Analysis method of shelterbelt degradation degree based on UAV multispectral data

Abstract

The present application relates to the field of image data processing technology, and specifically discloses a method for analyzing the degree of degradation of shelterbelts based on multispectral data of unmanned aerial vehicles, which uses multispectral unmanned aerial vehicles to collect multispectral images of shelterbelts in sample plots, and constructs a multi-source feature matrix of multispectral images by calculating the spectral reflectance, vegetation index and texture index of each local area in the multispectral images of shelterbelts in sample plots. Then, by performing deep learning and interactive fusion processing on the multi-source features of the multispectral images, the comprehensive characterization information of the vegetation status of the shelterbelts in sample plots is comprehensively obtained, and the intelligent decoding estimation of the degradation rate of the shelterbelts in sample plots is realized accordingly, and according to the degradation rate of the shelterbelts in the sample plots, a risk degradation area is constructed to accurately control the risk degradation area. In this way, the accuracy and efficiency of monitoring the degree of degradation of shelterbelts can be effectively improved, thereby providing more scientific and accurate decision support for forest resource management.

Description

Technical Field

[0001] The present application relates to the field of image data processing technology, and more specifically, to a method for analyzing the degree of protection forest degradation based on unmanned aerial vehicle multispectral data. Background Art

[0002] Shelterbelts play an irreplaceable role in resisting wind and sand, conserving water and soil, and improving the ecological environment. However, with the impact of global climate change and human activities, the health of shelterbelts has been challenged unprecedentedly. Specifically, due to natural factors (such as drought, pests and diseases) and human factors (such as overexploitation and irrational use), shelterbelts in many areas have experienced varying degrees of degradation. This not only weakens the function of shelterbelts, but also poses a threat to regional ecological security.

[0003] In forest resource management, monitoring and assessment of shelterbelt degradation is a key link in ensuring the sustainable use of forest resources. Traditional shelterbelt monitoring methods mostly rely on ground surveys and remote sensing satellite data. The former is time-consuming and labor-intensive and difficult to achieve large-scale coverage. Although the latter can provide vegetation information from a macro perspective, its spatial resolution is limited and cannot accurately reflect the detailed information of the shelterbelt vegetation status. In addition, satellite data acquisition is greatly restricted by weather conditions, and cloud cover will affect the quality and availability of data.

[0004] In recent years, the rapid development of drone technology has brought new solutions to this problem. Drones have the advantages of flexibility, low cost, and no terrain restrictions. They can carry multispectral sensors for high-frequency and high-resolution data collection, providing new possibilities for shelterbelt monitoring. Therefore, a shelterbelt degradation degree analysis method based on drone multispectral data is expected. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a method for analyzing the degree of degradation of shelterbelts based on multispectral data of unmanned aerial vehicles, which uses multispectral unmanned aerial vehicles to collect multispectral images of shelterbelts in sample plots, and constructs a multi-source feature matrix of multispectral images by calculating the spectral reflectance, vegetation index and texture index of each local area in the multispectral image of the shelterbelts in the sample plots. Then, by performing deep learning and interactive fusion processing on the multi-source features of the multispectral images, the comprehensive characterization information of the vegetation status of the shelterbelts in the sample plots is comprehensively obtained, and intelligent decoding estimation of the degradation rate of the shelterbelts in the sample plots is realized accordingly. In this way, the accuracy and efficiency of monitoring the degree of degradation of shelterbelts can be effectively improved, thereby providing more scientific and accurate decision support for forest resource management.

[0006] Accordingly, according to one aspect of the present application, a method for analyzing the degree of protection forest degradation based on multispectral data of unmanned aerial vehicles is provided, which includes:

[0007] Use multispectral drones to collect multispectral images of the shelterbelts in the sample plots;

[0008] Calculating the spectral reflectance of each area in the multispectral image of the sample plot shelterbelt to obtain a spectral reflectance distribution matrix of the sample plot shelterbelt;

[0009] Calculating the vegetation index of each area in the multispectral image of the sample plot shelterbelt to obtain a vegetation index distribution matrix of the sample plot shelterbelt;

[0010] Calculating the texture index of each area in the multispectral image of the sample plot shelterbelt to obtain a texture index distribution matrix of the sample plot shelterbelt;

[0011] The degradation rate of the sample plot protection forest is determined based on the spectral reflectance distribution matrix of the sample plot protection forest, the vegetation index distribution matrix of the sample plot protection forest and the texture index distribution matrix of the sample plot protection forest.

[0012] Compared with the prior art, the method for analyzing the degree of degradation of shelterbelts based on multispectral data of unmanned aerial vehicles provided in this application uses multispectral unmanned aerial vehicles to collect multispectral images of shelterbelts in sample plots, and constructs a multi-source feature matrix of multispectral images by calculating the spectral reflectance, vegetation index and texture index of each local area in the multispectral images of shelterbelts in sample plots. Then, the multi-source features of multispectral images are subjected to deep learning and interactive fusion processing to comprehensively obtain the comprehensive characterization information of the vegetation status of shelterbelts in sample plots, and accordingly realize the intelligent decoding estimation of the degradation rate of shelterbelts in sample plots. In this way, the accuracy and efficiency of monitoring the degree of degradation of shelterbelts can be effectively improved, thereby providing more scientific and accurate decision support for forest resource management. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] 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.

[0014] Figure 1 This is a flow chart of a method for analyzing the degree of protection forest degradation based on drone multispectral data according to an embodiment of the present application.

[0015] Figure 2 This is a flowchart of step S150 in the method for analyzing the degree of protection forest degradation based on drone multispectral data according to an embodiment of the present application.

[0016] Figure 3 Schematic diagram of data flow in step S150 of the method for analyzing the degree of protection forest degradation based on multispectral data of unmanned aerial vehicle according to an embodiment of the present application.

[0017] Figure 4 This is a flowchart of step S152 in the method for analyzing the degree of protection forest degradation based on drone multispectral data according to an embodiment of the present application. DETAILED DESCRIPTION

[0018] Below, the embodiments of the present application will be described in more detail in conjunction with the accompanying drawings, and the above and other purposes, features and advantages of the present application will become more apparent. 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 example embodiments described herein.

[0019] Figure 1 Flow chart of the method for analyzing the degree of protection forest degradation based on multispectral data of unmanned aerial vehicles according to an embodiment of the present application. Figure 1 As shown, according to the embodiment of the present application, the method for analyzing the degree of degradation of the shelterbelt based on multispectral data of unmanned aerial vehicles includes the following steps: S110, using a multispectral unmanned aerial vehicle to collect multispectral images of the sample plot shelterbelt; S120, calculating the spectral reflectance of each area in the multispectral image of the sample plot shelterbelt to obtain a spectral reflectance distribution matrix of the sample plot shelterbelt; S130, calculating the vegetation index of each area in the multispectral image of the sample plot shelterbelt to obtain a vegetation index distribution matrix of the sample plot shelterbelt; S140, calculating the texture index of each area in the multispectral image of the sample plot shelterbelt to obtain a texture index distribution matrix of the sample plot shelterbelt; S150, determining the degradation rate of the sample plot shelterbelt based on the spectral reflectance distribution matrix of the sample plot shelterbelt, the vegetation index distribution matrix of the sample plot shelterbelt and the texture index distribution matrix of the sample plot shelterbelt.

[0020] In the above-mentioned method for analyzing the degree of degradation of shelterbelts based on multispectral data of unmanned aerial vehicles, the step S110 uses a multispectral unmanned aerial vehicle to collect multispectral images of the shelterbelts in the sample plot. It should be understood that compared with satellite remote sensing technology, unmanned aerial vehicles can provide image data with higher spatial resolution, thereby obtaining more detailed ground information, which is particularly suitable for small sample plots or areas that require fine management. At the same time, by carrying multispectral sensors, unmanned aerial vehicles can capture the surface reflectance information of shelterbelts in different bands (such as visible light, near infrared, red edge, etc.), thereby reflecting the different physiological states of vegetation, thereby providing strong support for the analysis of the degree of degradation of shelterbelts.

[0021] Specifically, the implementation method of using a multispectral drone to collect multispectral images of the shelterbelts in the sample plot is as follows: First, use a multispectral drone to photograph the shelterbelts in the sample plot. In the embodiment of the present application, a 3M multispectral drone is selected, and the drone includes visible light, red light channel, near infrared channel, red edge channel and other bands. During data collection, representative sample plots are selected to ensure that the ecological characteristics of the sample plots can reflect the overall health of the shelterbelts. After the equipment is deployed, use the DJI GO software to set the route, set the route overlap and lateral overlap to more than 80%, the flight altitude to 60-80 meters, the flight speed to 3.5-4.0m / s, and fly under suitable meteorological conditions to avoid strong winds and strong sunlight to ensure the integrity of the orthophoto and the availability of the data.

[0022] After the drone shooting is completed, Agisoft Metashape software is used in the subsequent internal processing to perform image stitching and orthophoto processing. Specifically, orthophoto stitching is a complex and delicate process that aims to generate high-quality, high-precision geospatial data from the original images collected by the drone. First, the images acquired by the drone are imported into a dedicated software environment and a preliminary data check is performed to ensure that all image file formats are compatible and metadata are complete, while confirming that the image coverage meets the mission requirements and evaluating the overlap between images to ensure the smooth progress of subsequent stitching.

[0023] After the image is imported and the preliminary inspection is completed, the next step is radiometric calibration. Due to differences in the characteristics of drone sensors and changes in lighting conditions during shooting, there may be inconsistencies between images acquired at different time points or with different devices. Through radiometric calibration, the influence of these factors can be eliminated and the digital values ​​in the image can be converted into physical quantities - that is, ground reflectivity. This step not only improves the consistency between images, but also provides real spectral information for subsequent analysis. Radiometric calibration usually relies on auxiliary materials such as pre-calibrated sensor response curves and standard reflectors measured synchronously on site.

[0024] Next comes geometric correction, which is a process of adjusting the spatial position of the original image using ground control points (GCPs) and image registration algorithms. The instability of drone flight and the limitations of the sensor itself may cause varying degrees of spatial distortion in the image. By accurately measuring the selected GCPs and marking them on the image, combined with appropriate mathematical models, these distortions can be effectively corrected so that the image coordinate system corresponds to the actual geographic location. Good geometric correction improves the positioning accuracy of the image and lays the foundation for large-scale stitching.

[0025] Atmospheric correction is to remove the influence of atmospheric scattering and absorption on the image spectral signal, thereby improving the accuracy of surface reflectance estimation. Changes in atmospheric conditions will cause the reflectance of the same object in the same band to be different, so it is necessary to use an atmospheric radiation transfer model (such as the 6S model) for correction. This process involves the selection and setting of multiple parameters, such as atmosphere type, water vapor content, aerosol optical depth, etc. Correct atmospheric correction helps to reveal the true characteristics of the surface, which is particularly important for monitoring vegetation health.

[0026] Considering that the earth's surface is not completely flat, images taken in areas with complex terrain such as mountainous areas may produce shadows, stretching and other phenomena due to the undulating terrain. Therefore, in order to eliminate such effects, it is necessary to introduce a digital elevation model (DEM) to make additional adjustments to the image based on factors such as terrain slope and aspect, namely terrain correction, to ensure that the generated orthophotos can maintain high geometric accuracy even under undulating terrain conditions. Images after terrain correction not only have better visual effects, but are also more reliable for quantitative analysis.

[0027] After completing the above corrections, the stitching and seamless processing stage begins. The task at this point is to stitch each single image into a complete orthophoto. This includes automatically finding points of the same name, building a transformation matrix, and implementing mosaic operations. In order to avoid obvious seams or color mutations at the stitching boundaries, seamless processing is also required, such as feathering the transition area, applying a gradient filter, and adjusting the color balance to optimize the connection between adjacent images. In addition, the useless parts of the image edge should be cropped, the effective information area should be retained, and high-quality orthophotos should be output to provide accurate basic data for subsequent analysis.

[0028] In the above-mentioned shelterbelt degradation degree analysis method based on multispectral data of unmanned aerial vehicle, the step S120 calculates the spectral reflectance of each area in the multispectral image of the sample plot shelterbelt to obtain the spectral reflectance distribution matrix of the sample plot shelterbelt. It should be understood that the spectral reflectance is the proportion of solar radiation reflected by vegetation in different bands, which can directly reflect important information such as the health status, growth stage, moisture content, chlorophyll concentration, etc. of the vegetation. Therefore, the present application first calculates the spectral reflectance of each area in the multispectral image of the sample plot shelterbelt to reveal the spectral characteristics of the vegetation, and arranges and combines the obtained spectral reflectances of each area according to the regional position to form a spectral reflectance distribution matrix of the sample plot shelterbelt.

[0029] In an embodiment of the present application, the spectral reflectance of a selected area in an orthophoto is calculated by using ArcGIS Pro software. First, load the processed multispectral image data, and ensure that the image has completed preprocessing such as radiation calibration and atmospheric correction. Next, use the "Extract by Region" tool or the "Raster Calculator" tool to select an area of ​​interest (such as a shelterbelt plot or a specific forest stand), and limit the spectral band data to the target area by setting the corresponding mask area. Then, by setting the raster statistical parameters or extracting the pixel value, extract the spectral reflectance of different bands in the selected area. Among them, the spectral reflectance includes: the reflectance of the near-infrared band, the reflectance of the red band, the reflectance of the red edge band, and the reflectance of the green band. To ensure the accuracy of the reflectance, orthophotos are often used for registration, and geometric correction is performed in combination with existing ground control points to further improve the spatial accuracy of the data.

[0030] In the above-mentioned shelterbelt degradation degree analysis method based on multispectral data of unmanned aerial vehicles, the step S130 calculates the vegetation index of each area in the multispectral image of the shelterbelt of the sample plot to obtain the vegetation index distribution matrix of the shelterbelt of the sample plot. That is, based on the spectral reflectance data of each area obtained above, the vegetation index of each area in the multispectral image of the shelterbelt of the sample plot is further calculated by a specific mathematical formula to quantify the vegetation health and vitality status of each area, so as to obtain the vegetation index distribution matrix of the shelterbelt of the sample plot.

[0031] In an embodiment of the present application, based on the spectral reflectance data obtained above, NDVI (Normalized Difference Vegetation Index), NDWI (Normalized Difference Water Index), OSAVI (Optimized Soil-Adjusted Vegetation Index) and DVI (Difference Vegetation Index) are calculated. Among them, NDVI is a widely used vegetation health indicator, which evaluates the growth status of vegetation by comparing the reflectance difference between the near-infrared band and the red band. The value range of NDVI is from -1 to +1, where a positive value indicates the presence of vegetation, and the higher the value, the healthier the vegetation. NDWI is used to evaluate the moisture content in vegetation, and its calculation method is similar to NDVI, but uses the red edge band and the near-infrared band. OSAVI is a vegetation index optimized for soil background effects, which improves the accuracy of the vegetation index by adjusting the reflectance of the soil background. DVI is an index that evaluates vegetation growth by simply calculating the difference between the reflectance of the near-infrared band and the red band. By calculating the above vegetation index, we can objectively reflect the growth status, moisture content, influence of soil background and health of vegetation, which helps to accurately evaluate the key characteristics of vegetation health and degradation degree in the region, and provide solid data support for subsequent shelterbelt degradation assessment and related analysis work.

[0032] Specifically, the vegetation index calculation formula is as follows:

[0033] NDVI=(Nir-R) / (Nir+R)

[0034] NDWI=(Rededge-Nir) / (Rededge+Nir)

[0035] OSAVI=(Nir-Rededge) / (Nir+R+0.16)

[0036] DVI=Nir-R

[0037] Among them, Nir represents the reflectance of the near-infrared band, R represents the reflectance of the red band, Rededge represents the reflectance of the red edge band, NDVI represents the normalized difference vegetation index, NDWI represents the normalized difference water index, OSAVI represents the optimized soil adjusted vegetation index, and DVI represents the difference vegetation index.

[0038] In the above-mentioned shelterbelt degradation degree analysis method based on multispectral data of unmanned aerial vehicle, the step S140 calculates the texture index of each area in the multispectral image of the sample plot shelterbelt to obtain the sample plot shelterbelt texture index distribution matrix. It should be understood that texture features, as an important image feature, can reflect the roughness, uniformity and other characteristics of the vegetation coverage area in the image, thereby revealing the structure and morphological information of the vegetation. Therefore, the present application further calculates the texture index of each area in the multispectral image of the sample plot shelterbelt to obtain the sample plot shelterbelt texture index distribution matrix, thereby providing another dimension of reference information for the degradation degree analysis of the sample plot shelterbelt.

[0039] In an embodiment of the present application, ENVI software is used to calculate the texture index of a selected area in a multispectral image of the sample plot shelterbelt. Specifically, first, multispectral image data that has been radiometrically calibrated and atmospherically corrected is imported, and the spatial resolution of the image data is ensured to be suitable for texture analysis. Next, an area of ​​interest (such as a shelterbelt sample plot or a specific forest stand) is selected, and the area is delineated using the "ROI (Region of Interest)" tool in ENVI. Then, enter the "Texture Analysis" module, select an appropriate texture extraction method, such as gray level co-occurrence matrix (GLCM), local binary pattern (LBP) or waveform transformation, etc., to calculate texture features. In an embodiment of the present application, the texture features include mean (ME), variance (VAR), homogeneity (HOM) and contrast (CON). ENVI software provides a variety of algorithms to calculate these features, and texture analysis of local areas is performed by setting different window sizes (0°, 45°, 90°, 135°) and step sizes (3×3, 9×9, 12×12, etc.). The extraction of texture features usually requires local statistical operations on the image. Therefore, different step sizes and directions can be selected to extract texture information according to the analysis requirements to highlight the texture features at different scales and directions in the image.

[0040] The texture feature calculation formula is as follows:

[0041]

[0042] in, represents the mean, is the weighting factor, and Represent different gray levels. is the gray level co-occurrence matrix The element at position indicates the gray value and The probability that pixels appear at the same specific direction and distance at the same time, is the number of gray levels, represents the variance, Represents the average gray value of all pixels. Indicates homogeneity, Indicates contrast.

[0043] In the above-mentioned shelterbelt degradation degree analysis method based on multi-spectral data of unmanned aerial vehicles, the step S150 determines the degradation rate of the shelterbelt in the sample plot based on the spectral reflectance distribution matrix of the shelterbelt in the sample plot, the vegetation index distribution matrix of the shelterbelt in the sample plot, and the texture index distribution matrix of the shelterbelt in the sample plot. That is, in order to comprehensively consider the various information of the shelterbelt in the sample plot (spectral characteristics, vegetation health, surface texture) to achieve a comprehensive assessment of the degree of degradation of the shelterbelt, the present application further comprehensively analyzes the spectral reflectance distribution matrix of the shelterbelt in the sample plot, the vegetation index distribution matrix of the shelterbelt in the sample plot, and the texture index distribution matrix of the shelterbelt in the sample plot, so as to achieve an accurate assessment of the degree of degradation of the shelterbelt in the sample plot based on the various characteristic information of the shelterbelt.

[0044] Figure 2 This is a flowchart of step S150 in the method for analyzing the degree of protection forest degradation based on drone multispectral data according to an embodiment of the present application. Figure 3 FIG. 1 is a schematic diagram of data flow in step S150 of the method for analyzing the degree of protection forest degradation based on multispectral data of unmanned aerial vehicles according to an embodiment of the present application. Figure 2 and Figure 3 As shown, the step S150 includes: S151, respectively performing inter-regional semantic association coding on the sample plot protection forest spectral reflectance distribution matrix, the sample plot protection forest vegetation index distribution matrix and the sample plot protection forest texture index distribution matrix to obtain the sample plot protection forest spectral reflectance implicit association feature vector, the sample plot protection forest vegetation index implicit association feature vector and the sample plot protection forest texture index implicit association feature vector; S152, performing attention interaction based on external knowledge on the sample plot protection forest spectral reflectance implicit association feature vector and the sample plot protection forest vegetation index implicit association feature vector to obtain the sample plot protection forest spectral reflectance-vegetation index joint implicit association feature vector; S153, performing attention interaction based on external knowledge on the sample plot protection forest spectral reflectance-vegetation index joint implicit association feature vector and the sample plot protection forest texture index implicit association feature vector to obtain the sample plot protection forest multi-source data implicit association feature vector; S154, determining the degradation rate of the sample plot protection forest based on the sample plot protection forest multi-source data implicit association feature vector.

[0045] Specifically, in step S151, the spectral reflectance distribution matrix of the sample plot shelterbelt, the vegetation index distribution matrix of the sample plot shelterbelt, and the texture index distribution matrix of the sample plot shelterbelt are respectively subjected to inter-regional semantic association coding to obtain implicit association feature vectors of the spectral reflectance of the sample plot shelterbelt, implicit association feature vectors of the vegetation index of the sample plot shelterbelt, and implicit association feature vectors of the texture index of the sample plot shelterbelt. That is, in order to effectively capture the vegetation spatial layout and internal structure of the sample plot shelterbelt and to facilitate a more comprehensive assessment of the health status of the shelterbelt, the present application further performs spatial association analysis on the spectral reflectance, vegetation index, and texture index between each region in the multispectral image of the sample plot shelterbelt.

[0046] In a specific example of the present application, the sample plot shelterbelt spectral reflectance distribution matrix, the sample plot shelterbelt vegetation index distribution matrix and the sample plot shelterbelt texture index distribution matrix are respectively input into the inter-regional semantic association encoder based on the hole convolutional neural network model to obtain the sample plot shelterbelt spectral reflectance implicit association feature vector, the sample plot shelterbelt vegetation index implicit association feature vector and the sample plot shelterbelt texture index implicit association feature vector. Those of ordinary skill in the art should know that the hole convolutional neural network model can expand the receptive field without increasing the amount of calculation by adjusting the size and step size of the convolution kernel, thereby capturing a wider range of spatial context association information. Based on this, by using the hole convolutional neural network model to extract the spatial correlation characteristics of each area of ​​the sample plot shelterbelt in terms of spectral reflectance, vegetation index and texture index, it is helpful to more accurately reflect the relationship between the various areas of the shelterbelt, thereby enhancing the ability to describe the degradation state of the shelterbelt.

[0047] Specifically, in step S152, the implicitly associated feature vector of the spectral reflectance of the sample plot shelterbelt and the implicitly associated feature vector of the vegetation index of the sample plot shelterbelt are subjected to attention interaction based on external knowledge to obtain the joint implicitly associated feature vector of the spectral reflectance-vegetation index of the sample plot shelterbelt. Here, considering that the spectral reflectance directly reflects the reflection characteristics of the shelterbelt in different bands, and the vegetation index further quantifies the health status of the vegetation on the basis of the spectral reflectance, the two have a stronger correlation. Therefore, in the technical solution of the present application, the implicitly associated feature vector of the spectral reflectance of the sample plot shelterbelt and the implicitly associated feature vector of the vegetation index of the sample plot shelterbelt are first subjected to interactive fusion processing. In particular, in order to improve the effect of information interaction and fusion, the present application proposes an attention interaction method based on external knowledge, which guides the information interaction between the spectral reflectance and the vegetation index by introducing external knowledge, such as vegetation physiological ecology knowledge, forest ecology principles, etc., so that the fused feature representation can more accurately reflect the comprehensive state of vegetation.

[0048] Figure 4 FIG. 1 is a flowchart of step S152 in the method for analyzing the degree of protection forest degradation based on multispectral data of unmanned aerial vehicles according to an embodiment of the present application. Figure 4 As shown, the step S152 includes: S1521, based on external knowledge, performing feature modulation attention optimization on the implicitly associated feature vector of the sample plot protection forest spectral reflectance and the implicitly associated feature vector of the sample plot protection forest vegetation index respectively to obtain the optimized implicitly associated feature vector of the sample plot protection forest spectral reflectance and the optimized implicitly associated feature vector of the sample plot protection forest vegetation index; S1522, performing position-by-position semantic interactive encoding on the optimized implicitly associated feature vector of the sample plot protection forest spectral reflectance and the optimized implicitly associated feature vector of the sample plot protection forest vegetation index to obtain the joint implicitly associated feature vector of the sample plot protection forest spectral reflectance-vegetation index.

[0049] In a specific example of the present application, the step S1521 includes: first, inputting the implicitly associated feature vector of the spectral reflectance of the sample plot shelterbelt and the implicitly associated feature vector of the vegetation index of the sample plot shelterbelt into a fine-grained feature interaction network to obtain a spectral reflectance-vegetation index fine-grained interaction feature matrix, which is expressed as follows:

[0050]

[0051] in, represents the implicit correlation eigenvector of the spectral reflectance of the shelterbelt in the sample plot, represents the implicitly associated eigenvector of the shelterbelt vegetation index of the plot, represents the transpose of a vector, represents the matrix multiplication operation, Represents the spectral reflectance-vegetation index fine-grained interaction feature matrix.

[0052] Here, the fine-grained feature interaction network is used to calculate the product between the implicit correlation feature vector of the spectral reflectance of the sample plot shelterbelt and the transposed vector of the implicit correlation feature vector of the vegetation index of the sample plot shelterbelt to obtain the spectral reflectance-vegetation index fine-grained interaction feature matrix. In this way, the fine-grained feature interaction network can understand the correlation between the spectral reflectance of the shelterbelt and the vegetation index at a microscopic level, capture the fine-grained interaction information between the two, and generate the spectral reflectance-vegetation index fine-grained interaction feature matrix.

[0053] Next, the spectral reflectance-vegetation index fine-grained interactive feature matrix is ​​input into the attention unit based on external knowledge to obtain the external knowledge optimized spectral reflectance-vegetation index fine-grained interactive feature matrix, which is expressed as follows:

[0054]

[0055] in, and represents the learnable memory parameter matrix of the external knowledge-based attention unit, represents the normalization function, Representing external knowledge to optimize the spectral reflectance-vegetation index fine-grained interaction feature matrix.

[0056] That is, external knowledge is used to further optimize the spectral reflectance-vegetation index fine-grained interaction feature matrix, thereby enhancing the model's ability to understand vegetation ecology knowledge. In specific implementations, external knowledge such as vegetation physiological ecology knowledge and forest ecology principles are used to train the parameter matrix of the attention unit to ensure that it can accurately identify and strengthen feature interaction patterns that are closely related to the health status of vegetation, and guide the attention unit to perform weighted optimization on the spectral reflectance-vegetation index fine-grained interaction feature matrix, thereby capturing the potential connection between the two in a larger range.

[0057] Then, based on the external knowledge optimization of the fine-grained interactive feature matrix of spectral reflectance and vegetation index, the implicitly associated feature vector of the spectral reflectance of the sample plot shelterbelt and the implicitly associated feature vector of the vegetation index of the sample plot shelterbelt are respectively subjected to feature modulation optimization to obtain the optimized implicitly associated feature vector of the spectral reflectance of the sample plot shelterbelt and the optimized implicitly associated feature vector of the vegetation index of the sample plot shelterbelt. More specifically, the implicitly associated feature vector of the spectral reflectance of the sample plot shelterbelt is linearly transformed to obtain a first query feature vector and a first value feature vector, and the fine-grained interactive feature matrix of spectral reflectance and vegetation index optimized by the external knowledge is used as a key matrix, and the first query feature vector, the first value feature vector and the key matrix are input into a fine-grained modulation module based on a Transformer structure to obtain the optimized implicitly associated feature vector of the spectral reflectance of the sample plot shelterbelt, which is expressed by the formula:

[0058]

[0059]

[0060]

[0061] in, and denote the first query embedding matrix and the first value embedding matrix respectively, and denote the first query feature vector and the first value feature vector respectively, and Represent different bias terms, represents the normalized exponential function, represents the characteristic scale value of the external knowledge optimized fine-grained feature interaction matrix, Represents the implicit correlation eigenvector of the optimized shelterbelt spectral reflectance.

[0062] At the same time, the implicitly associated feature vector of the sample plot shelterbelt vegetation index is linearly transformed to obtain a second query feature vector and a second value feature vector, and the external knowledge is used to optimize the spectral reflectance-vegetation index fine-grained interactive feature matrix as a key matrix, and the second query feature vector, the second value feature vector and the key matrix are input into the fine-grained modulation module based on the Transformer structure to obtain the optimized implicitly associated feature vector of the sample plot shelterbelt vegetation index, which is expressed as follows:

[0063]

[0064]

[0065]

[0066] in, and denote the second query embedding matrix and the second value embedding matrix respectively, and denote the second query feature vector and the second value feature vector respectively, and Represent different bias terms, Represents the implicit correlation eigenvector of the optimized shelterbelt vegetation index of the sample plot.

[0067] That is, the spectral reflectance-vegetation index fine-grained interactive feature matrix optimized by external knowledge is used as the key matrix, and the first query feature vector and the first value feature vector are constructed based on the implicitly associated feature vector of the spectral reflectance of the sample plot protection forest, and the second query feature vector and the second value feature vector are constructed based on the implicitly associated feature vector of the vegetation index of the sample plot protection forest. The self-attention mechanism of the Transformer structure is used to realize information exchange and integration between internal and external knowledge of the features, thereby ensuring that the implicitly associated feature vector of the spectral reflectance of the sample plot protection forest and the implicitly associated feature vector of the vegetation index of the sample plot protection forest can benefit from external knowledge to improve the quality of their feature expression.

[0068] In a specific example of the present application, the step S1522 includes: calculating the implicit correlation feature vector of the optimized plot shelterbelt spectral reflectance and the implicit correlation feature vector of the optimized plot shelterbelt vegetation index by dividing the location point by the implicit correlation feature vector to obtain the plot shelterbelt spectral reflectance-vegetation index joint implicit correlation feature vector, which is expressed by the formula:

[0069]

[0070] in, Represents the implicit association feature vector of the spectral reflectance and vegetation index of the shelterbelt of the sample plot. That is, the implicit association feature vector of the spectral reflectance of the optimized shelterbelt of the sample plot and the implicit association feature vector of the vegetation index of the optimized shelterbelt of the sample plot are semantically interactively encoded by position by dividing by position, so as to comprehensively reflect the complementary information of the shelterbelt in terms of spectral characteristics and vegetation health status.

[0071] Specifically, in step S153, the attention interaction based on external knowledge is performed on the joint implicit correlation feature vector of the spectral reflectance-vegetation index of the sample plot shelterbelt and the implicit correlation feature vector of the texture index of the sample plot shelterbelt to obtain the implicit correlation feature vector of the multi-source data of the sample plot shelterbelt. That is, the above-mentioned attention interaction method based on external knowledge is also used to perform interactive fusion processing on the joint implicit correlation feature vector of the spectral reflectance-forest vegetation index of the sample plot shelterbelt and the implicit correlation feature vector of the texture index of the sample plot shelterbelt, so as to further integrate the texture distribution characteristics of the sample plot shelterbelt, and obtain the implicit correlation feature vector of the multi-source data of the sample plot shelterbelt, thereby realizing a multi-dimensional and in-depth analysis of the degree of degradation of the shelterbelt, and providing richer and more accurate data support for the subsequent degradation degree analysis.

[0072] Specifically, the step S154 determines the degradation rate of the sample plot shelterbelt based on the implicitly associated feature vector of the multi-source data of the sample plot shelterbelt. In a specific example of the present application, the implicitly associated feature vector of the multi-source data of the sample plot shelterbelt is input into a degradation rate estimation module based on a decoder to obtain the degradation rate of the sample plot shelterbelt. Specifically, the decoder adopts a multi-layer perceptron structure, which can extract and integrate the implicitly associated feature vector of the multi-source data of the sample plot shelterbelt layer by layer to achieve an accurate estimation of the degree of degradation of the shelterbelt. Specifically, the decoder first performs preliminary feature learning on the input multi-source data implicitly associated feature vector through a fully connected layer, and then gradually refines the feature representation through multiple hidden layers, and finally maps the feature representation extracted layer by layer to the numerical space of the degradation rate through a linear transformation in the output layer, thereby obtaining the estimated degradation rate of the sample plot shelterbelt. In this way, the multi-dimensional information of the shelterbelt can be effectively converted into an intuitive degradation rate indicator, thereby providing a scientific basis for forest resource management and protection.

[0073] In particular, in the technical solution of the present application, taking into account that the implicitly associated feature vector of the spectral reflectance of the sample plot shelterbelt, the implicitly associated feature vector of the vegetation index of the sample plot shelterbelt and the implicitly associated feature vector of the texture index of the sample plot shelterbelt respectively represent the semantic association characteristics of the position distribution of the spectral reflectance of the sample plot shelterbelt, the semantic association characteristics of the position distribution of the vegetation index of the sample plot shelterbelt and the semantic association characteristics of the position distribution of the texture index of the sample plot shelterbelt, when performing fine-grained feature interaction based on external knowledge modulation based on a dual structure, the two calls of external knowledge and the semantic differences in calls will cause the implicitly associated feature vector of the multi-source data of the sample plot shelterbelt to have a difference in the spatial structure of the semantic interaction fusion distribution of the multi-source data, affecting the decoding regression convergence consistency, thereby affecting the accuracy of the degradation rate of the sample plot shelterbelt obtained by the degradation rate estimation module based on the decoder.

[0074] Therefore, in a preferred example of the present application, the implicit correlation feature vector of the multi-source data of the sample plot shelterbelt is optimized, including:

[0075] First, the absolute value sum of all eigenvalues ​​of the implicit association eigenvectors of the multi-source data of the shelterbelt in the sample plot is calculated to obtain the first implicit association spatial structure value of the multi-source data of the shelterbelt in the sample plot, and the square root of the square sum is calculated to obtain the second implicit association spatial structure value of the multi-source data of the shelterbelt in the sample plot, that is:

[0076]

[0077]

[0078] in, Indicates the implicit association feature vector of the multi-source data of the sample plot shelterbelt The eigenvalues ​​at the positions, Indicates taking the absolute value, is the characteristic scale value of the implicit correlation characteristic vector of the multi-source data of the shelterbelt in the sample plot, Indicates the implicit correlation spatial structure value of the multi-source data of the first type of shelterbelt. It represents the implicit correlation spatial structure value of the multi-source data of the shelterbelt in the second sample plot;

[0079] Next, for each eigenvalue of the implicit association eigenvector of the sample plot shelterbelt multi-source data, the implicit association long-range dependency value of the first sample plot shelterbelt multi-source data is obtained by subtracting the product of the eigenvalue and the total number of eigenvalues ​​of the implicit association eigenvector of the sample plot shelterbelt multi-source data from the implicit association spatial structure value of the first sample plot shelterbelt data, that is:

[0080] in, Indicates the implicit association feature vector of the multi-source data of the sample plot shelterbelt The long-range dependency value of the first-level implicit correlation of the multi-source data of the protective forest corresponding to the characteristic value of each position;

[0081] At the same time, the implicit association long-range dependency value of the second sample plot shelterbelt multi-source data is obtained by calculating the product of the square root of the total number of eigenvalues ​​multiplied by the eigenvalue minus the implicit association spatial structure value of the second sample plot shelterbelt multi-source data, that is:

[0082] in, Indicates the implicit association feature vector of the multi-source data of the sample plot shelterbelt The long-range dependency value of the implicit association of the multi-source data of the shelterbelt in the second sample plot corresponding to the characteristic value of each position;

[0083] Secondly, the index value calculated by taking the implicit long-range dependency value of the first sample plot shelterbelt multi-source data as the exponent of the natural constant and the inverse of the implicit long-range dependency value of the second sample plot shelterbelt multi-source data are weighted summed to obtain the optimized eigenvalue corresponding to each eigenvalue, that is:

[0084]

[0085] in, Indicates the implicit association feature vector of the multi-source data of the sample plot shelterbelt The optimized eigenvalue corresponding to the eigenvalue at each position;

[0086] Furthermore, the optimized eigenvalues ​​are combined into an optimized implicitly associated eigenvector of the multi-source data of the sample plot protection forest.

[0087] Then, the optimized implicitly associated feature vector of the multi-source data of the sample plot protection forest is passed through a degradation rate estimation module based on a decoder to obtain the degradation rate of the sample plot protection forest.

[0088] Therefore, in view of the possible spatial structure loss in the high-dimensional space of the feature set of the implicitly associated feature vector of the sample plot protection forest multi-source data, which leads to the convergence inconsistency of the decoding regression based on the feature-based implicit inference of the spatial structure information, a long-distance feature dependency relationship is established based on the overall feature scale of the implicitly associated feature vector of the sample plot protection forest multi-source data relative to the spatial structure representation of the implicitly associated feature vector of the sample plot protection forest multi-source data, so as to establish the feature local connectivity of the implicitly associated feature vector of the sample plot protection forest multi-source data, and capture the spatial ambiguous information of the object feature value through the unstructured feature value point prediction of the implicitly associated feature vector of the sample plot protection forest multi-source data, thereby improving the spatial inductive bias perception ability of the feature set of the implicitly associated feature vector of the sample plot protection forest multi-source data, improving the convergence consistency of the decoding regression, and improving the accuracy of the degradation rate of the sample plot protection forest obtained by the degradation rate estimation module based on the decoder.

[0089] Furthermore, in an embodiment of the present application, the method further includes: constructing a risk degradation area according to the degradation rate of the shelterbelt in the sample plot, so as to accurately control the risk degradation area. Specifically, the risk degradation area in the shelterbelt can be determined based on the comparison between the degradation rate and the threshold. For example, if the degradation rate of a certain area in the shelterbelt is lower than the threshold, it means that the area is a risk degradation area. Further, the method may also include: generating corresponding management and disposal methods according to the degradation state of the risk degradation area. For example, the degradation state of the risk degradation area may include but is not limited to the degradation state of vegetation growth, the degradation state of leaf color change, the degradation state of pests and diseases, the degradation state of death, the degradation state of soil quality, and the degradation state of water source depletion. For each degradation state, a corresponding management and disposal method can be generated to achieve precise control. For example, for the degraded state of vegetation growth, the control and disposal methods can be: water and fertilizer management or branch pruning, etc. Water and fertilizer management can be to appropriately increase the supply of water and fertilizer to improve soil conditions, and branch pruning can be to appropriately prune trees and remove sick and weak branches to promote tree recovery. There is no limitation here.

[0090] In summary, the method for analyzing the degree of degradation of the shelterbelt based on multispectral data of drones according to the embodiment of the present application is explained, which uses a multispectral drone to collect multispectral images of the shelterbelts of the sample plots, and constructs a multi-source feature matrix of the multispectral images by calculating the spectral reflectance, vegetation index and texture index of each local area in the multispectral images of the shelterbelts of the sample plots, and then, through deep learning and interactive fusion processing of the multi-source features of the multispectral images, the comprehensive characterization information of the vegetation status of the shelterbelts of the sample plots is comprehensively obtained, and the intelligent decoding estimation of the degradation rate of the shelterbelts of the sample plots is realized accordingly. In this way, the accuracy and efficiency of monitoring the degree of degradation of the shelterbelts can be effectively improved, thereby providing more scientific and accurate decision support for forest resource management.

[0091] Furthermore, the embodiment of the present application also provides a system for analyzing the degree of protection forest degradation based on multispectral data of unmanned aerial vehicles, the system comprising:

[0092] An image acquisition module, used to collect multispectral images of the shelterbelt of the sample plot using a multispectral drone;

[0093] A reflectance calculation module, used to calculate the spectral reflectance of each area in the multispectral image of the sample plot shelterbelt to obtain a spectral reflectance distribution matrix of the sample plot shelterbelt;

[0094] A vegetation index calculation module, used to calculate the vegetation index of each area in the multispectral image of the sample plot shelterbelt to obtain a vegetation index distribution matrix of the sample plot shelterbelt;

[0095] A texture index calculation module, used to calculate the texture index of each area in the multispectral image of the sample plot shelterbelt to obtain a texture index distribution matrix of the sample plot shelterbelt;

[0096] A degradation rate calculation module, used to determine the degradation rate of the sample plot shelterbelt based on the sample plot shelterbelt spectral reflectance distribution matrix, the sample plot shelterbelt vegetation index distribution matrix and the sample plot shelterbelt texture index distribution matrix;

[0097] The degradation area management and control module is used to construct a risk degradation area according to the degradation rate of the shelterbelt in the sample plot, so as to accurately control the risk degradation area. The degradation area management and control module is used to generate corresponding management and disposal methods according to the degradation state of the risk degradation area. For example, the degradation state of the risk degradation area may include but is not limited to the degradation state of vegetation growth, the degradation state of leaf color change, the degradation state of pests and diseases, the degradation state of death, the degradation state of soil quality, and the degradation state of water source depletion. For each degradation state, a corresponding management and disposal method can be generated to achieve precise management and control. For example, for the degradation state of vegetation growth, the management and disposal methods may be: water and fertilizer management or pruning branches, etc. Water and fertilizer management may be to appropriately increase the supply of water and fertilizer to improve soil conditions, and pruning branches may be to appropriately prune trees and remove sick and weak branches to promote tree recovery. There is no limitation here.

[0098] The specific principles or functions of the shelterbelt degradation degree analysis system based on drone multispectral data in the embodiment of the present application can refer to the shelterbelt degradation degree analysis method based on drone multispectral data in the above-mentioned embodiment, and will not be repeated here.

[0099] The basic principle of the present invention is described above in conjunction with specific embodiments. However, it should be pointed out that the advantages, strengths, effects, etc. mentioned in the present invention are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. must be possessed by each embodiment of the present invention. In addition, the specific details of the above embodiments are only for the purpose of illustration and facilitation of understanding, rather than limitation, and the above details do not limit the present invention to being implemented by adopting the above specific details.

Claims

1. A method for analyzing the degree of protection forest degradation based on multispectral data of unmanned aerial vehicles, characterized in that: include: Use multispectral drones to collect multispectral images of the shelterbelts in the sample plots; Calculating the spectral reflectance of each area in the multispectral image of the sample plot shelterbelt to obtain a spectral reflectance distribution matrix of the sample plot shelterbelt; Calculating the vegetation index of each area in the multispectral image of the sample plot shelterbelt to obtain a vegetation index distribution matrix of the sample plot shelterbelt; Calculating the texture index of each area in the multispectral image of the sample plot shelterbelt to obtain a texture index distribution matrix of the sample plot shelterbelt; Determining the degradation rate of the sample plot shelterbelt based on the sample plot shelterbelt spectral reflectance distribution matrix, the sample plot shelterbelt vegetation index distribution matrix and the sample plot shelterbelt texture index distribution matrix; Determining the degradation rate of the sample plot shelterbelt based on the sample plot shelterbelt spectral reflectance distribution matrix, the sample plot shelterbelt vegetation index distribution matrix and the sample plot shelterbelt texture index distribution matrix includes: Performing inter-regional semantic association coding on the sample plot shelterbelt spectral reflectance distribution matrix, the sample plot shelterbelt vegetation index distribution matrix and the sample plot shelterbelt texture index distribution matrix respectively to obtain the sample plot shelterbelt spectral reflectance implicit association feature vector, the sample plot shelterbelt vegetation index implicit association feature vector and the sample plot shelterbelt texture index implicit association feature vector; Performing attention interaction based on external knowledge on the implicit correlation feature vector of the spectral reflectance of the shelterbelt forest of the sample plot and the implicit correlation feature vector of the vegetation index of the shelterbelt forest of the sample plot to obtain a joint implicit correlation feature vector of the spectral reflectance of the shelterbelt forest of the sample plot and the vegetation index; Performing attention interaction based on external knowledge on the combined implicit correlation feature vector of the sample plot shelterbelt spectral reflectance-vegetation index and the implicit correlation feature vector of the sample plot shelterbelt texture index to obtain the implicit correlation feature vector of the sample plot shelterbelt multi-source data; Determining the degradation rate of the shelterbelt of the sample plot based on the implicitly associated feature vector of the multi-source data of the shelterbelt of the sample plot; Performing attention interaction based on external knowledge on the implicit correlation feature vector of the spectral reflectance of the sample plot shelterbelt and the implicit correlation feature vector of the vegetation index of the sample plot shelterbelt to obtain a joint implicit correlation feature vector of the spectral reflectance of the sample plot shelterbelt and the vegetation index, including: Based on external knowledge, feature modulation attention optimization is performed on the implicit correlation feature vector of the spectral reflectance of the shelterbelt of the sample plot and the implicit correlation feature vector of the vegetation index of the shelterbelt of the sample plot to obtain an optimized implicit correlation feature vector of the spectral reflectance of the shelterbelt of the sample plot and an optimized implicit correlation feature vector of the vegetation index of the shelterbelt of the sample plot; Performing position-by-position semantic interactive encoding on the optimized implicitly associated feature vector of the spectral reflectance of the sample plot shelterbelt and the optimized implicitly associated feature vector of the vegetation index of the sample plot shelterbelt to obtain a joint implicitly associated feature vector of the spectral reflectance of the sample plot shelterbelt and the vegetation index; Based on external knowledge, the implicitly associated feature vector of the spectral reflectance of the sample plot shelterbelt and the implicitly associated feature vector of the vegetation index of the sample plot shelterbelt are respectively subjected to feature modulation attention optimization to obtain an optimized implicitly associated feature vector of the spectral reflectance of the sample plot shelterbelt and an optimized implicitly associated feature vector of the vegetation index of the sample plot shelterbelt, including: Inputting the implicitly associated feature vector of the spectral reflectance of the sample plot shelterbelt and the implicitly associated feature vector of the vegetation index of the sample plot shelterbelt into a fine-grained feature interaction network to obtain a spectral reflectance-vegetation index fine-grained interaction feature matrix; Inputting the spectral reflectance-vegetation index fine-grained interactive feature matrix into an attention unit based on external knowledge to obtain an external knowledge optimized spectral reflectance-vegetation index fine-grained interactive feature matrix; Based on the external knowledge, the spectral reflectance-vegetation index fine-grained interactive feature matrix is ​​optimized, and the implicit correlation feature vector of the spectral reflectance of the sample plot protection forest and the implicit correlation feature vector of the vegetation index of the sample plot protection forest are respectively subjected to feature modulation optimization to obtain the optimized implicit correlation feature vector of the spectral reflectance of the sample plot protection forest and the optimized implicit correlation feature vector of the vegetation index of the sample plot protection forest.

2. The method for analyzing the degree of protection forest degradation based on multispectral data of unmanned aerial vehicles according to claim 1 is characterized in that: The sample plot shelterbelt spectral reflectance distribution matrix, the sample plot shelterbelt vegetation index distribution matrix and the sample plot shelterbelt texture index distribution matrix are respectively subjected to inter-regional semantic association coding to obtain the sample plot shelterbelt spectral reflectance implicit association feature vector, the sample plot shelterbelt vegetation index implicit association feature vector and the sample plot shelterbelt texture index implicit association feature vector, including: The sample plot protection forest spectral reflectance distribution matrix, the sample plot protection forest vegetation index distribution matrix and the sample plot protection forest texture index distribution matrix are respectively input into the inter-regional semantic association encoder based on the void convolutional neural network model to obtain the sample plot protection forest spectral reflectance implicit association feature vector, the sample plot protection forest vegetation index implicit association feature vector and the sample plot protection forest texture index implicit association feature vector.

3. The method for analyzing the degree of protection forest degradation based on multispectral data of unmanned aerial vehicles according to claim 2 is characterized in that: The fine-grained feature interaction network is used to calculate the product between the implicitly associated feature vector of the sample plot shelterbelt spectral reflectance and the transposed vector of the implicitly associated feature vector of the sample plot shelterbelt vegetation index to obtain the spectral reflectance-vegetation index fine-grained interaction feature matrix.

4. The method for analyzing the degree of protection forest degradation based on multispectral data of unmanned aerial vehicles according to claim 3 is characterized in that: Based on the external knowledge, the spectral reflectance-vegetation index fine-grained interactive feature matrix is ​​optimized, and the implicit correlation feature vector of the spectral reflectance of the sample plot shelterbelt and the implicit correlation feature vector of the vegetation index of the sample plot shelterbelt are respectively subjected to feature modulation optimization to obtain the optimized implicit correlation feature vector of the spectral reflectance of the sample plot shelterbelt and the optimized implicit correlation feature vector of the vegetation index of the sample plot shelterbelt, including: Performing a linear transformation on the implicitly associated feature vector of the spectral reflectance of the shelterbelt forest of the sample plot to obtain a first query feature vector and a first value feature vector, and using the external knowledge to optimize the fine-grained interactive feature matrix of spectral reflectance-vegetation index as a key matrix, inputting the first query feature vector, the first value feature vector and the key matrix into a fine-grained modulation module based on a Transformer structure to obtain the optimized implicitly associated feature vector of the spectral reflectance of the shelterbelt forest of the sample plot; The implicitly associated eigenvector of the sample plot shelterbelt vegetation index is linearly transformed to obtain a second query eigenvector and a second value eigenvector, and the spectral reflectance-vegetation index fine-grained interactive feature matrix optimized by the external knowledge is used as a key matrix. The second query eigenvector, the second value eigenvector and the key matrix are input into the Transformer-based fine-grained modulation module to obtain the optimized implicitly associated eigenvector of the sample plot shelterbelt vegetation index.

5. The method for analyzing the degree of protection forest degradation based on multispectral data of unmanned aerial vehicles according to claim 4 is characterized in that: The optimized implicitly associated feature vector of the spectral reflectance of the sample plot shelterbelt and the optimized implicitly associated feature vector of the vegetation index of the sample plot shelterbelt are semantically interactively encoded position by position to obtain the combined implicitly associated feature vector of the spectral reflectance of the sample plot shelterbelt and the vegetation index, including: The optimized implicit correlation feature vector of the sample plot protection forest spectral reflectance and the optimized implicit correlation feature vector of the sample plot protection forest vegetation index are calculated and divided by the location points to obtain the combined implicit correlation feature vector of the sample plot protection forest spectral reflectance and vegetation index.

6. The method for analyzing the degree of protection forest degradation based on multispectral data of unmanned aerial vehicles according to claim 5 is characterized in that: Determining the degradation rate of the shelterbelt in the sample plot based on the implicitly associated feature vector of the multi-source data of the shelterbelt in the sample plot includes: The implicitly associated feature vector of the multi-source data of the sample plot protection forest is input into a degradation rate estimation module based on a decoder to obtain the degradation rate of the sample plot protection forest.

7. The method for analyzing the degree of protection forest degradation based on multispectral data of unmanned aerial vehicles according to claim 6 is characterized in that: The method also includes: constructing a risk degradation area according to the degradation rate of the protective forest in the sample plot to accurately control the risk degradation area.

Images