Jujube tree disease and insect pest monitoring and identifying method and system based on remote sensing technology

Abstract

The invention relates to the technical field of disease and pest monitoring and recognition, in particular to a jujube tree disease and pest monitoring and recognition method and system based on the remote sensing technology, and the method comprises the steps: collecting a jujube tree vegetation image, and obtaining a first feature map after the neural network downsampling; obtaining a multi-scale feature map of the first feature map, and recording the multi-scale feature map as each second feature map; a self-adaptive fine-grained attention module is added in the neural network and is used for carrying out weighted fusion on the second feature maps, specifically, superpixel segmentation is carried out on all the second feature maps, and the insect pest possibility of all superpixel blocks is obtained; disease spot areas in the second feature maps are obtained; determining the size of a local window divided by each second feature map, and adopting a multi-head self-attention mechanism to obtain the weight of each second feature map; and carrying out weighted fusion on all the second feature maps, and carrying out multi-stage feature extraction of a ResNet50 network structure to obtain a pest and disease monitoring result of the jujube tree. Therefore, the accuracy of monitoring and identifying jujube tree diseases and insect pests is improved.

Description

Technical Field

[0001] This application relates to the technical field of pest and disease monitoring and identification, and specifically to a method and system for monitoring and identifying jujube tree pests and diseases based on remote sensing technology. Background Art

[0002] In the process of the development of the jujube tree planting industry, pests and diseases will cause a decline in the quality and quantity of jujubes. Monitoring of jujube tree pests and diseases is one of the key measures to ensure the healthy growth of jujube trees, increase yields and improve quality. The monitoring of pests and diseases can timely detect the pest and disease problems existing in jujube trees in the early stage and carry out prevention and control, which is a very important part of the jujube tree planting industry. Remote sensing technology can cover a wide geographical area and can be monitored in real time through platforms such as satellites and drones. This enables the occurrence of pests and diseases to be detected quickly and comprehensively in a large-scale jujube tree planting area, without being restricted by human operations. Therefore, remote sensing technology is widely used in the field of jujube tree pest and disease monitoring.

[0003] After obtaining the jujube tree vegetation image by using remote sensing technology, a neural network is usually used to identify the pest and disease areas in the jujube tree vegetation image. The existing technology has a low accuracy in identifying fine-grained disease spot areas in the jujube tree vegetation image. And when the neural network performs multi-scale feature map fusion, the feature map is often divided into local windows of a fixed size, and the self-attention mechanism is used to divide weights for feature maps of different scales. However, the local windows of a fixed size may not be able to adapt to the scales or complexities of different features, resulting in the loss of information or the inability to capture sufficient context information, especially when dealing with high-resolution features, reducing the suitability of feature map weight setting, thereby affecting the accuracy of neural network feature extraction and the sensitivity of early jujube tree pest and disease identification, leading to the problem of low accuracy in jujube tree pest and disease monitoring and identification. Summary of the Invention

[0004] In order to solve the above technical problems, the purpose of this application is to provide a method and system for monitoring and identifying jujube tree pests and diseases based on remote sensing technology. The specific technical solutions adopted are as follows: In the first aspect, an embodiment of this application provides a method for monitoring and identifying jujube tree pests and diseases based on remote sensing technology. The method includes the following steps: Collect jujube tree vegetation images, input the jujube tree vegetation images into a neural network, and obtain a first feature map after downsampling; use convolutional layers with multiple different convolutional kernel sizes to obtain multi-scale feature maps of the first feature map, denoted as each second feature map; wherein, the neural network is a ResNet50 network structure; Add an adaptive fine-grained attention module in the neural network for weighted fusion of the second feature maps, specifically including: Perform superpixel segmentation on each second feature map, analyze the grayscale differences between each superpixel block and its adjacent superpixel blocks in each second feature map, and obtain the pest infestation possibility of each superpixel block; Based on the pest infestation possibility, screen the target superpixel blocks in each second feature map; based on the grayscale differences between the target superpixel blocks and their neighboring superpixel blocks, select the neighboring superpixel blocks of the target superpixel blocks to merge with the target superpixel blocks, and obtain the diseased spot regions in each second feature map according to the morphological features of the merged regions; Using the sizes of the diseased spot regions in each second feature map, determine the sizes of the local windows for dividing each second feature map. After dividing each second feature map into local windows, adopt a multi-head self-attention mechanism to obtain the weights of each second feature map; Perform weighted fusion on all second feature maps, and through multi-stage feature extraction of the ResNet50 network structure, obtain the output of the neural network and get the pest and disease monitoring results of the jujube tree.

[0005] In one embodiment, the determination of the pest infestation possibility of each superpixel block includes: For each second feature map, calculate the grayscale mean value of all pixel points within each superpixel block, calculate the sum value of the differences between the grayscale mean value of each superpixel block and the grayscale mean values of all its adjacent superpixel blocks, and determine the pest infestation possibility of each superpixel block through the proportion of the sum value of each superpixel block in the cumulative sum of the sum values of all superpixel blocks in its second feature map.

[0006] In one embodiment, the screening of the target superpixel blocks in each second feature map based on the pest infestation possibility includes: Obtain the segmentation threshold of the pest infestation possibility of all superpixel blocks in each second feature map, and take the superpixel blocks with pest infestation possibility greater than the segmentation threshold as the target superpixel blocks.

[0007] In one embodiment, the selection of the neighboring superpixel blocks of the target superpixel blocks to merge with the target superpixel blocks includes: For any target superpixel block, if the grayscale mean value of its neighboring superpixel block is greater than or equal to the grayscale mean value of the any target superpixel block, then merge its neighboring superpixel block with the any target superpixel block.

[0008] In one embodiment, the obtaining of the diseased spot regions in each second feature map includes: Obtain the fitting circles of the merged regions in each second feature map, calculate the proportion of the overlapping area between the merged regions and the fitting circles, and determine the diseased spot regions in each second feature map based on the overlapping area proportion.

[0009] In one embodiment, the lesion area in each second feature map is a merged area where the overlapping area ratio is greater than a preset first threshold.

[0010] In one embodiment, the determining the size of the local window divided for each second feature map includes: For each second feature map, obtain the minimum bounding rectangle of each lesion area therein, and use the length and width of the minimum bounding rectangle with the largest area as the size of the divided local window.

[0011] In one embodiment, the determination of the weights of the second feature maps includes: Use the multi-head self-attention mechanism to obtain the attention matrix of each second feature map, and the weight of each second feature map is the mean value of all elements in the attention matrix.

[0012] In one embodiment, the obtaining the pest and disease monitoring result of the jujube tree includes: The output of the neural network includes the coordinates of the predicted bounding boxes of the pest and disease areas in the jujube tree vegetation image, and determine the proportion of the union area of all the predicted bounding boxes of the pest and disease areas in the jujube tree vegetation image in the jujube tree vegetation image; When the proportion is greater than a preset second threshold, it is determined that there is a pest and disease area in the corresponding area of the jujube tree vegetation image, and pest and disease treatment needs to be carried out on the jujube tree. Otherwise, pest and disease treatment does not need to be carried out on the jujube tree.

[0013] In a second aspect, an embodiment of the present application further provides a jujube tree pest and disease monitoring and identification system based on remote sensing technology, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, the steps of the method described in any one of the above are implemented.

[0014] The present application has at least the following beneficial effects: The present application improves the adaptability to the diversity of sizes of pests and diseases in jujube trees through multi-branch feature extraction with different convolution kernel sizes, and simultaneously captures the pathological features of different scales such as local yellowing, blackening, and lesions of jujube leaves, avoiding the problem of insufficient coverage of complex symptoms by a single receptive field; further, the present application adds an adaptive fine-grained attention module to the neural network to perform weighted fusion on the second feature map, perform superpixel segmentation on each second feature map, analyze the grayscale difference between each superpixel block and its adjacent superpixel block in each second feature map, and obtain the possibility of pests in each superpixel block; the refined recognition of the edge of the lesion is enhanced. The recognition ability is improved to accurately locate the early lesions of jujube tree leaves, suppress the interference of background noise, and improve the recognition ability of fine-grained lesions on leaves; then, the target superpixel blocks in each second feature map are screened based on the possibility of the pest; based on the grayscale difference between the target superpixel block and its neighboring superpixel blocks, the neighboring superpixel blocks of the target superpixel block are selected to merge with the target superpixel block, and the lesion area in each second feature map is obtained according to the morphological characteristics of the merged area; the accuracy and reliability of lesion area recognition are improved; by adding the shape characteristics of the early lesion area of ​​the leaf, it is helpful to improve the recognition and timeliness; further, the size of the lesion area in each second feature map is used to determine the size of the local window divided by each second feature map. After each second feature map is divided into a local window, a multi-head self-attention mechanism is used to obtain the weight of each second feature map; the local window size is dynamically adjusted based on the size of the lesion area, which improves the ability of multi-head self-attention to capture long-range correlations of lesion features, optimizes the rationality of feature weight allocation, and helps to improve the recognition sensitivity of the neural network model in sparse lesion scenarios; finally, all second feature maps are weighted fused and extracted through the multi-stage feature extraction of the ResNet50 network structure. , the output of the neural network is obtained, and the pest and disease monitoring results of the jujube tree are obtained, which enhances the robustness of the neural network model to the complexity of the jujube tree growth environment. The high-level feature information is captured by the deep network, which effectively suppresses the negative impact of factors such as illumination changes and branch and leaf occlusion on the detection results. In addition, the grayscale difference analysis mechanism of the adaptive fine-grained attention module improves the early warning ability of latent pests and diseases of jujube trees. By identifying the weak grayscale gradient features between superpixel blocks, an early warning can be issued before the symptoms visible to the naked eye appear, which buys critical time for grasping the window period for jujube tree prevention and control, and improves the accuracy of jujube tree pest and disease monitoring and identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present application or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0016] Figure 1 The flowchart of the steps of the jujube tree pest and disease monitoring and identification method based on remote sensing technology provided by an embodiment of the present application; Figure 2 The functional flowchart of the adaptive fine-grained attention module; Figure 3 The weighted fusion relationship diagram of the second feature maps. Detailed implementation manners

[0017] In order to further elaborate on the technical means and effects adopted by the present application to achieve the intended invention purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation manners, structures, features and effects of the jujube tree pest and disease monitoring and identification method and system based on remote sensing technology proposed according to the present application. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which this application belongs.

[0019] The following specifically describes the specific solutions of the jujube tree pest and disease monitoring and identification method and system provided by the present application with reference to the accompanying drawings.

[0020] Please refer to Figure 1 , which shows the flowchart of the steps of the jujube tree pest and disease monitoring and identification method based on remote sensing technology provided by an embodiment of the present application. The method includes the following steps: S1. Collect jujube tree vegetation images, input the jujube tree vegetation images into a neural network, and obtain a first feature map after downsampling; use convolutional layers with multiple different convolutional kernel sizes to obtain multi-scale feature maps of the first feature map, denoted as respective second feature maps.

[0021] In this embodiment, a high-resolution RGB camera is mounted on a drone to capture jujube tree vegetation images. The jujube tree vegetation images are high-resolution images containing jujube tree leaves, fruits and trunks, and can present the characteristics of healthy plants and pest-infected plants.

[0022] During the growth process of jujube trees, black spot disease on the jujube tree leaves causes black lesions to appear on the leaves. The death of part of the leaf tissue indicates the disappearance of chlorophyll, and the characteristic of yellowing of the leaves appears around the lesions. Therefore, pest and disease detection and identification can be carried out according to the different pixel values of the diseased area and the healthy area in the jujube tree vegetation images.

[0023] Since the lesion sizes vary at different stages of jujube tree diseases, fine-grained lesion features exist in some of the large number of jujube tree vegetation images obtained, making it difficult to identify pests and diseases in the early stage.

[0024] For pest and disease monitoring and identification based on jujube tree vegetation images, in this embodiment, the jujube tree vegetation images are first preprocessed. Specifically, image enhancement technology is used to enhance the jujube tree vegetation images to solve the problem of affecting the clarity of the captured images under insufficient lighting conditions. In this embodiment, the Retinex image enhancement technology is adopted, which is a prior art and the specific process will not be elaborated. Implementers can choose other existing feasible image enhancement algorithms by themselves, and this embodiment does not limit this.

[0025] The enhanced jujube tree vegetation images are normalized to obtain the preprocessed jujube tree vegetation images, so that the image pixel values are normalized between [0, 1], which is convenient for better convergence in the subsequent neural network training process. In this embodiment, the maximum-minimum normalization method is adopted, and implementers can choose other existing feasible normalization methods by themselves.

[0026] Furthermore, in this embodiment, the preprocessed jujube tree vegetation images are manually labeled for pest and disease areas to obtain label data. Among them, the healthy vegetation areas in the jujube tree vegetation images are marked as 0, and the pest and disease vegetation areas are marked as 1.

[0027] In this embodiment, the preprocessed jujube tree vegetation images and the label data are input into the neural network model for training, and the pest and disease area features in the jujube tree vegetation images are extracted through a multi-scale fine-grained neural network. Since when the neural network extracts features, in the pixel grids of the same scale, the lesion areas of different regions of the jujube tree vegetation images are different, the neural network should allocate different weights to the multi-scale feature maps according to the image features to fuse the multi-scale information when capturing different scale information, so as to capture the early fine-grained pest and disease area features.

[0028] In this embodiment, the network structure of the neural network model is an encoder and a fully connected layer structure. The encoder specifically uses ResNet50 as the backbone network architecture, and the optimizer of the neural network model uses the Adam optimizer. The loss function of the neural network model uses the GIoU loss and the binary cross-entropy loss. The output of the neural network model is a structured tensor, specifically including the confidence, the coordinates of the predicted bounding box of the pest and disease area, and the classification probability.

[0029] As can be seen from the ResNet50 network structure, it is a feature extraction of a downsampling layer and 4-stage residual blocks, and the 4 stages are respectively denoted as , where the network structure of ResNet50 is a prior art and the specific content will not be elaborated.

[0030] The feature map obtained after the downsampling layer processes the jujube tree vegetation image input into the neural network model is denoted as the first feature map and represented as . The downsampling layer performs preliminary feature extraction on the jujube tree vegetation image to reduce the computational burden of subsequent layers.

[0031] Furthermore, the neural network model uses convolutional layers with different convolutional kernel sizes to obtain multi-scale feature maps of the first feature map . In this embodiment, the convolutional kernel sizes used are 1×1, 3×3, and 5×5 respectively, and the corresponding feature maps obtained are represented as , denoted as each second feature map. The multi-scale feature map contains feature information of different scales. In the case of the presence of fine-grained lesion features in the jujube tree vegetation image, a convolutional kernel with a smaller receptive field can focus on capturing the feature information of the fine-grained lesions.

[0032] S2. An adaptive fine-grained attention module is added to the neural network to perform weighted fusion on the second feature maps. Among them, the functional flowchart of the adaptive fine-grained attention module is as Figure 2 shown. The adaptive fine-grained attention module specifically includes: (1) Perform superpixel segmentation on each second feature map, analyze the gray-scale differences between each superpixel block and its adjacent superpixel blocks in each second feature map, and obtain the pest infestation possibility of each superpixel block.

[0033] The weighted fusion method of traditional multi-scale feature maps directly weights the feature maps, and the feature information of different scales is not distinguished, which will lead to the loss of fine-grained feature information and is not conducive to the detection and recognition of fine-grained lesion areas in jujube tree vegetation images. To better detect and recognize the fine-grained lesion areas in jujube tree vegetation images, this embodiment constructs an adaptive fine-grained attention module and inputs it into the adaptive fine-grained attention module for processing.

[0034] The adaptive fine-grained attention module can assign different weights to each second feature map for fusion according to the image feature information captured at different scales. A higher weight is assigned to the feature information of the fine-grained lesion area captured under the receptive field of the small-scale convolutional kernel, and the weight of the feature map containing a large number of non-lesion areas captured under the receptive field of the large-scale convolutional kernel in this area will be relatively reduced, so as to increase the attention to fine-grained features in the multi-scale feature map fusion.

[0035] The adaptive fine-grained attention module specifically includes the following content: First, for Perform position embedding using convolutional layers with a convolution kernel size of 1×1 respectively, and then segment each second feature map using the SLIC superpixel segmentation algorithm. Assume that in any second feature map containing small lesions, the lesions themselves appear black and the surrounding leaf tissue shows a yellow feature due to the disappearance of chlorophyll. Then, among the G superpixel blocks after segmentation, the th superpixel block contains lesion features, so the overall gray value of the superpixel block will be relatively low, and the gray values of the superpixel blocks around the superpixel block will be higher than the gray value of the superpixel block . In this embodiment, G = 100 is set, and the implementer can set it according to the actual situation. The SLIC superpixel segmentation algorithm is a well-known existing technology, and the specific process will not be elaborated.

[0036] For each second feature map, calculate the average gray value of all pixel points within each superpixel block as the gray feature value of each superpixel block. Evaluate the possibility of the presence of black spots within the th superpixel block based on the difference between the gray feature value of the th superpixel block and the gray feature values of all its adjacent superpixel blocks, denoted as the pest infestation possibility , and the specific expression is: ; where, is the pest infestation possibility of the gth superpixel block in each second feature map, is the gray feature value of the gth superpixel block in each second feature map, is the gray feature value of the zth adjacent superpixel block of the gth superpixel block in each second feature map, G is the number of superpixel blocks in each second feature map, is the number of superpixel blocks adjacent to the gth superpixel block in each second feature map.

[0037] It should be noted that in this embodiment, the concept of adjacent means that there is a common boundary between two superpixel blocks, that is, an adjacent superpixel block refers to a superpixel block that has a common boundary with the gth superpixel block; the difference represents the degree of difference between two variables, and specifically can be calculated in ways such as difference, absolute value of difference, square of difference, ratio, etc. This embodiment does not limit this.

[0038] (2), Screen the target superpixel blocks in each second feature map based on the pest infestation possibility; based on the gray difference between the target superpixel block and its neighboring superpixel blocks, select the neighboring superpixel blocks of the target superpixel block to merge with the target superpixel block, and obtain the lesion regions in each second feature map according to the morphological features of the merged region.

[0039] Secondly, for each second feature map, calculate the pest infestation probability of all superpixel blocks therein, and automatically obtain the segmentation threshold using the Otsu thresholding algorithm. Superpixel blocks with a pest infestation probability greater than the segmentation threshold are used as target superpixel blocks. The Otsu thresholding algorithm is a prior art, and its specific process will not be elaborated here.

[0040] For each target superpixel block, merge the superpixel blocks whose gray feature values in all adjacent superpixel blocks are greater than or equal to the gray feature value of the target superpixel block with the target superpixel block. Since this application aims to improve the detection accuracy of fine-grained pest and disease areas, that is, the early pest and disease areas are small and difficult to detect, and the early pest and disease areas generally show a circular shape. Therefore, in this embodiment, superpixel blocks containing disease spots are further screened according to whether the merged result presents an approximately circular shape.

[0041] For the merged superpixel block region, in this embodiment, the least squares circle fitting method is used to obtain the fitting circle of the merged region, and calculate the ratio of the overlapping area between the merged region and the fitting circle , if is greater than a preset first threshold, then it is determined that the merged region is a disease spot region; otherwise, it is determined that the merged region is a normal region. Least squares circle fitting is a well-known prior art, and its specific process will not be elaborated here. In this embodiment, the first threshold is set to 0.7, and the implementer can set it according to the situation.

[0042] (3) Use the size of the disease spot regions in each second feature map to determine the size of the local windows for dividing each second feature map. After dividing each second feature map into local windows, a multi-head self-attention mechanism is used to obtain the weights of each second feature map.

[0043] For all the disease spot regions screened out in each second feature map, obtain their minimum bounding matrices, and use the length and width of the matrix with the largest area among all the minimum bounding matrices in each second feature map as the size of the local windows that need to be divided for the corresponding second feature map.

[0044] For each second feature map after local window division, in each local window, use the multi-head self-attention mechanism to calculate the self-attention matrix of the second feature map to which it belongs respectively. Adopting the local block strategy of the second feature map can solve the problem that the traditional Softmax function lacks local bias in the global calculation of the entire second feature map, resulting in abnormal attention weights in some feature regions during the training process of the neural network model. Among them, the multi-head self-attention mechanism is a well-known prior art, and its specific process will not be elaborated in detail. The calculation formula of the self-attention mechanism can be expressed as: ; where represents the query vector, key vector, and value vector after the feature map is mapped by the linear layer, represents matrix transpose, is the dimension of the key vector, is the normalization function.

[0045] Then, the weights of each second feature map are calculated using the attention matrix of each second feature map. The weights can be calculated by the formula: ; where represents the number of rows and columns of the attention matrix, is the th row and th column element in the attention matrix.

[0046] After calculating the weights, the weights of are respectively . Therefore, the neural network model of this embodiment can adaptively and dynamically adjust the weights of each second feature map according to the feature information of the jujube tree vegetation image.

[0047] (4) Weighted fusion is performed on all second feature maps.

[0048] Through weighted fusion , the feature map is obtained, specifically:

[0049] The adaptive multi-scale feature map weight adjustment increases the weight of the local fine-grained feature map when there are a large number of small diseased areas in the jujube tree vegetation image, improves the attention of the neural network model to the fine-grained features, and reduces the influence of irrelevant background features. The weighted fusion relationship diagram of the second feature map is as shown in Figure 3 .

[0050] S3. Through multi-stage feature extraction of the ResNet50 network structure, the output of the neural network is obtained, and the pest and disease monitoring result of the jujube tree is obtained.

[0051] The feature map after weighted fusion enters the stage of the ResNet50 network structure for feature extraction, and the new feature map is obtained. Repeat the operations of the above second feature map to obtain the fused feature map .

[0052] Correspondingly, in the stage of the ResNet50 network structure, the above process of obtaining the feature map is sequentially repeated to obtain the final feature map Since the jujube tree vegetation images obtained by drone photography contain a large number of healthy leaves, leaves with pest and disease spots, and the structure, texture, and gray-scale features of the tree trunks, they constitute a scene with complex feature content. The neural network model in this embodiment uses multi-stage feature extraction of the ResNet50 network structure to obtain high-level features in the jujube tree vegetation images, and encodes the features in the complex scene to complete the fine-grained detection and recognition of pest and disease areas.

[0053] Feature map One-dimensional feature vectors are obtained through the global average pooling layer and the Flatten operation, and finally the detection results of the neural network model are output through the fully connected layer. The detection results of the neural network model and the label data are used to calculate the loss through the GIoU loss and the binary cross-entropy loss, and the parameters of the neural network model are optimized to complete the training of the neural network model. The training of the neural network is a well-known technology, and the specific process will not be elaborated here.

[0054] Finally, the output result of the trained neural network can determine whether there are pest and disease areas in the jujube tree vegetation image and outline the pest and disease areas. According to the coordinates of the bounding boxes of the outlined pest and disease areas, the ratio of the union area of all the bounding boxes in the jujube tree vegetation image to the overall image area of the jujube tree vegetation image is calculated to obtain the proportion of the pest and disease areas in the jujube tree vegetation image. For pest and disease monitoring and identification of jujube trees. The proportion The larger it is, the more pest and disease areas there are in the jujube tree vegetation image, and the more severely the jujube tree is disturbed by pests and diseases.

[0055] If the proportion is greater than the preset second threshold, it is determined that there are pest and disease areas in the corresponding area of the jujube tree vegetation image, and pest and disease treatment needs to be carried out on the jujube tree. Otherwise, pest and disease treatment does not need to be carried out on the jujube tree. In this embodiment, the second threshold is set to 0.3, and the implementer can set it according to the actual situation, and this embodiment does not limit it.

[0056] Based on the same inventive concept as the above method, the embodiment of the present application also provides a jujube tree pest and disease monitoring and identification system based on remote sensing technology, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any one of the above methods for jujube tree pest and disease monitoring and identification based on remote sensing technology.

[0057] It should be noted that: the above sequence of embodiments of the present application is only for description and does not represent the superiority or inferiority of the embodiments. And the above description of specific embodiments of this specification is made. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0058] Each embodiment in this specification is described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other, and the differences between each embodiment and other embodiments are emphasized.

[0059] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for monitoring and identifying jujube tree diseases and pests based on remote sensing technology, characterized in that, The method includes the following steps: Collect jujube tree vegetation images, input the jujube tree vegetation images into a neural network, and obtain a first feature map after downsampling; use convolutional layers with multiple different convolutional kernel sizes to obtain multi-scale feature maps of the first feature map, denoted as respective second feature maps; wherein, the neural network is a ResNet50 network structure; Add an adaptive fine-grained attention module to the neural network for weighted fusion of the second feature maps, specifically including: Perform superpixel segmentation on each second feature map, analyze the gray-scale differences between each superpixel block and its adjacent superpixel blocks in each second feature map, and obtain the pest infestation possibility of each superpixel block; Screen target superpixel blocks in each second feature map based on the pest infestation possibility; based on the gray-scale differences between the target superpixel blocks and their neighboring superpixel blocks, select neighboring superpixel blocks of the target superpixel blocks to merge with the target superpixel blocks, and obtain the diseased spot regions in each second feature map according to the morphological features of the merged regions; Use the sizes of the diseased spot regions in each second feature map to determine the sizes of the local windows for dividing each second feature map. After dividing each second feature map into local windows, use the multi-head self-attention mechanism to obtain the weights of each second feature map; Perform weighted fusion on all the second feature maps, and through multi-stage feature extraction of the ResNet50 network structure, obtain the output of the neural network to get the pest and disease monitoring result of the jujube tree.

2. The method for monitoring and identifying jujube tree diseases and pests based on remote sensing technology according to claim 1, characterized in that, The determination of the pest infestation possibility of each superpixel block includes: For each second feature map, calculate the gray-scale mean value of all pixel points within each superpixel block, calculate the sum value of the differences between the gray-scale mean value of each superpixel block and the gray-scale mean values of all its adjacent superpixel blocks, and determine the pest infestation possibility of each superpixel block through the proportion of the sum value of each superpixel block in the cumulative sum of the sum values of all superpixel blocks in its second feature map.

3. The method for monitoring and identifying jujube tree diseases and pests based on remote sensing technology according to claim 1, characterized in that, The screening of the target superpixel blocks in each second feature map based on the pest infestation possibility includes: Obtain the segmentation threshold of the pest infestation possibility of all superpixel blocks in each second feature map, and use the superpixel blocks with pest infestation possibility greater than the segmentation threshold as the target superpixel blocks.

4. The method for monitoring and identifying jujube tree diseases and pests based on remote sensing technology according to claim 2, wherein, The selection of neighboring superpixel blocks of the target superpixel blocks to merge with the target superpixel blocks includes: For any target superpixel block, if the gray-scale mean value of its neighboring superpixel block is greater than or equal to the gray-scale mean value of the any target superpixel block, then merge its neighboring superpixel block with the any target superpixel block.

5. The method for monitoring and identifying jujube tree diseases and pests based on remote sensing technology according to claim 1, characterized in that, The obtaining of the diseased spot regions in each second feature map includes: Obtain the fitting circle of the merged region in each second feature map, calculate the proportion of the overlapping area between the merged region and the fitting circle, and based on the proportion of the overlapping area, determine the diseased spot regions in each second feature map.

6. The method for monitoring and identifying jujube tree diseases and pests based on remote sensing technology according to claim 5, characterized in that, The diseased spot regions in each second feature map are the merged regions with the proportion of the overlapping area greater than a preset first threshold.

7. The method for monitoring and identifying jujube tree diseases and pests based on remote sensing technology according to claim 1, wherein The determination of the sizes of the local windows for dividing each second feature map includes: For each second feature map, obtain the minimum circumscribed matrix of each diseased spot region therein, and use the length and width of the minimum circumscribed matrix with the largest area as the sizes of the local windows for dividing.

8. The method for monitoring and identifying jujube tree diseases and pests based on remote sensing technology according to claim 1, wherein, The determination of the weights of each second feature map includes: The multi-head self-attention mechanism is adopted to obtain the attention matrix of each second feature map, and the weight of each second feature map is the mean value of all elements in the attention matrix.

9. The method for monitoring and identifying jujube tree diseases and pests based on remote sensing technology according to claim 1, characterized in that, The obtaining of the pest and disease monitoring results of the jujube tree includes: The output of the neural network includes the coordinates of the predicted bounding boxes of the pest and disease areas in the jujube tree vegetation image, and the proportion of the union area of all the predicted bounding boxes of the pest and disease areas in the jujube tree vegetation image in the jujube tree vegetation image is determined; When the proportion is greater than the preset second threshold, it is determined that there is a pest and disease area in the corresponding area of the jujube tree vegetation image, and pest and disease treatment needs to be carried out on the jujube tree. Otherwise, pest and disease treatment does not need to be carried out on the jujube tree.

10. A jujube tree pest and disease monitoring and identification system based on remote sensing technology, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, the steps of the method according to any one of claims 1-9 are implemented.

Images