.3;

[0073] According to the position of the insect image inserted in the background image, the multi-target supervised insect image required for the target detection model training is generated. For each insect image inserted into the background image, its maximum horizontal coordinate in the background image is marked as x max , the maximum ordinate is y max , the minimum horizontal coordinate is x min , the minimum ordinate is y min . Based on this, the corresponding coordinates of the four vertices of the bounding box are (x max ,y max ),(x min ,y min ),(x min ,y max ),(x max ,y min ), the category label corresponding to the bounding box is the category of the original single target insect specimen image. The vertex coordinates of the bounding boxes corresponding to all insect images inserted into the background image and the insect category in the bounding box are the labels of the constructed insect target detection samples;

[0074] Step 3.5, repeat the process from step 3.1 to step 3.4 until the maximum number of iterations is reached, and then obtain the insect target detection dataset;

[0075] Step 3.6, randomly divide the obtained insect target detection dataset into two parts: a training set and a test set;

[0076] Step 4: Build an insect target detection model, where the insect target detection model is YOLO, Faster R-CNN or SSD. The model parameters and hyperparameters can be adjusted according to actual conditions.

[0077] Use the training set to train the constructed insect target detection model;

[0078] Step 5: Use the test set to test the insect target detection model trained in step 4, and expand the training set data according to the test results. After expansion, return to step 4 for training;

[0079] Until the detection performance of the insect target detection model on the test set no longer improves, a trained insect target detection model is obtained; and insect target detection is performed on the image to be detected using the trained insect target detection model.

[0080] For example, you can use indicators such as accuracy, recall, and F1 value to evaluate the performance of the target detection model on the test set, and set a threshold or an iteration number. When the performance indicator reaches the threshold or the iteration number reaches the upper limit, stop the training process.

[0081] The specific process of expanding the training set data according to the test results is as follows:

[0082] For samples that are misdetected by the insect target detection model on the test set, new samples are constructed and added to the training set based on the background image used when constructing the misdetected samples and the insect subject image inserted into the background image.

[0083] That is, the insect main body image used to construct the detection error sample is randomly rotated and randomly scaled, and then the randomly rotated and randomly scaled insect main body image is inserted into the background image used to construct the detection error sample to obtain a new sample image, and then the new sample image is added to the training set. In this way, the model's detection ability for this type of sample can be further improved.

[0084] By using the method of the present invention to extract the main insect image, more detailed insect images can be extracted, thereby improving the quality of the subsequently constructed data set. The present invention proposes a complete process for training an insect target detection model using a single-target insect image, including data set construction, model training, data set optimization, and model optimization. The main innovation lies in the data set construction method and the overall process of the solution. The present invention can alleviate the problem of data scarcity in the field of insect target detection, and can use single-target insect specimen images to construct a highly realistic high-quality insect target detection data set and use it to complete the insect target detection model training, thereby obtaining an efficient insect target detection model.

[0085] Specific implementation method 2: In this implementation method, a deep learning-based insect target detection system is described, and the system includes an insect specimen image acquisition module, an insect specimen image processing module, a background image acquisition module, an insect target detection data set acquisition module, and an insect target detection module, wherein:

[0086] The insect specimen image acquisition module is used to acquire a single-target insect specimen image data set;

[0087] The insect specimen image processing module is used to process the images in the single-target insect specimen image data set to obtain the insect subject image in each specimen image;

[0088] The background image acquisition module is used to acquire a background image dataset of the insect's wild living environment;

[0089] The insect target detection data set acquisition module is used to construct a data set based on the images obtained by the insect specimen image processing module and the background image acquisition module;

[0090] The insect target detection module is trained based on the constructed data set. After the training is completed, the insect target detection module is used to perform insect target detection on the image to be detected.

[0091] Specific implementation method three: This implementation method is a computer storage medium, in which at least one instruction is stored. The at least one instruction is loaded and executed by a processor to implement the insect target detection method based on deep learning.

[0092] It should be understood that the instructions include computer program products, software or computerized methods corresponding to any method described in the present invention; the instructions can be used to program a computer system, or other electronic device. Computer storage media may include readable media on which instructions are stored, which may include but are not limited to magnetic storage media, optical storage media; magneto-optical storage media include read-only memory ROM, random access memory RAM, erasable programmable memory (e.g., EPROM and EEPROM) and flash memory layers, or other types of media suitable for storing electronic instructions.

[0093] Specific implementation method 4: This implementation method is an insect target detection device based on deep learning, and the device includes a processor and a memory. It should be understood that including any device including a processor and a memory described in the present invention, the device may also include other units and modules that perform display, interaction, processing, control, etc. and other functions through signals or instructions;

[0094] At least one instruction is stored in the memory, and the at least one instruction is loaded and executed by the processor to implement the insect target detection method based on deep learning.

[0095] like Figure 6, is a block diagram of an electronic device according to the insect image target detection method of the present invention. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown in the present invention, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required in the present invention.

[0096] A specific implementation environment of the embodiment of the present invention is shown in Table 1:

[0097] Table 1

[0098] Configuration Configuration parameters operating system Windows 10 Memory 64GB Graphics RTX-3090Ti CPU Intel i9-9900KF Software Environment Pycharm; Pytorch 1.4 Implementation Language Python3

[0099] The above calculation examples of the present invention are only used to explain the calculation model and calculation process of the present invention in detail, and are not intended to limit the implementation methods of the present invention. For ordinary technicians in the relevant field, other different forms of changes or modifications can be made based on the above description. It is impossible to list all the implementation methods here. All obvious changes or modifications derived from the technical solution of the present invention are still within the scope of protection of the present invention.

Claims

1. An insect target detection method based on deep learning, characterized in that, the method specifically includes the following steps: Step 1: Obtain a single-target insect specimen image dataset, and each single-target insect specimen image has a species label; Process each single-target insect specimen image in the dataset to obtain the insect main body image in each specimen image; specifically: Step 1.1: For any single-target insect specimen image in the dataset, use the Roberts operator, Prewitt operator, Sobel operator, Canny operator, and Laplacian operator to perform edge detection on the single-target insect specimen image respectively, so as to obtain the edge-detected image corresponding to each operator; Step 1.2: Combine the edge-detected images corresponding to each operator in Step 1.1 with the single-target insect specimen image before edge detection to obtain a combined image; Step 1.3: Input the combined image in Step 1.2 into the DeepLab-v3 semantic segmentation neural network, and output the contour localization result of the insect through the DeepLab-v3 semantic segmentation neural network; Step 1.4: Remove the background for the single-target insect specimen image according to the insect contour obtained in Step 1.3, and obtain an insect main body image containing only the insect main body; Step 1.5: Repeat the process of Step 1.1 to Step 1.4, and process each single-target insect specimen image in the dataset respectively, so as to obtain the insect main body image in each specimen image; Step 2: Obtain an insect wild survival environment background image dataset; Step 3: Construct an insect target detection dataset based on the images obtained in Step 1 and Step 2, and divide the constructed insect target detection dataset into a training set and a test set; The specific process of Step 3 is: Step 3.1: Randomly select N insect main body images from the insect main body images obtained in Step 1, and then preprocess the selected images to obtain preprocessed images; Step 3.2: Randomly select an insect wild survival environment background image from the image dataset obtained in Step 2; Step 3.3: Insert the preprocessed image in Step 3.1 into the selected insect wild survival environment background image; Step 3.4: Perform position and category annotation on the image obtained in Step 3.3; Step 3.5: Repeat the process of Step 3.1 to Step 3.4 until the set maximum number of iterations is reached and then stop to obtain an insect target detection dataset; Step 3.6: Randomly divide the obtained insect target detection dataset into a training set and a test set; Step 4: Construct an insect target detection model, and use the training set to train the constructed insect target detection model; Step 5: Use the test set to test the insect target detection model trained in Step 4, and expand the training set data according to the test results, and then return to Step 4 for training; A trained insect target detection model is obtained until the detection performance of the insect target detection model on the test set no longer improves; the trained insect target detection model is used to detect insect targets in the image to be detected.

2. A method for insect target detection based on deep learning according to claim 1, wherein, the preprocessing of the selected images includes random rotation and random scaling.

3. A method for insect target detection based on deep learning according to claim 2, wherein, the process of expanding the training set data according to the test results is as follows: For the samples misdetected by the insect target detection model on the test set, new samples are constructed based on the background images used when constructing the misdetected samples and the insect main body images inserted into the background images, and added to the training set; that is, the insect main body images used when constructing the misdetected samples are randomly rotated and scaled, and then the randomly rotated and scaled insect main body images are inserted into the background images used when constructing the misdetected samples to obtain new sample images, and the new sample images are added to the training set.

4. An insect target detection system based on deep learning, wherein, the system includes an insect specimen image acquisition module, an insect specimen image processing module, a background image acquisition module, an insect target detection data set acquisition module and an insect target detection module, wherein: the insect specimen image acquisition module is used to acquire a single-target insect specimen image data set; the insect specimen image processing module is used to process the images in the single-target insect specimen image data set to obtain the insect main body images in each specimen image; The working process of the insect specimen image processing module is as follows: Step 1.1: For any single-target insect specimen image in the data set, use the Roberts operator, Prewitt operator, Sobel operator, Canny operator and Laplacian operator to perform edge detection on the single-target insect specimen image to obtain the edge-detected images corresponding to each operator; Step 1.2: Combine the edge-detected images corresponding to each operator in Step 1.1 with the single-target insect specimen image before edge detection to obtain a combined image; Step 1.3: Input the combined image in Step 1.2 into the DeepLab-v3 semantic segmentation neural network, and output the contour localization result of the insect through the DeepLab-v3 semantic segmentation neural network; Step 1.4: Remove the background from the single-target insect specimen image according to the insect contour obtained in Step 1.3 to obtain an insect main body image containing only the insect main body; Step 1.5: Repeat the process of Step 1.1 to Step 1.4 to process each single-target insect specimen image in the data set to obtain the insect main body images in each specimen image; the background image acquisition module is used to acquire an insect wild survival environment background image data set; The insect target detection dataset acquisition module is used to construct a dataset according to the images obtained by the insect specimen image processing module and the background image acquisition module; The working process of the insect target detection dataset acquisition module is as follows: Step 3.1: Randomly select N insect body images from the insect body images obtained in Step 1, and then preprocess the selected images to obtain the preprocessed images; Step 3.2: Randomly select an insect wild survival environment background image from the image dataset obtained in Step 2; Step 3.3: Insert the preprocessed images in Step 3.1 into the selected insect wild survival environment background image; Step 3.4: Perform position and category annotation on the images obtained in Step 3.3; Step 3.5: Repeat the process of Step 3.1 to Step 3.4 until the set maximum iteration number is reached and then stop to obtain the insect target detection dataset; Step 3.6: Randomly divide the obtained insect target detection dataset into two parts: a training set and a test set; The insect target detection module is trained based on the constructed dataset. After training is completed, the insect target detection module is used to perform insect target detection on the image to be detected.

5. A computer storage medium, characterized in that, at least one instruction is stored in the storage medium, and the at least one instruction is loaded and executed by a processor to implement a deep learning-based insect target detection method as described in any one of claims 1 to 3.

6. A deep learning-based insect target detection device, characterized in that, the device includes a processor and a memory, and at least one instruction is stored in the memory, and the at least one instruction is loaded and executed by the processor to implement a deep learning-based insect target detection method as described in any one of claims 1 to 3.

Images