Intelligent monitoring and early warning system and method for forest plant diseases and insect pests

Abstract

The invention relates to the field of intelligent detection, and provides an intelligent monitoring and early warning system and method for forest plant diseases and insect pests, which adopts an image processing technology based on artificial intelligence to analyze a high-dimensional feature space of a multispectral remote sensing image in a target forest monitoring area. And further identifying whether the target forest monitoring area has a pest and disease damage condition or not. Therefore, large-scale rapid intelligent monitoring of forest tree diseases and insect pests can be realized, and the problem of low efficiency of a traditional ground survey method is solved.

Description

Technical Field

[0001] The present application relates to the field of intelligent detection, and more specifically, to an intelligent monitoring and early warning system and method for forest pests and diseases. Background Art

[0002] As an important part of the ecological environment, the healthy and sustainable development of forestry is of great significance for regulating the climate, protecting the environment, and providing ecological products. However, forest pests and diseases are one of the most common constraints in forestry production, which not only affects the growth quality of trees, but also may lead to the deterioration of the ecological environment. In order to effectively prevent and control the occurrence of forest pests and diseases and improve the efficiency of forestry production, it is necessary to monitor and warn of forest pests and diseases. In the monitoring and early warning of forest pests and diseases, although the traditional ground survey method can accurately detect the occurrence of pests and diseases, it is limited by the monitoring efficiency and quality of large-scale forest areas.

[0003] Therefore, an optimized intelligent monitoring and early warning solution for forest pests and diseases is needed. Summary of the invention

[0004] In view of the shortcomings of the prior art, this application provides a forest pest and disease intelligent monitoring and early warning system and method, which adopts artificial intelligence-based image processing technology to analyze the multi-spectral remote sensing images in the target forest monitoring area in high-dimensional feature space, and then identify whether the target forest monitoring area is invaded by pests and diseases. In this way, large-scale and rapid intelligent monitoring of forest pests and diseases can be achieved, overcoming the problem of low efficiency of traditional ground survey methods.

[0005] According to one aspect of the present application, there is provided an intelligent monitoring and early warning system for forest pests and diseases, comprising: The forest image data acquisition module is used to obtain multispectral remote sensing images of the target forest monitoring area; A forest image preprocessing module, used for performing image preprocessing on the multispectral remote sensing image of the target forest monitoring area to obtain a preprocessed forest multispectral remote sensing image; A forest image feature encoding module is used to perform feature encoding on the preprocessed forest multispectral remote sensing image to obtain a forest multispectral remote sensing global image feature matrix; A forest image feature enhancement module, used for performing feature enhancement on the forest multispectral remote sensing global image feature matrix to obtain a forest multispectral remote sensing global feature enhancement matrix; The forest pest and disease monitoring result generation module is used to obtain the result of whether the target forest monitoring area is invaded by pests and diseases based on the forest multi-spectral remote sensing global feature enhancement matrix.

[0006] According to another aspect of the present application, a method for intelligent monitoring and early warning of forest pests and diseases is provided, which comprises: Acquire multispectral remote sensing images of the target forest monitoring area; Performing image preprocessing on the multispectral remote sensing image of the target forest monitoring area to obtain a preprocessed forest multispectral remote sensing image; Performing feature encoding on the preprocessed forest multispectral remote sensing image to obtain a forest multispectral remote sensing global image feature matrix; Performing feature enhancement on the forest multispectral remote sensing global image feature matrix to obtain a forest multispectral remote sensing global feature enhancement matrix; Based on the forest multispectral remote sensing global feature enhancement matrix, the result of whether the target forest monitoring area is invaded by pests and diseases is obtained.

[0007] This application has significant technical effects due to the adoption of the above technical solutions: The intelligent monitoring and early warning system and method for forest pests and diseases provided by the present application adopts an image processing technology based on artificial intelligence, and analyzes the multi-spectral remote sensing images in the target forest monitoring area in a high-dimensional feature space to identify whether there is an invasion of pests and diseases in the target forest monitoring area. In this way, large-scale and rapid intelligent monitoring of forest pests and diseases can be achieved, overcoming the problem of low efficiency of traditional ground survey methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] 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.

[0009] Figure 1 The present invention is a flowchart of a method for intelligent monitoring and early warning of forest pests and diseases according to an embodiment of the present application.

[0010] Figure 2 The present invention is a flowchart of performing feature encoding on the pre-processed forest multispectral remote sensing image to obtain a forest multispectral remote sensing global image feature matrix in the forest disease and insect pest intelligent monitoring and early warning method according to an embodiment of the present application.

[0011] Figure 3 The present invention is a flowchart of performing feature extraction on the image blocks of the preprocessed forest multispectral remote sensing image to obtain multiple forest multispectral remote sensing local feature matrices in the intelligent monitoring and early warning method for forest pests and diseases according to an embodiment of the present application.

[0012] Figure 4This is a flowchart of obtaining the result of whether there is pest and disease invasion in the target forest monitoring area based on the forest multi-spectral remote sensing global feature enhancement matrix in the forest pest and disease intelligent monitoring and early warning method according to the embodiment of the present application.

[0013] Figure 5 This is a system block diagram of the intelligent monitoring and early warning system for forest pests and diseases according to an embodiment of the present application. DETAILED DESCRIPTION

[0014] 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.

[0015] As mentioned in the above background technology, forestry is an important part of the ecological environment, and its healthy and sustainable development is of great significance for regulating climate, protecting the environment, and providing ecological products. Forests are the "green lungs" of the earth. They absorb carbon dioxide and release oxygen through photosynthesis, helping to maintain the carbon-oxygen balance of the earth. At the same time, forests can also maintain water and soil, regulate water cycles, and provide habitats for biodiversity, playing a key role in maintaining ecological balance.

[0016] However, forest pests and diseases are one of the most common constraints in forestry production. These pests and diseases not only affect the growth quality of trees, but may also lead to the deterioration of the ecological environment. For example, certain pests such as pine wood nematodes and pine caterpillars can reproduce rapidly and destroy forests on a large scale, causing serious ecological and economic losses. In order to effectively prevent and control the occurrence of forest pests and diseases and improve the efficiency of forestry production, it is necessary to monitor and warn of forest pests and diseases. Although traditional ground survey methods can accurately detect the occurrence of pests and diseases, they are limited by the monitoring efficiency and quality of large-scale forest areas. These methods often require a lot of manpower and material resources, and may result in incomplete monitoring coverage and untimely information updates due to factors such as complex terrain and inconvenient transportation. Therefore, an optimized intelligent monitoring and early warning solution for forest pests and diseases is expected.

[0017] In recent years, deep learning and neural networks have been widely used in computer vision, natural language processing, text signal processing and other fields. In addition, deep learning and neural networks have also demonstrated a level close to or even beyond that of humans in image classification, object detection, semantic segmentation, text translation and other fields. The development of deep learning and neural networks has provided new solutions and solutions for intelligent monitoring and early warning of forest pests and diseases.

[0018] Figure 1 Flow chart of the intelligent monitoring and early warning method for forest pests and diseases according to the embodiment of the present application. Figure 1As shown, according to the embodiment of the present application, the intelligent monitoring and early warning method for forest pests and diseases includes: S110, acquiring a multispectral remote sensing image of a target forest monitoring area; S120, performing image preprocessing on the multispectral remote sensing image of the target forest monitoring area to obtain a preprocessed forest multispectral remote sensing image; S130, performing feature encoding on the preprocessed forest multispectral remote sensing image to obtain a forest multispectral remote sensing global image feature matrix; S140, performing feature enhancement on the forest multispectral remote sensing global image feature matrix to obtain a forest multispectral remote sensing global feature enhancement matrix; S150, obtaining a result of whether the target forest monitoring area is invaded by pests and diseases based on the forest multispectral remote sensing global feature enhancement matrix.

[0019] Specifically, in the technical solution of the present application, it is first necessary to obtain a multispectral remote sensing image of the target forest monitoring area. It should be understood that the multispectral remote sensing image contains spectral reflectance data of different bands (such as visible light, near infrared and short-wave infrared), and the health of vegetation can be reflected by analyzing it. Different vegetation types (including healthy trees and trees attacked by pests and diseases) have different reflectances in these bands. Healthy plants usually have a higher reflectivity in the near-infrared band, while plants attacked by pests and diseases will show changes in reflectivity. Multispectral remote sensing images also contain texture information such as the spatial distribution, shape, and size of trees. Certain pests and diseases can cause discoloration, shedding or morphological changes in tree leaves, and these changes will be shown in specific texture patterns on remote sensing images. In general, the multispectral remote sensing images of the target forest monitoring area contain rich spectral information and spatial information, which are closely related to the occurrence and development of pests and diseases. By deeply analyzing this information, intelligent monitoring and early warning of forest pests and diseases can be achieved.

[0020] In an embodiment of the present application, one possible implementation method for obtaining multispectral remote sensing images of a target forest monitoring area may be: utilizing earth observation satellites, such as the Landsat series, the Sentinel series, etc. These satellites are equipped with multispectral sensors and are capable of obtaining large-scale, high-resolution multispectral remote sensing images. Satellite remote sensing has the advantages of wide coverage and short data acquisition cycle, and is suitable for monitoring large-scale forest areas. According to actual needs, the scope of the forest area to be monitored is clearly defined, including latitude and longitude coordinates, area and other information. Multispectral remote sensing image data of the target forest monitoring area can be obtained from these remote sensing platforms.

[0021] Considering that in the process of data acquisition, remote sensing satellites are affected by various factors such as the rotation of the earth, satellite orbit, curvature of the earth, atmospheric refraction, etc., which will cause geometric distortion of the acquired images, such as image distortion, displacement and tilt, etc., these distortions will seriously affect the accuracy of subsequent analysis of the image; and the atmosphere has different absorption and scattering effects on electromagnetic radiation of different wavelengths, which will cause atmospheric effects in remote sensing images, such as atmospheric scattering and absorption, which will change the reflectivity and radiation brightness of surface objects, thereby affecting the quality of multispectral remote sensing images. Therefore, in the technical solution of this application, it is necessary to perform image preprocessing on the multispectral remote sensing images of the target forest monitoring area to obtain preprocessed forest multispectral remote sensing images. Preprocessing can reduce problems such as geometric distortion and atmospheric effects in the image, improve the quality of multispectral remote sensing images of the target forest monitoring area, make the preprocessed images closer to the actual situation of surface objects, and provide a more accurate and reliable data basis for subsequent pest and disease analysis.

[0022] In an embodiment of the present application, a possible implementation method for performing image preprocessing on the multispectral remote sensing image of the target forest monitoring area to obtain a preprocessed forest multispectral remote sensing image can be: selecting a suitable geometric correction model to process the original image based on the terrain characteristics of the target monitored forest area and the characteristics of the collected remote sensing image. The polynomial model is suitable for areas with relatively flat terrain, while the rational function model is more suitable for complex terrain; a method based on the radiation transfer model can be selected to perform atmospheric correction processing on the original image.

[0023] Figure 2 The present invention is a flowchart of performing feature encoding on the pre-processed forest multispectral remote sensing image to obtain a forest multispectral remote sensing global image feature matrix in the forest pest intelligent monitoring and early warning method according to the embodiment of the present application. Figure 2 As shown, the feature encoding of the preprocessed forest multispectral remote sensing image to obtain the forest multispectral remote sensing global image feature matrix includes: S131, respectively performing feature extraction on the image blocks after the preprocessed forest multispectral remote sensing image is divided to obtain multiple forest multispectral remote sensing local feature matrices; S132, splicing the multiple forest multispectral remote sensing local feature matrices into the forest multispectral remote sensing global image feature matrix.

[0024] Figure 3 The flowchart is a method for intelligent monitoring and early warning of forest pests and diseases according to an embodiment of the present application, wherein feature extraction is performed on the image blocks of the pre-processed forest multispectral remote sensing image to obtain multiple forest multispectral remote sensing local feature matrices. Figure 3As shown, the feature extraction is performed on the image blocks after the preprocessed forest multispectral remote sensing image is divided into blocks to obtain multiple forest multispectral remote sensing local feature matrices, including: S1311, dividing the preprocessed forest multispectral remote sensing image into blocks to obtain a sequence of forest multispectral remote sensing local image blocks, wherein the sizes of each forest multispectral remote sensing local image block in the sequence of forest multispectral remote sensing local image blocks are equal; S1312, inputting each forest multispectral remote sensing local image block in the sequence of forest multispectral remote sensing local image blocks into a forest remote sensing image feature encoder to obtain the multiple forest multispectral remote sensing local feature matrices.

[0025] Considering that the original complete remote sensing image usually contains a large number of pixels and data, directly processing the entire image requires huge computing resources. In order to decompose the large-scale computing task into multiple small-scale subtasks, thereby reducing the computational complexity of a single processing and improving the processing efficiency, the pre-processed forest multispectral remote sensing image can be processed in blocks to obtain a sequence of forest multispectral remote sensing local image blocks, wherein each forest multispectral remote sensing local image block in the sequence of forest multispectral remote sensing local image blocks is of equal size. It is also worth noting that the manifestation of forest diseases and insect pests in remote sensing images is often local and subtle. By dividing the image into blocks, by processing each image block, it is possible to focus more on the local area of ​​the image, thereby extracting more detailed and accurate forest features, which are then used to determine whether it is suffering from diseases and insect pests, which is conducive to improving the accuracy of disease and insect pest invasion identification.

[0026] Each forest multispectral remote sensing local image block contains the spectral information of the trees in the local area. These spectral information reflect the reflection, absorption and radiation characteristics of the trees in different bands, and also contain the texture information of the trees in the local area. This information is very important for analyzing the pest and disease invasion of the trees. In order to extract feature information useful for pest and disease analysis from the original image block, it is necessary to input each forest multispectral remote sensing local image block in the sequence of the forest multispectral remote sensing local image blocks into the forest remote sensing image feature encoder to obtain multiple forest multispectral remote sensing local feature matrices. The feature information processed by the forest remote sensing image feature encoder is more representative and discriminative than the original pixel value, which is helpful for subsequent pest and disease identification. The forest remote sensing image feature encoder in this application is a convolutional neural network model including a deep and shallow feature fusion module. In the shallow part of the convolutional neural network, low-level visual features such as edges, corners, and textures are mainly extracted from forest satellite remote sensing images. These features describe the local details and patterns of the image, and are important for the preliminary judgment of the basic morphology and health of the forest. As the number of network layers increases, the convolutional neural network can learn more abstract and advanced feature representations. In the deep part, semantic information in forest satellite remote sensing images is extracted. That is, the shallow features provide local details and texture information of local image blocks of forest multispectral remote sensing, while the deep features capture the semantics and abstract information of local image blocks of forest multispectral remote sensing. The fusion of the two can fully utilize the advantages of features at different levels and improve the accuracy and robustness of pest and disease identification.

[0027] In particular, considering that the process of encoding and processing each forest multispectral remote sensing local image block in the sequence of the forest multispectral remote sensing local image block by the forest remote sensing image feature encoder is the same, the present application takes the processing of a local image block as an example. In an embodiment of the present application, a possible implementation method of inputting the forest multispectral remote sensing local image block into the forest remote sensing image feature encoder to obtain the forest multispectral remote sensing local feature matrix may be: extracting the forest multispectral remote sensing shallow feature matrix from the i-th layer of the forest remote sensing image feature encoder, the i-th layer being the first to sixth layers of the forest remote sensing image feature encoder; extracting the forest multispectral remote sensing deep feature matrix from the j-th layer of the forest remote sensing image feature encoder, the ratio between the j-th layer and the i-th layer being greater than or equal to 5; using the deep and shallow feature fusion module to fuse the forest multispectral remote sensing shallow feature matrix and the forest multispectral remote sensing deep feature matrix to obtain the forest multispectral remote sensing local feature matrix.

[0028] Considering that the original forest multispectral remote sensing image is processed in blocks is to reduce the computational complexity and enhance the local feature extraction capability. However, each image block is only a part of the original image, and processing these image blocks individually will lose the spatial coherence and contextual information between images. In order to reconstruct the global representation of the original image in the feature space, thereby maintaining the spatial coherence and contextual information of the image, in the technical solution of the present application, the multiple forest multispectral remote sensing local feature matrices are spliced ​​into a forest multispectral remote sensing global image feature matrix. It should be understood that by splicing local feature matrices to form a global feature matrix, information from the entire monitoring area can be comprehensively utilized to improve the accuracy of pest and disease monitoring and early warning. The global feature matrix contains more comprehensive and rich feature information, which can more accurately reflect the growth status and pest and disease conditions of the forest.

[0029] Considering that the forest multispectral remote sensing global image feature matrix contains the information of the entire monitoring area, the local features therein may not be able to fully express the long-range dependencies between trees. In order to further capture these long-range dependencies and improve the richness and accuracy of feature representation, in the technical solution of the present application, the forest multispectral remote sensing global image feature matrix is ​​passed through a non-local neural network to obtain a forest multispectral remote sensing global feature enhancement matrix. Those of ordinary skill in the art should understand that a non-local neural network is a neural network architecture that can capture long-range dependencies in an image or sequence. It is inspired by the non-local mean algorithm, and directly captures long-range dependencies by calculating the similarity between any two positions in an image or sequence, and using these similarities as weights for weighted summation. Non-local neural networks do not rely on repeated operations (such as convolution or loop) within a local neighborhood, but can consider the global information of the entire input data at one time. Through the processing of non-local neural networks, the features in the forest multispectral remote sensing global image feature matrix can be enhanced, making the feature representation more robust and discriminative.

[0030] Figure 4 The flowchart is a method for intelligent monitoring and early warning of forest pests and diseases according to an embodiment of the present application, which obtains the result of whether the target forest monitoring area is invaded by pests and diseases based on the forest multispectral remote sensing global feature enhancement matrix. Figure 4As shown, the result of obtaining whether the target forest monitoring area is invaded by pests and diseases based on the forest multispectral remote sensing global feature enhancement matrix includes: S151, based on the feature deviation information between the forest multispectral remote sensing global image feature matrix and the forest multispectral remote sensing global feature enhancement matrix, the forest multispectral remote sensing global feature enhancement matrix is ​​subjected to shift information feature compensation to obtain an optimized forest multispectral remote sensing global feature enhancement matrix; S152, the optimized forest multispectral remote sensing global feature enhancement matrix is ​​passed through a pest and disease detection classifier to obtain a classification result, and the classification result is used to indicate whether the target forest monitoring area is invaded by pests and diseases.

[0031] In particular, in the technical solution of the present application, the original multispectral remote sensing image is first converted into a global feature matrix, then feature enhancement is performed through a non-local neural network, and finally a pest and disease detection classifier is used for classification judgment. The purpose of feature enhancement is to extract more significant and highly distinguishable features in order to better identify and classify targets. Non-local neural networks usually enhance features by considering the relationship between pixels at different positions in the image. This network structure is able to capture long-distance dependencies in the image. Then, during the feature enhancement process, the non-local neural network may overemphasize certain features, causing the center of the feature distribution to shift. This shift may be due to the network's over-response to certain features, or because the network has learned some features that are not related to pests and diseases. Feature enhancement may introduce noise or irrelevant information, which interferes with the purity of the feature, causing the useful information in the feature matrix to be mixed with the noise information, thereby affecting the performance of the pest and disease detection classifier. That is, the forest multispectral remote sensing global feature enhancement matrix has a feature distribution shift deviation relative to the forest multispectral remote sensing global image feature matrix, which causes the forest multispectral remote sensing global feature enhancement matrix to have an extreme category offset when performing classification judgment through a classifier, resulting in a decrease in the accuracy of the classification result. In order to solve this problem, in the technical solution of the present application, based on the feature deviation information between the forest multispectral remote sensing global image feature matrix and the forest multispectral remote sensing global feature enhancement matrix, the forest multispectral remote sensing global feature enhancement matrix is ​​compensated for the shift information characteristics to obtain an optimized forest multispectral remote sensing global feature enhancement matrix.

[0032] Specifically, based on the feature deviation information between the forest multispectral remote sensing global image feature matrix and the forest multispectral remote sensing global feature enhancement matrix, the forest multispectral remote sensing global feature enhancement matrix is ​​subjected to shift information feature compensation to obtain an optimized forest multispectral remote sensing global feature enhancement matrix, including: calculating a shift difference representation matrix between the forest multispectral remote sensing global image feature matrix and the forest multispectral remote sensing global feature enhancement matrix; performing probabilistic activation function on each eigenvalue of the shift difference representation matrix to obtain a probabilistic shift difference representation matrix; inputting the forest multispectral remote sensing global feature enhancement matrix into the pest and disease detection classifier to obtain a probabilistic shift difference representation matrix. Obtain an initial classification probability value; determine a backpropagation compensation representation matrix based on a comparison between the eigenvalues ​​of each position of the probabilistic shifted difference representation matrix and the initial classification probability value and the weight matrix of the pest and disease detection classifier; calculate the exponential function values ​​of each eigenvalue of the backpropagation compensation representation matrix to obtain a backpropagation compensation class representation matrix; calculate the position-by-position point multiplication between the backpropagation compensation class representation matrix and the forest multispectral remote sensing global feature enhancement matrix to obtain a point multiplication result matrix, and calculate the weighted sum between the point multiplication result matrix and the forest multispectral remote sensing global feature enhancement matrix to obtain the optimized forest multispectral remote sensing global feature enhancement matrix.

[0033] In an embodiment of the present application, based on the feature deviation information between the forest multispectral remote sensing global image feature matrix and the forest multispectral remote sensing global feature enhancement matrix, another expression of performing shift information feature compensation on the forest multispectral remote sensing global feature enhancement matrix to obtain an optimized forest multispectral remote sensing global feature enhancement matrix is: performing operation on the forest multispectral remote sensing global feature enhancement matrix using the following formula to obtain the optimized forest multispectral remote sensing global feature enhancement matrix; wherein the formula is:

[0034] in, represents the normalized exponential function, represents the forest multispectral remote sensing global image feature matrix, represents the forest multispectral remote sensing global feature enhancement matrix, It means subtracting by position. represents the probabilistic shift difference representation matrix, and denote the first weighted hyperparameter and the second weighted hyperparameter respectively, represents the weight matrix of the pest detection classifier, represents the eigenvalues ​​of the shifted difference representation matrix, represents the initial classification probability value, It means point multiplication by position. represents matrix multiplication, represents the natural exponential function, represents the symbolic function, Represents the optimized forest multispectral remote sensing global feature enhancement matrix.

[0035] That is, in view of the distribution shift deviation problem between feature matrices, the forest multispectral remote sensing global feature enhancement matrix is ​​compensated for the shift information feature based on the feature deviation information between the forest multispectral remote sensing global image feature matrix and the forest multispectral remote sensing global feature enhancement matrix. The difference between the forest multispectral remote sensing global image feature matrix and the forest multispectral remote sensing global feature enhancement matrix is ​​calculated to obtain a shift difference representation matrix to quantify the deviation of feature distribution. The shift difference representation matrix reveals the alignment of the two feature matrices in the feature space, providing a basis for subsequent compensation. Then, the probability activation function is applied to each eigenvalue of the shift difference representation matrix for probabilistic processing, so that the difference value is converted into a probability distribution, so that the model can understand the deviation between features in a probabilistic form. Then, the original forest multispectral remote sensing global feature enhancement matrix is ​​input into the pest detection classifier to obtain the initial classification probability value, wherein the initial classification probability value provides the model with a basic classification performance indicator before feature compensation. Next, the back propagation compensation representation matrix is ​​determined by combining the probabilistic shift difference representation matrix and the initial classification probability value, as well as the weight matrix of the pest and disease detection classifier. That is, the feature matrix is ​​compensated by using the model's own weight and bias information to reduce the classification error. Next, the exponential function value is calculated for each eigenvalue of the back propagation compensation representation matrix, thereby enhancing the expression range of the eigenvalue and providing a mechanism for enlarging or reducing the feature compensation. Finally, the optimized forest multispectral remote sensing global feature enhancement matrix is ​​obtained by calculating the point product result between the back propagation compensation class representation matrix and the forest multispectral remote sensing global feature enhancement matrix, and then calculating the weighted sum with the original forest multispectral remote sensing global feature enhancement matrix. It should be understood that the point product and weighted sum operations combine the compensation information and the original feature information to generate an adjusted feature matrix, which aims to reduce the extreme deviation of the category and improve the accuracy of classification. The entire data processing process achieves the goal of optimizing the feature representation from identifying the feature distribution shift deviation to compensating for these deviations. This process not only improves the adaptability of the pest detection classifier to changes in feature distribution, but also enhances the model's ability to distinguish different categories. Through the shift information feature compensation mechanism, the model can more accurately capture the subtle differences between feature matrices, thereby achieving higher accuracy and better generalization performance in classification tasks.

[0036] Finally, the optimized forest multispectral remote sensing global feature enhancement matrix is ​​passed through the pest detection classifier to obtain a classification result, and the classification result is used to indicate whether the target forest monitoring area is invaded by pests and diseases. It should be understood that the optimized forest multispectral remote sensing global feature enhancement matrix contains the enhanced features obtained after a series of processing of the forest multispectral remote sensing image. These features not only cover the basic morphology and texture of the forest, but also integrate the spectral features in the multispectral data and the global dependencies captured by the non-local neural network. These features together constitute a comprehensive description of the growth status of the trees in the target forest monitoring area. The pest detection classifier is a model based on machine learning, which can automatically determine whether the target forest monitoring area is invaded by pests and diseases based on the input feature matrix. The classification results can provide important decision support for forestry managers. According to the classification results, managers can timely understand the growth status and pest and disease conditions of the trees and formulate corresponding prevention and control strategies and measures.

[0037] In an embodiment of the present application, the optimized forest multispectral remote sensing global feature enhancement matrix is ​​passed through a pest and disease detection classifier to obtain a classification result, and the classification result is used to indicate whether there is pest and disease invasion in the target forest monitoring area. One possible implementation method may be: expanding the optimized forest multispectral remote sensing global feature enhancement matrix into a one-dimensional feature vector according to a row vector or a column vector to obtain a forest pest and disease monitoring classification feature vector; using the fully connected layer of the pest and disease detection classifier to fully connect encode the forest pest and disease monitoring classification feature vector to obtain an encoded classification feature vector; inputting the encoded classification feature vector into the Softmax classification function of the pest and disease detection classifier to obtain the classification result.

[0038] In summary, the intelligent monitoring and early warning method for forest pests and diseases based on the embodiment of the present application is explained, which uses artificial intelligence-based image processing technology to analyze the multispectral remote sensing images in the target forest monitoring area in a high-dimensional feature space to identify whether the target forest monitoring area is invaded by pests and diseases. In this way, large-scale and rapid intelligent monitoring of forest pests and diseases can be achieved, overcoming the problem of low efficiency of traditional ground survey methods.

[0039] Figure 5 : is a system block diagram of the intelligent monitoring and early warning system for forest pests and diseases according to an embodiment of the present application. Figure 5As shown, according to the embodiment of the present application, the intelligent monitoring and early warning system 100 for forest pests and diseases includes: a forest image data acquisition module 110, which is used to obtain a multispectral remote sensing image of a target forest monitoring area; a forest image preprocessing module 120, which is used to perform image preprocessing on the multispectral remote sensing image of the target forest monitoring area to obtain a preprocessed forest multispectral remote sensing image; a forest image feature encoding module 130, which is used to perform feature encoding on the preprocessed forest multispectral remote sensing image to obtain a forest multispectral remote sensing global image feature matrix; a forest image feature enhancement module 140, which is used to perform feature enhancement on the forest multispectral remote sensing global image feature matrix to obtain a forest multispectral remote sensing global feature enhancement matrix; a forest pest and disease monitoring result generation module 150, which is used to obtain a result of whether the target forest monitoring area is invaded by pests and diseases based on the forest multispectral remote sensing global feature enhancement matrix.

[0040] Here, those skilled in the art will appreciate that the specific functions and operations of the various units and modules in the above-mentioned forest pest intelligent monitoring and early warning system 100 have been described in detail above. Figures 1 to 4 The invention has been introduced in detail in the description of the intelligent monitoring and early warning method for forest pests and diseases, and therefore, its repeated description will be omitted.

[0041] In summary, the forest pest intelligent monitoring and early warning system 100 based on the embodiment of the present application is explained, which adopts the image processing technology based on artificial intelligence, and analyzes the multi-spectral remote sensing image in the target forest monitoring area in a high-dimensional feature space to identify whether the target forest monitoring area is invaded by pests and diseases. In this way, large-scale and rapid intelligent monitoring of forest pests and diseases can be achieved, overcoming the problem of low efficiency of traditional ground survey methods.

[0042] As described above, the intelligent monitoring and early warning system 100 for forest pests and diseases according to the embodiment of the present application can be implemented in various wireless terminals, such as a server for intelligent monitoring and early warning of forest pests and diseases. In one example, the intelligent monitoring and early warning system 100 for forest pests and diseases according to the embodiment of the present application can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the intelligent monitoring and early warning system 100 for forest pests and diseases can be a software module in the operating system of the wireless terminal, or can be an application developed for the wireless terminal; of course, the intelligent monitoring and early warning system 100 for forest pests and diseases can also be one of the many hardware modules of the wireless terminal.

[0043] Alternatively, in another example, the intelligent monitoring and early warning system 100 for forest pests and diseases and the wireless terminal may also be separate devices, and the intelligent monitoring and early warning system 100 for forest pests and diseases may be connected to the wireless terminal via a wired and / or wireless network, and transmit interactive information in accordance with an agreed data format.

[0044] The above is only a preferred embodiment of the present application and does not constitute any form of limitation to the present application. Although the present application has been disclosed as a preferred embodiment as above, it is not intended to limit the present application. Any technician familiar with the profession can make some changes or modifications to equivalent embodiments of the technical contents disclosed above without departing from the scope of the technical solution of the present application. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present application without departing from the content of the technical solution of the present application still falls within the scope of the technical solution of the present application.

Claims

1. A method for intelligent monitoring and early warning of forest pests and diseases, characterized in that: include: Acquire multispectral remote sensing images of the target forest monitoring area; Performing image preprocessing on the multispectral remote sensing image of the target forest monitoring area to obtain a preprocessed forest multispectral remote sensing image; Performing feature encoding on the preprocessed forest multispectral remote sensing image to obtain a forest multispectral remote sensing global image feature matrix; Performing feature enhancement on the forest multispectral remote sensing global image feature matrix to obtain a forest multispectral remote sensing global feature enhancement matrix; Based on the forest multispectral remote sensing global feature enhancement matrix, the result of whether the target forest monitoring area is invaded by pests and diseases is obtained.

2. The intelligent monitoring and early warning method for forest pests and diseases according to claim 1 is characterized in that: The preprocessed forest multispectral remote sensing image is feature encoded to obtain a forest multispectral remote sensing global image feature matrix, including: Performing feature extraction on the image blocks of the preprocessed forest multispectral remote sensing image to obtain a plurality of forest multispectral remote sensing local feature matrices; The multiple forest multispectral remote sensing local feature matrices are spliced ​​into the forest multispectral remote sensing global image feature matrix.

3. The intelligent monitoring and early warning method for forest pests and diseases according to claim 2 is characterized in that: Feature extraction is performed on the image blocks of the preprocessed forest multispectral remote sensing image to obtain multiple forest multispectral remote sensing local feature matrices, including: The preprocessed forest multispectral remote sensing image is divided into blocks to obtain a sequence of forest multispectral remote sensing local image blocks, wherein the sizes of the forest multispectral remote sensing local image blocks in the sequence of forest multispectral remote sensing local image blocks are equal; Each forest multispectral remote sensing local image block in the sequence of forest multispectral remote sensing local image blocks is input into a forest remote sensing image feature encoder to obtain the plurality of forest multispectral remote sensing local feature matrices.

4. The intelligent monitoring and early warning method for forest pests and diseases according to claim 3 is characterized in that: The feature enhancement is performed on the forest multispectral remote sensing global image feature matrix to obtain the forest multispectral remote sensing global feature enhancement matrix, including: the forest multispectral remote sensing global image feature matrix is ​​passed through a non-local neural network to obtain the forest multispectral remote sensing global feature enhancement matrix.

5. The intelligent monitoring and early warning method for forest pests and diseases according to claim 4 is characterized in that: The forest remote sensing image feature encoder is a convolutional neural network model that includes a deep and shallow feature fusion module.

6. The intelligent monitoring and early warning method for forest pests and diseases according to claim 5 is characterized in that: Based on the forest multispectral remote sensing global feature enhancement matrix, the result of whether the target forest monitoring area is invaded by pests and diseases is obtained, including: Based on the feature deviation information between the forest multispectral remote sensing global image feature matrix and the forest multispectral remote sensing global feature enhancement matrix, the forest multispectral remote sensing global feature enhancement matrix is ​​subjected to shift information feature compensation to obtain an optimized forest multispectral remote sensing global feature enhancement matrix; The optimized forest multispectral remote sensing global feature enhancement matrix is ​​passed through a pest and disease detection classifier to obtain a classification result, and the classification result is used to indicate whether the target forest monitoring area is invaded by pests and diseases.

7. The intelligent monitoring and early warning method for forest pests and diseases according to claim 6 is characterized in that: Based on the feature deviation information between the forest multispectral remote sensing global image feature matrix and the forest multispectral remote sensing global feature enhancement matrix, the forest multispectral remote sensing global feature enhancement matrix is ​​subjected to shift information feature compensation to obtain an optimized forest multispectral remote sensing global feature enhancement matrix, including: Calculating a shift difference representation matrix between the forest multispectral remote sensing global image feature matrix and the forest multispectral remote sensing global feature enhancement matrix; Probabilize each eigenvalue of the shifted difference representation matrix based on a probability activation function to obtain a probabilized shifted difference representation matrix; Inputting the forest multispectral remote sensing global feature enhancement matrix into the pest detection classifier to obtain an initial classification probability value; Determine a back-propagation compensation representation matrix based on a comparison between the eigenvalues ​​of each position of the probabilistic shift difference representation matrix and the initial classification probability value and a weight matrix of the pest detection classifier; Calculating the exponential function value of each eigenvalue of the back-propagation compensation representation matrix to obtain a back-propagation compensation class representation matrix; The position-based point multiplication between the back-propagation compensation class representation matrix and the forest multispectral remote sensing global feature enhancement matrix is ​​calculated to obtain a point product result matrix, and the weighted sum between the point product result matrix and the forest multispectral remote sensing global feature enhancement matrix is ​​calculated to obtain the optimized forest multispectral remote sensing global feature enhancement matrix.

8. An intelligent monitoring and early warning system for forest pests and diseases, characterized in that: include: The forest image data acquisition module is used to obtain multispectral remote sensing images of the target forest monitoring area; A forest image preprocessing module, used for performing image preprocessing on the multispectral remote sensing image of the target forest monitoring area to obtain a preprocessed forest multispectral remote sensing image; A forest image feature encoding module is used to perform feature encoding on the preprocessed forest multispectral remote sensing image to obtain a forest multispectral remote sensing global image feature matrix; A forest image feature enhancement module, used for performing feature enhancement on the forest multispectral remote sensing global image feature matrix to obtain a forest multispectral remote sensing global feature enhancement matrix; The forest pest and disease monitoring result generation module is used to obtain the result of whether the target forest monitoring area is invaded by pests and diseases based on the forest multi-spectral remote sensing global feature enhancement matrix.

9. The intelligent monitoring and early warning system for forest pests and diseases according to claim 8 is characterized in that: The forest image feature encoding module includes: A forest image block feature extraction unit is used to extract features from the image blocks of the preprocessed forest multispectral remote sensing image to obtain a plurality of forest multispectral remote sensing local feature matrices; The forest image block feature integration unit is used to splice the multiple forest multispectral remote sensing local feature matrices into the forest multispectral remote sensing global image feature matrix.

10. The intelligent monitoring and early warning system for forest pests and diseases according to claim 9 is characterized in that: The forest image block feature extraction unit comprises: The forest image blocking subunit is used to block the preprocessed forest multispectral remote sensing image to obtain a sequence of forest multispectral remote sensing local image blocks, wherein the sizes of the forest multispectral remote sensing local image blocks in the sequence of forest multispectral remote sensing local image blocks are equal; The forest image block feature generating subunit is used to input each forest multispectral remote sensing local image block in the sequence of forest multispectral remote sensing local image blocks into the forest remote sensing image feature encoder to obtain the multiple forest multispectral remote sensing local feature matrices.

Images