Intelligent identification method and system of forest pests and diseases based on remote sensing data

Abstract

The present application relates to the field of pest and disease identification, and provides a method and system for intelligent identification of forest pests and diseases based on remote sensing data, which uses image analysis and processing technology based on deep learning to perform image preprocessing and remote sensing color and texture feature extraction on the acquired remote sensing images of the original forest area, so as to automatically judge whether there are pests and diseases based on the significant sparse interactive fusion features between the remote sensing color features and the remote sensing texture features of the forest area. Through remote sensing technology, large areas of forest areas can be covered to achieve real-time and comprehensive monitoring without waiting for manual inspection cycles. In addition, pests and diseases can be accurately identified through deep learning technology, so that measures can be taken quickly to achieve efficient, accurate and intelligent monitoring of pests and diseases in forest areas.

Description

Technical Field

[0001] The present application relates to the field of pest and disease identification, and more specifically, to a method and system for intelligently identifying forest pests and diseases based on remote sensing data. Background Art

[0002] Forest pests and diseases pose a huge threat to forest ecosystems, especially diseases such as pine wilt disease, which can cause trees to die quickly and spread quickly, making them difficult to control. Pine wilt disease is spread by vector insects (such as Monochamus alternatus). Once infected, trees usually wither and die within a few months, seriously affecting the ecological balance and economic value of the forest. Therefore, timely detection and control of pests and diseases is one of the key measures to protect forest resources, which can effectively prevent the spread of diseases and ensure the sustainable development of forests.

[0003] However, traditional pest and disease identification is highly dependent on manual inspections and visual inspections. Specifically, manual inspections can usually only cover limited areas, especially in large forest areas. This detection method is inefficient and difficult to detect pests and diseases in a timely and comprehensive manner. In addition, inspectors need to spend a lot of time and energy on on-foot inspections, and are limited by human physical strength and field of vision, making it difficult to conduct frequent and detailed inspections. In addition, traditional manual detection methods are difficult to achieve real-time monitoring. Once a pest and disease breaks out, it often takes a long time to discover and take measures, missing the best time for prevention and control, thereby causing serious damage to forest resources.

[0004] Therefore, an optimized forestry pest and disease identification scheme is desired. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present application provides a method and system for intelligent identification of forestry pests and diseases based on remote sensing data.

[0006] According to one aspect of the present application, a method for intelligent identification of forestry pests and diseases based on remote sensing data is provided, which comprises:

[0007] Obtain remote sensing images of virgin forest areas;

[0008] Preprocessing the remote sensing image of the original forest area to obtain a preprocessed remote sensing image of the original forest area;

[0009] Extracting color and texture features from the preprocessed original forest area remote sensing image to obtain a forest area remote sensing color feature map and a forest area remote sensing texture feature map;

[0010] The forest area remote sensing color feature map and the forest area remote sensing texture feature map are subjected to feature significant joint interactive coding to obtain forest area remote sensing color-texture significant sparse interactive fusion features, which includes:

[0011] Performing kernel feature distillation and global association on the forest area remote sensing color feature map and the forest area remote sensing texture feature map to obtain a global encoding matrix between the forest area remote sensing color and texture kernel features;

[0012] Based on the global coding matrix between the forest area remote sensing color texture kernel features, the forest area remote sensing color feature map and the forest area remote sensing texture feature map are subjected to correlation sparse interactive fusion to obtain the forest area remote sensing color-texture significant sparse interactive fusion feature;

[0013] Based on the forest area remote sensing color-texture significant sparse interactive fusion features, a recognition result is obtained, and the recognition result is used to indicate whether there are pests and diseases.

[0014] According to another aspect of the present application, a forestry pest and disease intelligent identification system based on remote sensing data is provided, which comprises:

[0015] Forest area remote sensing image collection module, used to obtain remote sensing images of original forest areas;

[0016] A forest area remote sensing image preprocessing module, used for preprocessing the original forest area remote sensing image to obtain a preprocessed original forest area remote sensing image;

[0017] A forest area remote sensing image color and texture feature extraction module is used to extract color and texture features from the preprocessed original forest area remote sensing image to obtain a forest area remote sensing color feature map and a forest area remote sensing texture feature map;

[0018] A forest area remote sensing image color-texture feature interactive coding module is used to perform feature-significant joint interactive coding on the forest area remote sensing color feature map and the forest area remote sensing texture feature map to obtain forest area remote sensing color-texture significant sparse interactive fusion features, wherein the forest area remote sensing image color-texture feature interactive coding module includes: a forest area remote sensing feature kernel distillation global association unit, which is used to perform kernel feature distillation and global association on the forest area remote sensing color feature map and the forest area remote sensing texture feature map to obtain a global coding matrix between forest area remote sensing color texture kernel features; a forest area remote sensing color texture feature interactive fusion unit, which is used to perform association sparse interactive fusion on the forest area remote sensing color feature map and the forest area remote sensing texture feature map based on the global coding matrix between forest area remote sensing color texture kernel features to obtain the forest area remote sensing color-texture significant sparse interactive fusion features;

[0019] The pest and disease identification result generation module is used to obtain the identification result based on the forest area remote sensing color-texture significant sparse interactive fusion feature, and the identification result is used to indicate whether there are pests and diseases.

[0020] This application has significant technical effects due to the adoption of the above technical solutions:

[0021] Compared with the prior art, the intelligent identification method and system of forest pests and diseases based on remote sensing data provided by this application adopts image analysis and processing technology based on deep learning to perform image preprocessing and remote sensing color and texture feature extraction on the acquired remote sensing images of the original forest area, so as to automatically judge whether there are pests and diseases based on the significant sparse interactive fusion features between the remote sensing color features and the remote sensing texture features of the forest area. In this way, remote sensing technology can cover a large area of ​​forest area and realize real-time and comprehensive monitoring without waiting for the manual inspection cycle. In addition, pests and diseases can be accurately identified through deep learning technology, so that measures can be taken quickly to realize efficient, accurate and intelligent monitoring of pests and diseases in forest areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0023] Figure 1 The present invention is a flowchart of a method for intelligently identifying forest pests and diseases based on remote sensing data according to an embodiment of the present application.

[0024] Figure 2 This is a data flow diagram of the intelligent identification method of forestry pests and diseases based on remote sensing data according to an embodiment of the present application.

[0025] Figure 3 In the intelligent identification method of forestry pests and diseases based on remote sensing data according to an embodiment of the present application, the remote sensing color feature map of the forest area and the remote sensing texture feature map of the forest area are subjected to feature-significant joint interactive encoding to obtain a flowchart of the remote sensing color-texture significant interactive fusion feature of the forest area.

[0026] Figure 4 In the method for intelligent identification of forest pests and diseases based on remote sensing data according to an embodiment of the present application, kernel feature distillation and global correlation are performed on the forest area remote sensing color feature map and the forest area remote sensing texture feature map to obtain a global coding matrix between the forest area remote sensing color and texture kernel features.

[0027] Figure 5 In the intelligent identification method of forestry pests and diseases based on remote sensing data according to an embodiment of the present application, based on the global coding matrix between the kernel features of the remote sensing color and texture of the forest area, the remote sensing color feature map of the forest area and the remote sensing texture feature map of the forest area are correlated and sparsely interactively fused to obtain a flowchart of the significant sparse interactive fusion features of the remote sensing color and texture of the forest area.

[0028] Figure 6 This is a system block diagram of an intelligent identification system for forestry pests and diseases based on remote sensing data according to an embodiment of the present application. DETAILED DESCRIPTION

[0029] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0030] In view of the problems in the background technology, the technical concept of this application is to obtain remote sensing images of original forest areas, and use image analysis and processing technology based on deep learning to perform image preprocessing and remote sensing color and texture feature extraction on the remote sensing images of original forest areas, so as to automatically judge whether there are pests and diseases based on the significant sparse interactive fusion features between the remote sensing color features and the remote sensing texture features of the forest areas. In this way, remote sensing technology can cover a large area of ​​forest areas and realize real-time and comprehensive monitoring without waiting for manual inspection cycles. In addition, pests and diseases can be accurately identified through deep learning technology, so that measures can be taken quickly to realize efficient, accurate and intelligent monitoring of pests and diseases in forest areas.

[0031] Figure 1 The present invention is a flowchart of a method for intelligently identifying forest pests and diseases based on remote sensing data according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of the method for intelligent identification of forestry pests and diseases based on remote sensing data according to an embodiment of the present application. Figure 1 and Figure 2 As shown, according to the embodiment of the present application, the intelligent identification method of forestry pests and diseases based on remote sensing data includes: S110, acquiring a remote sensing image of a primitive forest area; S120, preprocessing the remote sensing image of the primitive forest area to obtain a preprocessed remote sensing image of the primitive forest area; S130, respectively performing color and texture feature extraction on the preprocessed remote sensing image of the primitive forest area to obtain a forest area remote sensing color feature map and a forest area remote sensing texture feature map; S140, performing feature significant joint interactive encoding on the forest area remote sensing color feature map and the forest area remote sensing texture feature map to obtain a forest area remote sensing color-texture significant sparse interactive fusion feature; S150, obtaining a recognition result based on the forest area remote sensing color-texture significant sparse interactive fusion feature, and the recognition result is used to indicate whether there are pests and diseases.

[0032] In step S110, a remote sensing image of a virgin forest area is obtained. Specifically, when trees are infected by pests and diseases, the color of the trees will change, usually manifested as yellowing or browning of the leaves. This color change can be detected in remote sensing images, especially in the visible light band. At the same time, pests and diseases can also cause damage to the bark and leaves, that is, the texture information of the trees can also reflect the growth status and health status of the trees. In general, obtaining remote sensing images of virgin forest areas and analyzing the color and texture information in the remote sensing images of virgin forest areas can help accurately judge the health status of trees, thereby realizing intelligent identification of pests and diseases.

[0033] In step S120, the remote sensing image of the original forest area is preprocessed to obtain a preprocessed remote sensing image of the original forest area. Specifically, in an embodiment of the present application, the preprocessing includes radiation correction, atmospheric correction and geometric correction. Specifically, during the acquisition process of the remote sensing image of the original forest area, due to the influence of factors such as sensor characteristics, atmospheric conditions, and lighting, the original image may have some distortion or deviation. Therefore, in order to eliminate or reduce these influences and make the image more accurately reflect the actual situation on the ground, thereby improving the accuracy and reliability of subsequent analysis, in the technical solution of the present application, the remote sensing image of the original forest area is preprocessed to obtain a preprocessed remote sensing image of the original forest area, wherein the preprocessing includes radiation correction, atmospheric correction and geometric correction, so that the high-quality image after preprocessing can more accurately capture the subtle changes caused by pests and diseases, thereby improving the accuracy of subsequent symptom judgment.

[0034] In step S130, the preprocessed original forest area remote sensing image is subjected to color and texture feature extraction to obtain a forest area remote sensing color feature map and a forest area remote sensing texture feature map. Specifically, in an embodiment of the present application, the preprocessed original forest area remote sensing image is subjected to color and texture feature extraction to obtain a forest area remote sensing color feature map and a forest area remote sensing texture feature map, including: inputting the preprocessed original forest area remote sensing image into a color feature extractor based on a first hole convolutional neural network model to obtain the forest area remote sensing color feature map; inputting the preprocessed original forest area remote sensing image into a texture feature extractor based on a second hole convolutional neural network model to obtain the forest area remote sensing texture feature map.

[0035] Specifically, the preprocessed original forest remote sensing image contains rich color information, and these color features reflect the spectral characteristics of surface objects. In particular, in a forest environment, color features can provide important clues about the health status of vegetation. Considering that the hole convolution is a special convolution operation, it expands the receptive field by introducing intervals (holes) between the elements of the convolution kernel, so that the model can capture a wider range of contextual information, which helps to extract features from remote sensing images with complex structures. Based on this, in the technical solution of the present application, the preprocessed original forest remote sensing image is input into a color feature extractor based on the first hole convolution neural network model to obtain a forest remote sensing color feature map. Similarly, the preprocessed original forest remote sensing image contains rich texture features that can reflect the subtle structure and pattern of the surface of surface objects. In forest pest detection, texture features can provide important information about the health status of vegetation. Therefore, in order to more clearly understand the subtle texture changes caused by pests and diseases, in the technical solution of the present application, the preprocessed original forest area remote sensing image is input into a texture feature extractor based on the second hole convolutional neural network model to obtain a remote sensing texture feature map of the forest area.

[0036] In step S140, the forest area remote sensing color feature map and the forest area remote sensing texture feature map are subjected to feature significant joint interactive coding to obtain forest area remote sensing color-texture significant sparse interactive fusion features. Accordingly, considering that the forest area remote sensing color feature map and the forest area remote sensing texture feature map each contain different types of information, and there are interactions and associations between color features and texture features, such as when pests and diseases begin to affect trees, it may cause changes in leaf color, such as yellowing, reddening or browning. At the same time, these color changes are usually accompanied by changes in texture, such as spots, stripes or unevenness on the surface of leaves. Therefore, in order to achieve effective fusion of key significant information in color features and texture features, and thus obtain more reliable recognition results, in the technical solution of the present application, the forest area remote sensing color feature map and the forest area remote sensing texture feature map are subjected to feature significant joint interactive coding to obtain forest area remote sensing color-texture significant sparse interactive fusion feature maps as forest area remote sensing color-texture significant sparse interactive fusion features.

[0037] In detail, first, the forest area remote sensing color feature map and the forest area remote sensing texture feature map are input into the kernel feature distillation network to extract the most core and discriminative features from the original features, while reducing redundant information, and retaining the core content and essential information in the original features to improve the effectiveness of the features, thereby obtaining the forest area remote sensing color kernel feature vector and the forest area remote sensing texture kernel feature vector. Next, the forest area remote sensing color kernel feature vector and the forest area remote sensing texture kernel feature vector are globally associated and encoded to better reflect the significant features between the two, and the potential interactive correlation between color features and texture features is mined to obtain the global encoding matrix between the forest area remote sensing color texture kernel features, which provides data support for subsequent feature sparseness. Subsequently, the global encoding matrix between kernel features is input into the sparse mask unit to convert it into a sparse weight matrix associated with the forest area remote sensing color texture kernel features. Specifically, the sparse mask unit can help select the most representative features and represent them as a sparse weight matrix, so as to reduce redundant features, retain and highlight the most critical and significant features for pest and disease identification, and emphasize those features that show significant changes in both color and texture features. At the same time, in order to pay more attention to and focus on the information interaction and influence between color features and texture features in a more fine-grained manner, the forest area remote sensing color feature map and the forest area remote sensing texture feature map are input into the position-by-position association interaction network to focus on the interaction between the local features of the two, and the forest area remote sensing color-texture position-by-position interaction fusion feature map is obtained. Finally, the forest area remote sensing color-texture position-by-position interaction fusion feature map is subjected to feature significant sparse modulation processing by using the association sparse weight matrix between the kernel features of forest area remote sensing color and texture as the weight matrix to significantly modulate the influence of each position in the fusion feature map according to the importance of each position in the weight matrix, highlighting the most significant and representative areas and features in color and texture features, and obtaining the forest area remote sensing color-texture significant sparse interaction fusion feature map.

[0038] Specifically, Figure 3 The flowchart is a method for intelligently identifying forest pests and diseases based on remote sensing data according to an embodiment of the present application, in which the forest area remote sensing color feature map and the forest area remote sensing texture feature map are subjected to feature significant joint interactive coding to obtain forest area remote sensing color-texture significant sparse interactive fusion features. Figure 3As shown, the forest area remote sensing color feature map and the forest area remote sensing texture feature map are subjected to feature significant joint interactive coding to obtain the forest area remote sensing color-texture significant sparse interactive fusion feature, including: S141, kernel feature distillation and global correlation are performed on the forest area remote sensing color feature map and the forest area remote sensing texture feature map to obtain the global coding matrix between the forest area remote sensing color texture kernel features; S142, based on the global coding matrix between the forest area remote sensing color texture kernel features, the forest area remote sensing color feature map and the forest area remote sensing texture feature map are subjected to correlation sparse interactive fusion to obtain the forest area remote sensing color-texture significant sparse interactive fusion feature.

[0039] More specifically, Figure 4 This is a flowchart of performing kernel feature distillation and global association on the forest area remote sensing color feature map and the forest area remote sensing texture feature map to obtain a global coding matrix between the forest area remote sensing color and texture kernel features in the forest area remote sensing data intelligent identification method according to the embodiment of the present application. Figure 4 As shown, kernel feature distillation and global association are performed on the forest area remote sensing color feature map and the forest area remote sensing texture feature map to obtain a global coding matrix between the forest area remote sensing color and texture kernel features, including: S1411, inputting the forest area remote sensing color feature map and the forest area remote sensing texture feature map into the kernel feature distillation network to obtain a forest area remote sensing color kernel feature vector and a forest area remote sensing texture kernel feature vector; S1412, performing global association encoding on the forest area remote sensing color kernel feature vector and the forest area remote sensing texture kernel feature vector to obtain a global coding matrix between the forest area remote sensing color and texture kernel features.

[0040] In particular, the processing steps are the same when the forest area remote sensing color feature map and the forest area remote sensing texture feature map are input into the kernel feature distillation network for processing, so the processing of the forest area remote sensing color feature map is used to illustrate here. More specifically, in an embodiment of the present application, the forest area remote sensing color feature map and the forest area remote sensing texture feature map are input into the kernel feature distillation network to obtain the forest area remote sensing color kernel feature vector and the forest area remote sensing texture kernel feature vector, including: performing feature fine-grained decomposition on the forest area remote sensing color feature map to obtain a set of forest area remote sensing color feature vectors; calculating the feature difference coefficients of each forest area remote sensing color feature vector in the set of forest area remote sensing color feature vectors relative to other forest area remote sensing color feature vectors to obtain a set of forest area remote sensing color feature difference coefficients; and normalizing the set of forest area remote sensing color feature difference coefficients to obtain a set of forest area remote sensing color feature difference coefficients. A set of normalized forest area remote sensing color feature difference coefficients is obtained; the set of normalized forest area remote sensing color feature difference coefficients is used as weights to calculate the weighted sum of the set of forest area remote sensing color feature vectors to obtain the forest area remote sensing color kernel feature vector. Specifically, it can be understood that, assuming that the set of normalized forest area remote sensing color feature difference coefficients is (A1, A2, A3...An), and the set of forest area remote sensing color feature vectors is (B1, B2, B3...Bn), then A1, A2, A3...An are used as the weights of B1, B2, B3...Bn to calculate the weighted sum, which can be expressed by the formula: .

[0041] More specifically, in an embodiment of the present application, the characteristic difference coefficient of each forest area remote sensing color feature vector in the set of forest area remote sensing color feature vectors relative to other forest area remote sensing color feature vectors is calculated to obtain a set of forest area remote sensing color feature difference coefficients, including: extracting a predetermined forest area remote sensing color feature vector from the set of forest area remote sensing color feature vectors; subtracting the predetermined forest area remote sensing color feature vector from other forest area remote sensing color feature vectors in the set of forest area remote sensing color feature vectors by position, and calculating a norm of each forest area remote sensing color feature difference vector in the obtained set of forest area remote sensing color feature difference vectors to obtain a set of forest area remote sensing color feature norm values; calculating the global mean of the set of forest area remote sensing color feature norm values ​​to obtain the forest area remote sensing color feature difference coefficient corresponding to the predetermined forest area remote sensing color feature vector.

[0042] More specifically, Figure 5 In the method for intelligent identification of forest pests and diseases based on remote sensing data according to the embodiment of the present application, based on the global coding matrix between the forest area remote sensing color texture kernel features, the forest area remote sensing color feature map and the forest area remote sensing texture feature map are subjected to correlation sparse interactive fusion to obtain the forest area remote sensing color-texture significant sparse interactive fusion feature. Figure 5 As shown, based on the global coding matrix between the kernel features of the forest area remote sensing color texture, the forest area remote sensing color feature map and the forest area remote sensing texture feature map are associated and sparsely interactively fused to obtain the forest area remote sensing color-texture significant sparse interactive fusion features, including: S1421, inputting the global coding matrix between the kernel features of the forest area remote sensing color texture into a sparse mask unit to obtain the associated sparse weight matrix between the kernel features of the forest area remote sensing color texture; S1422, inputting the forest area remote sensing color feature map and the forest area remote sensing texture feature map into a position-by-position associated interactive network to obtain a forest area remote sensing color-texture position-by-position interactive fusion feature map; S1423, based on the associated sparse weight matrix between the kernel features of the forest area remote sensing color texture, performing feature significant sparse modulation on the forest area remote sensing color-texture position-by-position interactive fusion feature map to obtain a forest area remote sensing color-texture significant sparse interactive fusion feature map as the forest area remote sensing color-texture significant sparse interactive fusion feature.

[0043] More specifically, in an embodiment of the present application, the global coding matrix between the kernel features of the remote sensing color texture of the forest area is input into a sparse mask unit to obtain a sparse weight matrix associated with the kernel features of the remote sensing color texture of the forest area, including: comparing the eigenvalues ​​of each position in the global coding matrix between the kernel features of the remote sensing color texture of the forest area with a preset threshold to obtain a sparse weight matrix associated with the kernel features of the remote sensing color texture of the forest area composed of multiple updated eigenvalues; wherein, in response to the eigenvalue being greater than the second preset threshold, the product of the eigenvalue and a constant three is used as the updated eigenvalue; in response to the eigenvalue being greater than the first preset threshold and less than or equal to the second preset threshold, the original value of the eigenvalue is used as the updated eigenvalue; in response to the eigenvalue being less than or equal to the first preset threshold, the value obtained by dividing the eigenvalue by the constant three is used as the updated eigenvalue.

[0044] More specifically, in an embodiment of the present application, the forest area remote sensing color feature map and the forest area remote sensing texture feature map are input into a position-by-position association interaction network to obtain a forest area remote sensing color-texture position-by-position interactive fusion feature map, including: performing positional point multiplication on the eigenvalues ​​of corresponding positions in each feature matrix along the channel dimension in the forest area remote sensing color feature map and each feature matrix along the channel dimension in the forest area remote sensing texture feature map to obtain the forest area remote sensing color-texture position-by-position interactive fusion feature map.

[0045] More specifically, in an embodiment of the present application, based on the sparse weight matrix associated between the kernel features of the forest area remote sensing color texture, the forest area remote sensing color-texture position-by-position interactive fusion feature map is subjected to feature-significant sparse modulation to obtain the forest area remote sensing color-texture significant sparse interactive fusion feature map, including: matrix multiplying each feature matrix along the channel dimension in the forest area remote sensing color-texture position-by-position interactive fusion feature map with the sparse weight matrix associated between the kernel features of the forest area remote sensing color texture to obtain the forest area remote sensing color-texture significant sparse interactive fusion feature map.

[0046] In the embodiment of the present application, specifically, the forest area remote sensing color feature map and the forest area remote sensing texture feature map are subjected to feature significant joint interactive coding, which can be expressed by the formula:

[0047] ;

[0048] ;

[0049] ;

[0050] ;

[0051] ;

[0052] ;

[0053] ;

[0054] in, and are respectively the forest area remote sensing color feature map and the forest area remote sensing texture feature map, For the feature map Perform kernel feature distillation. and are the forest area remote sensing color kernel feature vector and forest area remote sensing texture kernel feature vector, respectively. For the feature map Perform feature fine-grained decomposition operation, that is, the feature fine-grained decomposition operation is represented by the feature map Each characteristic matrix of is expanded into characteristic vectors, is the first, second, ..., first eigenvector in the set of decomposed eigenvectors ,..., ,..., After decomposition, the eigenvectors represents the norm of the eigenvector, is the set of decomposed eigenvectors, represents the number of vectors in the set of decomposed eigenvectors minus one, yes The corresponding characteristic difference coefficient, Indicated by natural constant The exponential function with base , is the number of vectors in the set of decomposed eigenvectors, is the kernel eigenvector, yes The transposed vector of is matrix multiplication, is the global coding matrix between the kernel features of the forest area remote sensing color texture, Expressed as and Perform global association coding to obtain , is a sparse masking operation, yes The eigenvalues ​​at each position in , is the preset threshold, is the correlation sparse weight matrix between the forest area remote sensing color texture kernel features, For point-by-position multiplication, The forest area remote sensing color-texture position-by-position interactive fusion feature map, is the forest area remote sensing color-texture significant sparse interactive fusion feature map, Expressed as and Input the position-by-position association interaction network to obtain , that is, the position-by-position interaction network can be used in the input and Then multiply the two by position and get .

[0055] In step S150, based on the forest area remote sensing color-texture significant sparse interactive fusion feature, a recognition result is obtained, and the recognition result is used to indicate whether there are pests and diseases. Specifically, in an embodiment of the present application, based on the forest area remote sensing color-texture significant sparse interactive fusion feature, a recognition result is obtained, including: inputting the forest area remote sensing color-texture significant sparse interactive fusion feature map into a forestry pest and disease intelligent recognition module based on a classifier to obtain the recognition result. That is, the forest area remote sensing color feature map and the forest area remote sensing texture feature map are significantly combined and interactively encoded to obtain the forest area remote sensing color-texture significant sparse interactive fusion feature for classification processing, so as to automatically determine whether there are pests and diseases, so that a large area of ​​forest area can be covered by remote sensing technology, and real-time and comprehensive monitoring can be achieved without waiting for a manual inspection cycle. In addition, pests and diseases can be accurately identified through deep learning technology, so that measures can be taken quickly to achieve efficient, accurate and intelligent monitoring of forest area pests and diseases.

[0056] In particular, the present application takes into account that the forest area remote sensing color feature map and the forest area remote sensing texture feature map respectively represent the forest area remote sensing image color features and forest area remote sensing image texture features of the preprocessed original forest area remote sensing image. When they are subjected to feature significant joint interactive encoding, the distribution differences of image semantic features at different scales and depths will also have kernel-correlated sparse mask perceptual differences. Therefore, it is expected to improve the complex spatial structure regression convergence and generalization effect of the forest area remote sensing color-texture significant sparse interactive fusion feature map based on the joint perceptual interaction of different spatial distribution structures.

[0057] Preferably, in an example of the present application, the forest area remote sensing color-texture significant sparse interactive fusion feature map is input into a classifier-based forestry pest and disease intelligent recognition module to obtain a recognition result, including: calculating the sum of the absolute values ​​of each eigenvalue of the forest area remote sensing color-texture significant sparse interactive fusion feature map to obtain a first forest area remote sensing color-texture significant sparse interactive fusion spatial structure value, and calculating the square root of the sum of the squares of each eigenvalue of the forest area remote sensing color-texture significant sparse interactive fusion feature map to obtain a second forest area remote sensing color-texture significant sparse interactive fusion spatial structure value; the forest area remote sensing color-texture significant sparse interactive fusion feature map is input into a classifier-based forestry pest and disease intelligent recognition module to obtain a recognition result, including: calculating the sum of the absolute values ​​of each eigenvalue of the forest area remote sensing color-texture significant sparse interactive fusion feature map to obtain a first forest area remote sensing color-texture significant sparse interactive fusion spatial structure value, and calculating the square root of the sum of the squares of each eigenvalue of the forest area remote sensing color-texture significant sparse interactive fusion feature map to obtain a second forest area remote sensing color-texture significant sparse interactive fusion spatial structure value. Each eigenvalue of the sparse interactive fusion feature map is multiplied by the first forest area remote sensing color-texture significant sparse interactive fusion spatial structure value and the second forest area remote sensing color-texture significant sparse interactive fusion spatial structure value to obtain the first forest area remote sensing color-texture significant sparse interactive fusion structure reference value and the second forest area remote sensing color-texture significant sparse interactive fusion structure reference value corresponding to each eigenvalue; each eigenvalue of the forest area remote sensing color-texture significant sparse interactive fusion feature map is multiplied by the scale of the forest area remote sensing color-texture significant sparse interactive fusion feature map and the square root of the scale to obtain the first forest area remote sensing color-texture significant sparse interactive fusion structure reference value and the second forest area remote sensing color-texture significant sparse interactive fusion structure reference value corresponding to each eigenvalue. The first forest area remote sensing color-texture significant sparse interactive fusion scale transformation value and the second forest area remote sensing color-texture significant sparse interactive fusion scale transformation value; the first forest area remote sensing color-texture significant sparse interactive fusion structure reference value is divided by the difference between the first forest area remote sensing color-texture significant sparse interactive fusion spatial structure value and the first forest area remote sensing color-texture significant sparse interactive fusion scale transformation value to obtain the first forest area remote sensing color-texture significant sparse interactive fusion transformation adjustment value; the second forest area remote sensing color-texture significant sparse interactive fusion structure reference value is divided by the second forest area remote sensing color-texture significant sparse interactive fusion spatial structure value The difference between the intermediate structure value and the second forest area remote sensing color-texture significant sparse interactive fusion scale transformation value is used to obtain the second forest area remote sensing color-texture significant sparse interactive fusion transformation adjustment value; the weighted sum of the first forest area remote sensing color-texture significant sparse interactive fusion transformation adjustment value and the second forest area remote sensing color-texture significant sparse interactive fusion transformation adjustment value is calculated to obtain each eigenvalue of the optimized forest area remote sensing color-texture significant sparse interactive fusion feature map; the optimized forest area remote sensing color-texture significant sparse interactive fusion feature map is input into the classifier-based forestry pest and disease intelligent recognition module to obtain the recognition result.

[0058] The optimized representation of the forest area remote sensing color-texture significant sparse interactive fusion feature map is as follows:

[0059] ;

[0060] ;

[0061] ;

[0062] ;

[0063] ;

[0064] ;

[0065] ;

[0066] ;

[0067] ;

[0068] in, is the forest area remote sensing color-texture significant sparse interactive fusion feature map, and Respectively represent the width and height of each feature matrix of the forest area remote sensing color-texture significant sparse interactive fusion feature map along the channel dimension, represents the number of channels of the forest area remote sensing color-texture significant sparse interactive fusion feature map, represents the field of real numbers, represents the scale of the forest area remote sensing color-texture significant sparse interactive fusion feature map, Indicates the feature values ​​of each position in the forest area remote sensing color-texture significant sparse interactive fusion feature map, Indicates the spatial structure value of remote sensing color-texture significant sparse interactive fusion in the first forest area, It represents the spatial structure value of remote sensing color-texture significant sparse interactive fusion in the second forest area, It represents the first forest area remote sensing color-texture significant sparse interactive fusion transformation adjustment value of each position in the first forest area remote sensing color-texture significant sparse interactive fusion transformation adjustment map, It represents the second forest area remote sensing color-texture significant sparse interactive fusion transformation adjustment value of each position in the second forest area remote sensing color-texture significant sparse interactive fusion transformation adjustment map, It represents the remote sensing color-texture significant sparse interactive fusion transformation adjustment map of the first forest area, Represents the remote sensing color-texture significant sparse interactive fusion transformation adjustment map of the second forest area, Indicates adding by position point. It means point multiplication by position. represents the weighted hyperparameter, Represents the optimized forest area remote sensing color-texture significant sparse interactive fusion feature map.

[0069] In the preferred example, for the spatial structure information of the feature set of the forest area remote sensing color-texture significant sparse interactive fusion feature map in the high-dimensional space, the scale-based frame transformation of each eigenvalue of the forest area remote sensing color-texture significant sparse interactive fusion feature map is performed by taking the class norm space structured representation of the forest area remote sensing color-texture significant sparse interactive fusion feature map as the reference window, and the spatial structure-based frame attention weight adjustment of each eigenvalue of the forest area remote sensing color-texture significant sparse interactive fusion feature map is realized to ensure the invariance of spatial transformation (translation, scaling and rotation) of the forest area remote sensing color-texture significant sparse interactive fusion feature map under feature space interaction, thereby improving the convergence and generalization effect of classification regression of the feature set of the forest area remote sensing color-texture significant sparse interactive fusion feature map under complex spatial structure representation, and improving the accuracy of the recognition result obtained by inputting it into the classifier-based forestry pest and disease intelligent recognition module. In this way, large areas of forest areas can be covered by remote sensing technology to achieve real-time and comprehensive monitoring without waiting for manual inspection cycles. In addition, deep learning technology can accurately identify pests and diseases, so that measures can be taken quickly to achieve efficient, accurate and intelligent monitoring of pests and diseases in forest areas.

[0070] In summary, the intelligent identification method of forest pests and diseases based on remote sensing data based on the embodiment of the present application is explained, which uses image analysis and processing technology based on deep learning to perform image preprocessing and remote sensing color and texture feature extraction on the acquired original forest remote sensing images, so as to automatically judge whether there are pests and diseases based on the significant sparse interactive fusion features between the remote sensing color features and the remote sensing texture features of the forest area. Through remote sensing technology, large areas of forest areas can be covered to achieve real-time and comprehensive monitoring without waiting for manual inspection cycles. In addition, pests and diseases can be accurately identified through deep learning technology, so that measures can be taken quickly to achieve efficient, accurate and intelligent monitoring of pests and diseases in forest areas.

[0071] Figure 6 FIG. 1 is a system block diagram of a forestry pest and disease intelligent identification system based on remote sensing data according to an embodiment of the present application. Figure 6As shown, according to the embodiment of the present application, the intelligent identification system 100 of forest pests and diseases based on remote sensing data includes: a forest area remote sensing image collection module 110, which is used to obtain the original forest area remote sensing image; a forest area remote sensing image preprocessing module 120, which is used to preprocess the original forest area remote sensing image to obtain the preprocessed original forest area remote sensing image; a forest area remote sensing image color and texture feature extraction module 130, which is used to extract color and texture features of the preprocessed original forest area remote sensing image respectively to obtain a forest area remote sensing color feature map and a forest area remote sensing texture feature map; a forest area remote sensing image color-texture feature interactive coding module 140, which is used to perform feature significant joint interactive coding on the forest area remote sensing color feature map and the forest area remote sensing texture feature map to obtain a forest area remote sensing color-texture significant interactive fusion feature; The pest identification result generation module 150 is used to obtain the identification result based on the forest area remote sensing color-texture significant sparse interactive fusion feature, and the identification result is used to indicate whether there are pests and diseases; wherein the forest area remote sensing image color-texture feature interactive coding module includes: a forest area remote sensing feature kernel distillation global association unit, which is used to perform kernel feature distillation and global association on the forest area remote sensing color feature map and the forest area remote sensing texture feature map to obtain the global coding matrix between the forest area remote sensing color texture kernel features; a forest area remote sensing color texture feature interactive fusion unit, which is used to perform association and sparse interactive fusion on the forest area remote sensing color feature map and the forest area remote sensing texture feature map based on the global coding matrix between the forest area remote sensing color texture kernel features to obtain the forest area remote sensing color-texture significant sparse interactive fusion feature.

[0072] The specific functions and operations of each unit and module in the above-mentioned forestry pest intelligent identification system 100 based on remote sensing data have been referred to above. Figures 1 to 5 The description of the intelligent identification method of forestry pests and diseases based on remote sensing data has been introduced in detail, and therefore, its repeated description will be omitted.

[0073] In summary, the forestry pest and disease intelligent identification system 100 based on remote sensing data according to the embodiment of the present application is explained, which uses deep learning-based image analysis and processing technology to perform image preprocessing and remote sensing color and texture feature extraction on the acquired original forest remote sensing image, so as to automatically judge whether pests and diseases exist according to the significant sparse interactive fusion features between the forest remote sensing color features and the forest remote sensing texture features. Through remote sensing technology, large areas of forest can be covered to achieve real-time and comprehensive monitoring without waiting for manual inspection cycles. In addition, pests and diseases can be accurately identified through deep learning technology, so that measures can be taken quickly to achieve efficient, accurate and intelligent monitoring of forest pests and diseases.

[0074] In the several embodiments provided in this application, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are only schematic, for example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0075] It is obvious to those skilled in the art that the present application is not limited to the details of the above exemplary embodiments, and that the present application can be implemented in other specific forms without departing from the spirit or essential features of the present application.

[0076] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present application and are not intended to limit it. Although the present application has been described in detail with reference to the preferred embodiments, a person of ordinary skill in the art should understand that the technical solution of the present application may be modified or replaced by equivalents without departing from the spirit of the technical solution of the present application.

Claims

1. An intelligent identification method for forestry pests and diseases based on remote sensing data, characterized in that: include: Obtain remote sensing images of virgin forest areas; Preprocessing the remote sensing image of the original forest area to obtain a preprocessed remote sensing image of the original forest area; Extracting color and texture features from the preprocessed original forest area remote sensing image to obtain a forest area remote sensing color feature map and a forest area remote sensing texture feature map; The forest area remote sensing color feature map and the forest area remote sensing texture feature map are subjected to feature significant joint interactive coding to obtain forest area remote sensing color-texture significant sparse interactive fusion features, which includes: Performing kernel feature distillation and global association on the forest area remote sensing color feature map and the forest area remote sensing texture feature map to obtain a global encoding matrix between the forest area remote sensing color and texture kernel features; Based on the global coding matrix between the forest area remote sensing color texture kernel features, the forest area remote sensing color feature map and the forest area remote sensing texture feature map are subjected to correlation sparse interactive fusion to obtain the forest area remote sensing color-texture significant sparse interactive fusion feature; Based on the forest area remote sensing color-texture significant sparse interactive fusion features, a recognition result is obtained, and the recognition result is used to indicate whether there is a disease or insect pest; Based on the global coding matrix between the forest area remote sensing color texture kernel features, the forest area remote sensing color feature map and the forest area remote sensing texture feature map are subjected to associated sparse interactive fusion to obtain the forest area remote sensing color-texture significant sparse interactive fusion features, including: Inputting the global encoding matrix between the forest area remote sensing color texture kernel features into the sparse mask unit to obtain the correlation sparse weight matrix between the forest area remote sensing color texture kernel features; Inputting the forest area remote sensing color feature map and the forest area remote sensing texture feature map into a position-by-position association interaction network to obtain a forest area remote sensing color-texture position-by-position interactive fusion feature map; Based on the correlation sparse weight matrix between the kernel features of the forest area remote sensing color and texture, the forest area remote sensing color-texture position-by-position interactive fusion feature map is subjected to feature significant sparse modulation to obtain the forest area remote sensing color-texture significant sparse interactive fusion feature map as the forest area remote sensing color-texture significant sparse interactive fusion feature.

2. The method for intelligent identification of forestry pests and diseases based on remote sensing data according to claim 1 is characterized in that: The preprocessing includes radiation correction, atmospheric correction and geometric correction.

3. The method for intelligent identification of forestry pests and diseases based on remote sensing data according to claim 2 is characterized in that: The pre-processed original forest remote sensing image is subjected to color and texture feature extraction to obtain a forest remote sensing color feature map and a forest remote sensing texture feature map, including: Inputting the preprocessed original forest area remote sensing image into a color feature extractor based on a first hole convolutional neural network model to obtain the forest area remote sensing color feature map; The preprocessed original forest area remote sensing image is input into a texture feature extractor based on a second hole convolutional neural network model to obtain the forest area remote sensing texture feature map.

4. The method for intelligent identification of forestry pests and diseases based on remote sensing data according to claim 3 is characterized in that: The forest area remote sensing color feature map and the forest area remote sensing texture feature map are subjected to kernel feature distillation and global association to obtain a global coding matrix between forest area remote sensing color and texture kernel features, including: Inputting the forest area remote sensing color feature map and the forest area remote sensing texture feature map into a kernel feature distillation network to obtain a forest area remote sensing color kernel feature vector and a forest area remote sensing texture kernel feature vector; The forest area remote sensing color kernel feature vector and the forest area remote sensing texture kernel feature vector are globally correlated and encoded to obtain a global encoding matrix between the forest area remote sensing color and texture kernel features.

5. The method for intelligent identification of forestry pests and diseases based on remote sensing data according to claim 4 is characterized in that: Inputting the forest area remote sensing color feature map and the forest area remote sensing texture feature map into a kernel feature distillation network to obtain a forest area remote sensing color kernel feature vector and a forest area remote sensing texture kernel feature vector, including: Performing feature fine-grained decomposition on the forest area remote sensing color feature map to obtain a set of forest area remote sensing color feature vectors; Calculating the characteristic difference coefficient of each forest area remote sensing color feature vector in the set of forest area remote sensing color feature vectors relative to other forest area remote sensing color feature vectors to obtain a set of forest area remote sensing color feature difference coefficients; Normalizing the set of forest area remote sensing color feature difference coefficients to obtain a set of normalized forest area remote sensing color feature difference coefficients; The set of normalized forest area remote sensing color feature difference coefficients is used as weights to calculate the weighted sum of the set of forest area remote sensing color feature vectors to obtain the forest area remote sensing color kernel feature vector.

6. The method for intelligent identification of forestry pests and diseases based on remote sensing data according to claim 5 is characterized in that: Calculating the characteristic difference coefficient of each forest area remote sensing color feature vector in the set of forest area remote sensing color feature vectors relative to other forest area remote sensing color feature vectors to obtain a set of forest area remote sensing color feature difference coefficients, including: Extracting a predetermined forest area remote sensing color feature vector from the set of forest area remote sensing color feature vectors; Subtracting the predetermined forest area remote sensing color feature vector from other forest area remote sensing color feature vectors in the set of forest area remote sensing color feature vectors by position, and calculating a norm of each forest area remote sensing color feature difference vector in the obtained set of forest area remote sensing color feature difference vectors to obtain a set of forest area remote sensing color feature norm values; The global mean of the set of norm values ​​of the forest area remote sensing color features is calculated to obtain the forest area remote sensing color feature difference coefficient corresponding to the predetermined forest area remote sensing color feature vector.

7. The method for intelligent identification of forestry pests and diseases based on remote sensing data according to claim 6 is characterized in that: The global coding matrix between the forest area remote sensing color texture kernel features is input into the sparse mask unit to obtain the correlation sparse weight matrix between the forest area remote sensing color texture kernel features, including: Compare the eigenvalues ​​of each position in the global coding matrix between the forest area remote sensing color texture kernel features with a preset threshold to obtain a correlation sparse weight matrix between the forest area remote sensing color texture kernel features composed of a plurality of updated eigenvalues; In response to the eigenvalues ​​of each position in the global coding matrix between the kernel features of the forest area remote sensing color texture being greater than a second preset threshold, the product of the eigenvalues ​​of each position in the global coding matrix between the kernel features of the forest area remote sensing color texture and a constant three is used as the updated eigenvalue; In response to the eigenvalues ​​of each position in the global coding matrix between the forest area remote sensing color texture kernel features being greater than a first preset threshold and less than or equal to the second preset threshold, using the original values ​​of the eigenvalues ​​of each position in the global coding matrix between the forest area remote sensing color texture kernel features as the updated eigenvalues; In response to the eigenvalues ​​of each position in the global coding matrix between the kernel features of the forest area remote sensing color texture being less than or equal to the first preset threshold, the values ​​obtained by dividing the eigenvalues ​​of each position in the global coding matrix between the kernel features of the forest area remote sensing color texture by three constants are used as the updated eigenvalues.

8. The method for intelligent identification of forestry pests and diseases based on remote sensing data according to claim 7 is characterized in that: Based on the forest area remote sensing color-texture significant sparse interactive fusion features, an identification result is obtained, including: inputting the forest area remote sensing color-texture significant sparse interactive fusion feature map into a classifier-based forestry pest and disease intelligent identification module to obtain the identification result.

9. Intelligent identification system of forest pests and diseases based on remote sensing data, characterized by: include: Forest area remote sensing image collection module, used to obtain remote sensing images of original forest areas; A forest area remote sensing image preprocessing module, used for preprocessing the original forest area remote sensing image to obtain a preprocessed original forest area remote sensing image; A forest area remote sensing image color and texture feature extraction module is used to extract color and texture features from the preprocessed original forest area remote sensing image to obtain a forest area remote sensing color feature map and a forest area remote sensing texture feature map; A forest area remote sensing image color-texture feature interactive coding module is used to perform feature-significant joint interactive coding on the forest area remote sensing color feature map and the forest area remote sensing texture feature map to obtain forest area remote sensing color-texture significant sparse interactive fusion features, wherein the forest area remote sensing image color-texture feature interactive coding module includes: a forest area remote sensing feature kernel distillation global association unit, which is used to perform kernel feature distillation and global association on the forest area remote sensing color feature map and the forest area remote sensing texture feature map to obtain a global coding matrix between forest area remote sensing color texture kernel features; a forest area remote sensing color texture feature interactive fusion unit, which is used to perform association sparse interactive fusion on the forest area remote sensing color feature map and the forest area remote sensing texture feature map based on the global coding matrix between forest area remote sensing color texture kernel features to obtain the forest area remote sensing color-texture significant sparse interactive fusion features; A pest and disease identification result generation module is used to obtain an identification result based on the forest area remote sensing color-texture significant sparse interactive fusion feature, and the identification result is used to indicate whether there is a pest and disease; The forest area remote sensing color and texture feature interactive fusion unit is used to: Inputting the global encoding matrix between the forest area remote sensing color texture kernel features into the sparse mask unit to obtain the correlation sparse weight matrix between the forest area remote sensing color texture kernel features; Inputting the forest area remote sensing color feature map and the forest area remote sensing texture feature map into a position-by-position association interaction network to obtain a forest area remote sensing color-texture position-by-position interactive fusion feature map; Based on the correlation sparse weight matrix between the kernel features of the forest area remote sensing color and texture, the forest area remote sensing color-texture position-by-position interactive fusion feature map is subjected to feature significant sparse modulation to obtain the forest area remote sensing color-texture significant sparse interactive fusion feature map as the forest area remote sensing color-texture significant sparse interactive fusion feature.

Images