Remote sensing image processing method and system based on ai and computer vision fusion

Abstract

The present application relates to the technical field of image analysis, in particular to a remote sensing image processing method and system based on AI and computer vision fusion, comprising the following steps: based on input multispectral image data, analyzing the spectral distribution data of the image, grouping and numerical operation on the multispectral channel matrix, extracting the distribution characteristic value and the explicit and implicit parameters, completing matrix decomposition through matrix calculation and classification operation, and generating spectral sparse matrix eigenvalue. In the present application, through spectral distribution analysis and channel matrix numerical operation on multispectral image data, deeper spectral and implicit features can be mined, the analysis accuracy is improved, the spectral and spatial characteristics are effectively decoupled by combining the spatial position matrix, the accuracy and efficiency of feature extraction are improved, the characteristic distribution expression ability is optimized through the screening and weighted enhancement of explicit and implicit characteristics, the target recognition accuracy is improved, the dynamic mapping of spectral and spatial characteristics accurately captures the changes of the ground surface, and the stability and accuracy of classification are enhanced.

Description

Technical Field

[0001] The present invention relates to the field of image analysis technology, and in particular to a remote sensing image processing method and system based on the fusion of AI and computer vision. Background Art

[0002] The field of remote sensing image processing technology involves analyzing and processing image data acquired from remote sensing equipment using methods such as computer vision, image processing, and artificial intelligence. The core of this technology involves decoding, feature extraction, classification, segmentation, and enhancement of remote sensing images to obtain target information and specific features contained in the images, thereby achieving comprehensive perception and analysis of surface information. Remote sensing image processing technology encompasses image data preprocessing, geometric correction, spectral analysis, and automated interpretation of image content, encompassing a complete technical chain from data acquisition, storage, and transmission to subsequent in-depth analysis and application development.

[0003] Remote sensing image processing refers to the process of processing and parsing acquired remote sensing image data through specific image analysis and computational methods to extract target information. This method addresses technical issues such as geometric distortion correction, spectral feature analysis, and precise extraction of spatial information from remote sensing images. It primarily utilizes deep learning-based target recognition technology, region segmentation algorithms, and image registration methods, combined with the unique multispectral and high-resolution data characteristics of remote sensing images for feature modeling and data analysis, ensuring accurate processing of image information at the pixel or object level.

[0004] Existing technologies often suffer from information loss and inaccurate feature decoupling due to high complexity when analyzing spectral characteristics. Reliance on a single technology in feature extraction and classification limits accurate processing of complex areas, impacting precision and consistency. Traditional regional segmentation and image registration methods struggle to dynamically reflect subtle surface changes, impacting classification accuracy. Furthermore, inefficient methods for eliminating redundant information in geometric correction and spectral analysis increase computational complexity, reduce processing efficiency, and limit the practical effectiveness of image data in applications. Summary of the Invention

[0005] The purpose of the present invention is to solve the shortcomings of the existing technology and propose a remote sensing image processing method and system based on the fusion of AI and computer vision.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a remote sensing image processing method based on the fusion of AI and computer vision, comprising the following steps:

[0007] S1: Based on the input multispectral image data, the spectral distribution data of the image is analyzed, the multispectral channel matrix is ​​grouped and numerically operated, the distribution characteristic values ​​and explicit and implicit parameters are extracted, and the matrix decomposition is completed through matrix calculation and classification operations to generate the spectral sparse matrix eigenvalues;

[0008] S2: Based on the spectral sparse matrix eigenvalues, the spatial position matrix of the image is called to analyze the cross-distribution characteristics of the spectral channels and the position coordinates, calculate the characteristic values ​​within the partition, remove the redundant components and complete the normalization and feature extraction operations to generate the spectral and spatial feature decoupling values;

[0009] S3: Based on the spectral and spatial feature decoupling values, call the image feature set, analyze the multidimensional feature components, perform partition screening on the explicit and implicit feature values, perform weighting and enhancement operations on the distribution matrix of the feature components, reconstruct the extracted features through feature mapping, and generate feature dictionary enhancement values;

[0010] S4: Based on the feature dictionary enhancement value, call the regional pixel characteristics of each pixel point, reconstruct the distribution parameters and calculate the characteristic values ​​of the pixel points and the spectral matrix, normalize and map the parameter matrix, calculate the characteristic change distribution, and generate the spectral and spatial characteristic change matrix;

[0011] S5: Based on the spectral and spatial characteristic change matrix, the structural change matrix in the image is called, the distribution parameters of multiple partitions in the matrix are classified and compared with the matrix, the classification results are extracted by calculating the regional change relationship, the characteristic classification operation is completed, and the image characteristic classification matrix is ​​generated.

[0012] The spectral sparse matrix eigenvalues ​​include distribution characteristic values, explicit and implicit parameters, matrix grouping results, and matrix operation results. The spectral and spatial feature decoupling values ​​are specifically spectral partition characteristic values, spatial position matrix cross-distribution characteristic values, normalized characteristic values, and de-redundant characteristic values. The feature dictionary enhancement values ​​include characteristic component distribution matrix, explicit and implicit characteristic partition values, characteristic mapping reconstruction values, and weighted enhancement characteristic values. The spectral and spatial characteristic change matrix specifically refers to pixel point area distribution parameters, characteristic value change matrix, normalized parameter matrix, and characteristic distribution mapping matrix. The image characteristic classification matrix includes multi-partition distribution parameters, structural change matrix, regional change relationship classification results, and characteristic classification comparison results.

[0013] As a further solution of the present invention, the step of obtaining the eigenvalues ​​of the spectral sparse matrix is ​​specifically as follows:

[0014] S111: performing spectral distribution analysis on the input multispectral image data, analyzing the spectral distribution characteristics of multiple channels in the multispectral channel matrix, and generating a multispectral channel spectral characteristic matrix by calculating the relationship between the normalized spectral values ​​and the wavelength distribution among the multiple channels;

[0015] S112: Calculating a fitting matrix between channels based on the multi-spectral channel spectral characteristic matrix, performing standardization processing on the matrix, combining the explicit and implicit parameters of the multiple channels, and adjusting the weights of the matrix component values ​​according to the weight distribution rule of the characteristic parameters to generate an explicit and implicit weighted characteristic matrix;

[0016] S113: Performing singular value decomposition on the explicit and implicit weighted characteristic matrix, and combining the decomposed characteristic value matrix to screen the judgment relationship between the proportion of multiple characteristic values ​​and the threshold, using the formula:

[0017] ;

[0018] Calculate and generate spectral sparse matrix eigenvalues;

[0019] in, Representative The sparse eigenvalues ​​of the channel, Represents the channel in the explicit and implicit weighted characteristic matrix With channel The component value of Represents the channel in the explicit and implicit weighted characteristic matrix Corresponding to The square of the characteristic value, represents the explicit and implicit weight components, Represents the number of channels, Represents the total number of characteristic values.

[0020] As a further solution of the present invention, the step of obtaining the spectral and spatial feature decoupling value is specifically as follows:

[0021] S211: Based on the spectral sparse matrix eigenvalues, calling the spatial position matrix of the image, analyzing the cross-distribution characteristics of the spectral channels and the position coordinates, calculating the spatial position characteristic value corresponding to each spectral channel, and generating a spectral space cross-distribution matrix;

[0022] S212: Eliminating redundant components in the distribution matrix based on the partition characteristic values ​​of the spectral spatial cross-distribution matrix, and determining and adjusting normalization weights based on the frequency distribution characteristics of the characteristic values ​​and a normalization standard to generate a characteristic normalized distribution matrix;

[0023] S213: Perform cross-validation and parameter extraction on the characteristic components in the characteristic normalized distribution matrix, construct an explicit and implicit weight matrix for the multi-partition characteristic components, and calculate the normalized decoupling ratio of the components using the formula:

[0024] ;

[0025] Generate decoupled values ​​of spectral and spatial features;

[0026] in, Represents spectral channels and spatial coordinates The decoupled eigenvalue of Spectral channels in the representative characteristic normalized distribution matrix In quantity The value on Represents the components in the explicit and implicit weight matrix In spatial coordinates The weight value on is the characteristic adjustment coefficient, used to correct the ratio of deviation, Normalization adjustment parameters are used to control the numerical balance during the denominator normalization process. is the deviation correction value of the characteristic component, is the number of spectral components, is the number of coordinate components, Spectral channels in the representative characteristic normalized distribution matrix In spatial components The characteristic value on .

[0027] As a further solution of the present invention, the step of obtaining the feature dictionary enhancement value is specifically as follows:

[0028] S311: Based on the spectral and spatial feature decoupling values, call the associated image feature set, analyze the data structure and correlation degree in the multi-dimensional feature components, and generate a pre-processed explicit and implicit feature value matrix through mapping analysis and explicit and implicit comparison of the feature components;

[0029] S312: performing partition screening on the pre-processed explicit and implicit characteristic value matrix, based on the frequency distribution characteristics of the characteristic values, comparing the standard deviation and mean of the characteristic values ​​in the partition, and combining the validity judgment of the characteristic values ​​to eliminate abnormal values, thereby generating a filtered characteristic partition matrix;

[0030] S313: Perform weighting and enhancement operations on the filtered feature partition matrix, and calculate the dynamic weight coefficients of the feature components using the formula:

[0031] ;

[0032] Calculate the feature map to reconstruct the extracted features and generate feature dictionary enhancement values;

[0033] in, Representative feature partition and characteristic dimensions The enhancement value, Characteristic component The weight of Partitioning features Middle The value of the component, Characteristic component The mean of To avoid small positive numbers with zero denominator, is the total number of characteristic components.

[0034] As a further solution of the present invention, the steps of obtaining the spectrum and spatial characteristic change matrix are specifically as follows:

[0035] S411: Based on the feature dictionary enhancement value, calling the regional pixel characteristic data associated with each pixel point, parsing the multidimensional distribution information of the regional pixel characteristic data, extracting the statistical distribution parameters of the regional pixel characteristic data and calculating the correlation matrix value, reconstructing each dimension of the spectral matrix segment by segment using the distribution parameters, dynamically matching and comparing the pixel values ​​according to the reconstructed parameter matrix, and generating a preliminary characteristic value calculation matrix;

[0036] S412: performing normalization processing on the preliminary characteristic value calculation matrix, adjusting the distribution ratio by calculating the minimum and maximum ranges of the characteristic values, optimizing the characteristic value distribution weights in combination with the mapping rules of the characteristic value distribution, and obtaining a normalized parameter matrix;

[0037] S413: Extract characteristic change data from the normalized parameter matrix, perform dynamic analysis based on the fluctuation of distribution parameters, and calculate the standardized change distribution of characteristic values ​​of multiple pixel points using the formula:

[0038] ;

[0039] Generate spectral and spatial characteristic variation matrix;

[0040] in, Represents pixel points and characteristic dimensions The change value of is the first parameter in the normalized parameter matrix The pixel characteristic value, Characteristic component The mean of Characteristic component The standard deviation of is the total number of characteristic components.

[0041] As a further solution of the present invention, the steps of obtaining the image characteristic classification matrix are specifically as follows:

[0042] S511: Based on the spectral and spatial characteristic change matrix, the structural change matrix in the image is called, and a preliminary classification result matrix is ​​generated by comparing the distribution parameters of multiple partitions in the matrix and using characteristic comparison and difference calculation in the partition change relationship;

[0043] S512: Optimizing the data in the preliminary classification result matrix, adjusting the thresholds of the partition parameters and refining the classification rules, adjusting and classifying the distribution characteristic values ​​of multiple partitions, and obtaining an optimized classification result matrix;

[0044] S513: Performing a characteristic classification operation on the optimized classification result matrix, calculating the classification characteristic value and characteristic weight response value of each partition, and performing a comprehensive analysis in combination with the environmental parameter differences, using the formula:

[0045] ;

[0046] Calculate the comprehensive classification results of the characteristic values ​​and generate the image characteristic classification matrix;

[0047] in, Indicates the Partition No. The categorical value of the characteristic, For the Partition No. The response value of the parameter, For the The environmental parameter value of the characteristic, is the average value of environmental parameters in all partitions, is the total number of parameters.

[0048] A remote sensing image processing system based on the fusion of AI and computer vision, wherein the remote sensing image processing system based on the fusion of AI and computer vision is used to execute the above-mentioned remote sensing image processing method based on the fusion of AI and computer vision, and the system comprises:

[0049] The spectral characteristic analysis module analyzes the spectral distribution data based on the input multispectral image data, completes the grouping operation of the spectral channel matrix, extracts the distribution characteristic values ​​between channels, classifies the explicit and implicit parameters, decomposes the classified matrix data, and generates the spectral sparse matrix eigenvalues;

[0050] The spatial feature extraction module calls the spatial position matrix based on the spectral sparse matrix eigenvalues, analyzes the cross-distribution characteristics of the spectral channels and position coordinates, calculates the partition cross-characteristic values, eliminates redundant components, normalizes the spatial position characteristics, extracts the spectral and spatial decoupling characteristics, and generates the spectral and spatial feature decoupling values;

[0051] The feature mapping enhancement module calls the image feature set based on the spectral and spatial feature decoupling value, analyzes the characteristic component distribution, screens the distribution area of ​​the explicit and implicit characteristic values, weights the distribution matrix, enhances the explicit and implicit characteristic components, reconstructs the spectral and spatial characteristic distribution through mapping, and generates a feature dictionary enhancement value;

[0052] The pixel characteristic reconstruction module calls the pixel characteristics in the region based on the feature dictionary enhancement value, calculates the distribution parameters of the pixels on the spectral matrix, reconstructs the characteristic values ​​of the pixels and recalculates the distribution parameters, completes the parameter matrix normalization and mapping operation, and generates the spectral and spatial characteristic change matrix;

[0053] The classification matrix generation module calls the structural change matrix based on the spectral and spatial characteristic change matrix, calculates the distribution parameters within the partition, completes the matrix classification operation of the regional distribution parameters, compares the partition distribution relationship in the matrix, extracts the classification parameters of the change relationship, and generates the image characteristic classification matrix.

[0054] Compared with the prior art, the advantages and positive effects of the present invention are:

[0055] This method, through spectral distribution analysis and channel matrix numerical calculations of multispectral image data, can uncover deeper spectral and latent features, improving analysis accuracy. Combined with the spatial position matrix, it effectively decouples spectral and spatial characteristics, enhancing the accuracy and efficiency of feature extraction. By screening and weighted enhancement of explicit and implicit features, it optimizes the expressive power of feature distribution and improves target recognition accuracy. Dynamic mapping of spectral and spatial features accurately captures surface changes, enhancing the stability and accuracy of classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a schematic diagram of the workflow of the present invention;

[0057] Figure 2 Flowchart of the steps for obtaining the eigenvalues ​​of the spectrum sparse matrix of the present invention;

[0058] Figure 3 Flowchart of the steps for obtaining the decoupling value of the spectrum and spatial characteristics of the present invention;

[0059] Figure 4 Flowchart of the steps for obtaining the feature dictionary enhancement value of the present invention;

[0060] Figure 5 Flowchart of the steps for obtaining the spectrum and spatial characteristic variation matrix of the present invention;

[0061] Figure 6 Flowchart of the steps for obtaining the image characteristic classification matrix of the present invention. DETAILED DESCRIPTION

[0062] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0063] In the description of the present invention, it should be understood that the terms "length," "width," "up," "down," "front," "back," "left," "right," "vertical," "horizontal," "top," "bottom," "inside," "outside," and the like, indicating positions or relationships, are based on the positions or relationships shown in the accompanying drawings and are intended only to facilitate the description of the present invention and simplify the description. They do not indicate or imply that the devices or elements referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limiting the present invention. Furthermore, in the description of the present invention, "plurality" means two or more, unless otherwise expressly and specifically defined.

[0064] Example 1:

[0065] See also Figure 1 The present invention provides a technical solution: a remote sensing image processing method based on the fusion of AI and computer vision, comprising the following steps:

[0066] S1: Based on the input multispectral image data, the spectral distribution data of the image is analyzed, the multispectral channel matrix is ​​grouped and numerically operated, the distribution characteristic values ​​and explicit and implicit parameters are extracted, and the matrix decomposition is completed through matrix calculation and classification operations to generate the spectral sparse matrix eigenvalues;

[0067] S2: Based on the spectral sparse matrix eigenvalues, the spatial position matrix of the image is called to analyze the cross-distribution characteristics of the spectral channels and position coordinates, calculate the characteristic values ​​within the partition, remove the redundant components, complete the normalization and feature extraction operations, and generate the spectral and spatial feature decoupling values;

[0068] S3: Based on the decoupling values ​​of spectral and spatial features, the image feature set is called to analyze the multidimensional feature components, the explicit and implicit feature values ​​are partitioned and screened, and the distribution matrix of the feature components is weighted and enhanced. The extracted features are reconstructed through feature mapping to generate feature dictionary enhancement values.

[0069] S4: Based on the feature dictionary enhancement value, the regional pixel characteristics of each pixel are called, the distribution parameters of the pixel and the spectral matrix are reconstructed and the characteristic values ​​are calculated, the parameter matrix is ​​normalized and mapped, the characteristic change distribution is calculated, and the spectral and spatial characteristic change matrix is ​​generated;

[0070] S5: Based on the spectral and spatial characteristic change matrix, the structural change matrix in the image is called, and the distribution parameters of multiple partitions in the matrix are classified and compared with the matrix. The classification results are extracted by calculating the regional change relationship, the characteristic classification operation is completed, and the image characteristic classification matrix is ​​generated.

[0071] The spectral sparse matrix eigenvalues ​​include distribution characteristic values, explicit and implicit parameters, matrix grouping results, and matrix operation results. The spectral and spatial feature decoupling values ​​are specifically spectral partition characteristic values, spatial position matrix cross-distribution characteristic values, normalized characteristic values, and de-redundant characteristic values. The feature dictionary enhancement values ​​include characteristic component distribution matrix, explicit and implicit feature partition values, feature mapping reconstruction values, and weighted enhancement feature values. The spectral and spatial feature change matrix specifically refers to pixel point area distribution parameters, feature value change matrix, normalized parameter matrix, and feature distribution mapping matrix. The image feature classification matrix includes multi-partition distribution parameters, structural change matrix, regional change relationship classification results, and feature classification comparison results.

[0072] See also Figure 2 , the specific steps for obtaining the eigenvalues ​​of the spectral sparse matrix are:

[0073] S111: performing spectral distribution analysis on the input multispectral image data, analyzing the spectral distribution characteristics of multiple channels in the multispectral channel matrix, and generating a multispectral channel spectral characteristic matrix by calculating the relationship between the normalized spectral values ​​and the wavelength distribution among the multiple channels;

[0074] The multispectral image data is decomposed into multiple spectral distribution units channel by channel, and the data value of each unit is normalized to eliminate the influence of amplitude differences between different channels. The normalization result is called and the spectral distribution within the wavelength range of each channel is discretized and calculated. The spectral characteristic matrix of each channel is constructed through the discretized spectral distribution array. The spectral characteristic matrix is ​​used as the characteristic expression of the spectral characteristic distribution within the channel, and the coupling degree of the spectral distribution between channels is defined according to the interactive relationship between the data values ​​of each row and each column in the matrix. Specifically, the distribution characteristic value of each channel is calculated after segmenting the data, and finally the spectral characteristic matrix of the multispectral channel is formed.

[0075] S112: Calculating the inter-channel fitting matrix based on the multi-spectral channel spectral characteristic matrix, performing standardization on the matrix, combining the explicit and implicit parameters of the multiple channels, adjusting the weights of the matrix component values ​​according to the weight distribution rule of the characteristic parameters, and generating an explicit and implicit weighted characteristic matrix;

[0076] First, the row vector and column vector data in the matrix are extracted, and these data vectors are converted into points in a multidimensional vector space. The Euclidean distance matrix is ​​calculated for the relative distribution relationship between the points, and the similarity between the channels is calculated based on the normalized distance matrix. By calling the spectral characteristic values ​​in the characteristic matrix, it is used as the basis for the explicit and implicit parameters. At the same time, the weighting rules for each parameter are set, and the explicit and implicit weighted matrix is ​​generated for each vector data through matrix weighting. Specifically, the weight of each channel data is multiplied by its spectral distribution characteristic value to finally form the explicit and implicit weighted characteristic matrix.

[0077] S113: By performing singular value decomposition on the explicit and implicit weighted feature matrix, and combining the decomposed feature value matrix to screen the judgment relationship between the proportion of multiple feature values ​​and the threshold, the formula is used:

[0078] ;

[0079] Calculate and generate spectral sparse matrix eigenvalues;

[0080] in, Representative The sparse eigenvalues ​​of the channel, Represents the channel in the explicit and implicit weighted characteristic matrix With channel The component value of Represents the channel in the explicit and implicit weighted characteristic matrix Corresponding to The square of the characteristic value, represents the explicit and implicit weight components, Represents the number of channels, Represents the total number of characteristic values.

[0081] formula:

[0082] ;

[0083] The benefit of the formula is that, by performing a weighted summation operation on the explicit and implicit weighted characteristic matrix and combining it with a square normalization operation, the accuracy of the sparse expression of the spectral distribution characteristics between channels is effectively improved, and the eigenvalue distribution characteristics of the data are optimized.

[0084] Detailed explanation of the formula and the process of formula calculation and derivation:

[0085] According to the component data of the explicit and implicit weighted characteristic matrix, the channel sparse characteristic value of the first characteristic matrix is ​​calculated. The explicit and implicit weighted values ​​of each channel are calculated by absolutizing each row value of the explicit and implicit matrix and summing the values ​​in the column direction. Perform explicit and implicit distribution ratio allocation, introduce each weight value into the absolute value and calculate , by calculating the sum of the squares of the components in the denominator Take the square root to complete the calculation of the sparse feature value. The example calculation is as follows:

[0086] Assumptions , in the explicit and implicit characteristic matrix For the following data:

[0087] ;

[0088] The weight value is , then the sparse feature value is calculated as:

[0089] ;

[0090] Repeat the above calculation steps to obtain the sparse feature values ​​of the remaining channels:

[0091] ;

[0092] The results show that the sparse characteristic value reflects the sparsity of the explicit and implicit distribution coupling characteristics between channels. The higher the value, the more significant the sparsity. It is directly related to the eigenvalue of the spectral sparse matrix, providing a basis for subsequent partitioning or classification.

[0093] See also Figure 3 , the specific steps for obtaining the decoupling value of spectral and spatial characteristics are:

[0094] S211: Based on the spectral sparse matrix eigenvalues, the spatial position matrix of the image is called, the cross-distribution characteristics of the spectral channels and the position coordinates are analyzed, the spatial position characteristic value corresponding to each spectral channel is calculated, and the spectral space cross-distribution matrix is ​​generated;

[0095] The spectral sparse matrix eigenvalues ​​are obtained by decomposing the sparse characteristics of different spectral channels in the multispectral image. The spatial position matrix of the image is called, and the spatial position characteristic value corresponding to each spectral channel is calculated by analyzing the cross-distribution relationship between the distribution characteristics of different partitions in the spatial position matrix and the spectral characteristic components. The sparse characteristic components of the spectral channel are obtained through pre-order operations, specifically including matrix normalization and singular value decomposition based on the spectral intensity distribution, and classifying the sparse spectral channels after eliminating the components with singular value contributions below a set threshold. The spatial position matrix is ​​further combined, and the partition position information is used as the weight factor of the matrix operation. The spatial distribution characteristics of each spectral channel are obtained through weighted calculation, and then the mean and variance are calculated to reflect the overall distribution trend to generate a spectral space cross-distribution matrix.

[0096] S212: Eliminating redundant components in the distribution matrix based on the partition characteristic values ​​of the spectral spatial cross-distribution matrix, and determining and adjusting the normalization weights based on the frequency distribution characteristics of the characteristic values ​​and a normalization standard to generate a characteristic normalized distribution matrix;

[0097] First, the high-frequency and low-frequency component characteristics are extracted from the partition characteristic values ​​of the matrix, and the component characteristic values ​​are classified according to the frequency distribution range. The classification standard of the frequency distribution is calculated based on the standard deviation and mean of the spectral signal of each partition. The significant low-frequency components below the lower limit of the characteristic value frequency distribution range are eliminated, and the normalization operation is completed by adjusting the component frequency distribution weight coefficient. The normalization operation is specifically to normalize the absolute value of each characteristic component by weighting so that its distribution range falls within a specific numerical interval. The contribution rate of the cross-contribution value of the spectral and spatial components in the normalized distribution matrix is ​​further calculated. The significant component is judged by the contribution rate and the redundant component is eliminated to generate a characteristic normalized distribution matrix.

[0098] S213: Perform cross-validation and parameter extraction on the characteristic components in the characteristic normalized distribution matrix, construct an explicit and implicit weight matrix for the multi-partition characteristic components, and calculate the normalized decoupling ratio of the components using the formula:

[0099] ;

[0100] Generate decoupled values ​​of spectral and spatial features;

[0101] in, Represents spectral channels and spatial coordinates The decoupled eigenvalue of Spectral channels in the representative characteristic normalized distribution matrix In quantity The value on Represents the components in the explicit and implicit weight matrix In spatial coordinates The weight value on is the characteristic adjustment coefficient, used to correct the ratio of deviation, Normalization adjustment parameters are used to control the numerical balance during the denominator normalization process. is the deviation correction value of the characteristic component, is the number of spectral components, is the number of coordinate components, Spectral channels in the representative characteristic normalized distribution matrix In spatial components The characteristic value on .

[0102] formula:

[0103] ;

[0104] The benefit of the formula is that by introducing the synergistic effect of sparse feature components and explicit and implicit weight components in the feature normalized distribution matrix, the ability of the decoupled value to identify the feature weight distribution in multidimensional matrix calculations is enhanced, thereby further improving the numerical stability and resolution of the decoupled value.

[0105] Detailed explanation of the formula and the process of formula calculation and derivation:

[0106] It is obtained by calculating the square mean of the partition matrix difference, specifically:

[0107] ;

[0108] in and are the values ​​of specific components of the spectral and spatial partition matrices, is the number of components in the matrix, The normalization adjustment parameter is used to control the normalization range of the denominator to prevent the ratio from being too large due to a too small value. It is set to 0.01 through experimental verification. represents the weighted summation calculation of all components in the spectral channel, Represents the calculation of the sum of squares of the weighted contributions of the spatial components.

[0109] Calculation process:

[0110] Calculate the characteristic normalized distribution matrix components:

[0111] for The specific component value of is 0.8. The value is 0.6, The value is 0.05, Calculated by the above formula, it is 0.02;

[0112] Calculate the molecular part:

[0113] ;

[0114] Calculate the denominator:

[0115] ;

[0116] Calculate the decoupling value:

[0117] ;

[0118] The results show that:

[0119] Calculated Represents spectral channels and spatial coordinates The decoupling characteristic value in the normalized distribution of characteristics is at a high level, further demonstrating the contribution of the synergistic effect of sparse characteristic components and explicit and implicit weights in the calculation of decoupling characteristics. This value will serve as an important input parameter in subsequent spectral and spatial characteristic analysis to screen significant characteristic partitions and channels.

[0120] See also Figure 4 , the steps for obtaining the feature dictionary enhancement value are as follows:

[0121] S311: Based on the decoupling values ​​of spectral and spatial features, the associated image feature set is called to analyze the data structure and correlation degree in the multi-dimensional feature components. Through mapping analysis and explicit and implicit comparison of the feature components, a pre-processed explicit and implicit feature value matrix is ​​generated.

[0122] First, it is necessary to obtain the partition feature data in the spectral and spatial feature decoupling values ​​and the corresponding spatial feature components. Through the correlation analysis of the partition feature components, the distribution trend of each feature value is analyzed. The correlation coefficient is calculated using the linear characteristics of the data distribution. The linear relationship between the components is measured by calculating the Pearson correlation coefficient. Secondly, the components with high correlation are normalized. The normalization method can be selected to rescale the data according to the maximum and minimum value range. The calculation formula is: ,in and Represent the minimum and maximum values ​​of the data components respectively. The normalized data can eliminate the dimension effect and improve the consistency of data comparison. Then, combining the distribution trend with the normalized explicit and implicit characteristic data, the explicit and implicit comparison of each characteristic component is performed, and the significance factor is calculated using the explicit and implicit value of the characteristic value in the partition. The formula for calculating the significance factor can be: ,in represents the component mean, represents the component standard deviation, Represents the significant factor, The characteristic components above a certain threshold are further extracted to form a highly significant characteristic component matrix. Finally, the above matrix is ​​mapped to the explicit and implicit characteristic value matrix. The calculation and reorganization of the explicit and implicit characteristic values ​​are completed in combination with the mapping rules to generate the preprocessed explicit and implicit characteristic value matrix.

[0123] S312: Performing partition screening on the pre-processed explicit and implicit feature value matrix. Based on the frequency distribution characteristics of the feature values, comparing the standard deviation and mean of the feature values ​​within the partition, and judging the validity of the feature values, outliers are eliminated to generate a filtered feature partition matrix.

[0124] First, the distribution matrix needs to be divided into multiple feature subsets according to the partitioning rules. Each subset contains the data structure and frequency distribution pattern of the feature values ​​in the partition. The distribution of the feature values ​​in each subset is statistically analyzed through the frequency histogram, and the mean and variance of the frequency distribution are obtained. The formula for calculating the standard deviation of the feature values ​​in the partition is: ,in is the standard deviation, is the mean, is the number of components, For each characteristic value, after the calculation is completed, the standard deviation of each partition is compared with the mean, and the partition characteristic values ​​with variance lower than a specific threshold are screened out; secondly, outliers are eliminated based on the validity of the characteristic value. The outlier range can be identified by the box plot method, and the outlier value is set to be higher than the upper quartile. Plus 1.5 times the interquartile range or below the lower quartile Subtract 1.5 times the interquartile range from the original value and eliminate these values. Finally, use the characteristic components after effectiveness screening to reconstruct the partition characteristic matrix and generate the filtered characteristic partition matrix.

[0125] S313: Perform weighting and enhancement operations on the filtered feature partition matrix and calculate the dynamic weight coefficients of the feature components using the formula:

[0126] ;

[0127] Calculate the feature map to reconstruct the extracted features and generate feature dictionary enhancement values;

[0128] in, Representative feature partition and characteristic dimensions The enhancement value, Characteristic component The weight of Partitioning features Middle The value of the component, Characteristic component The mean of To avoid small positive numbers with zero denominator, is the total number of characteristic components.

[0129] formula:

[0130] ;

[0131] The benefit of the formula is that it enhances the influence of low-amplitude feature components by dynamically adjusting weight parameters combined with the deviation calculation of feature values, thereby improving the accuracy and robustness of the feature map reconstruction process.

[0132] Detailed explanation of the formula and the process of formula calculation and derivation:

[0133] set up , represents the total number of characteristic components, , are the weights of the three characteristic components, , are the means of each characteristic component, , is the adjustment coefficient, the partition characteristic value matrix ;

[0134] Calculate component deviation:

[0135] ;

[0136] Calculate the normalized bias term:

[0137] ;

[0138] Compute the logarithmic term:

[0139] ;

[0140] Compute the weighted sum:

[0141] ;

[0142] The results show that the enhanced value extracted after feature map reconstruction is 14.85. This result can be further used for feature dictionary optimization and enhancement calculation to ensure the data feature integrity and weight distribution rationality of the mapping process.

[0143] See also Figure 5 , the specific steps for obtaining the spectral and spatial characteristic change matrix are:

[0144] S411: Based on the feature dictionary enhancement value, call the regional pixel characteristic data associated with each pixel point, analyze the multidimensional distribution information of the regional pixel characteristic data, extract the statistical distribution parameters of the regional pixel characteristic data and calculate the correlation matrix value, reconstruct each dimension of the spectral matrix segment by segment using the distribution parameters, dynamically match and compare the pixel values ​​based on the reconstructed parameter matrix, and generate a preliminary characteristic value calculation matrix;

[0145] First, we need to clarify the key feature dimensions involved in the feature dictionary, analyze the regional pixel characteristic data associated with each pixel point one by one, and obtain the multidimensional distribution characteristics of the data. This process can be implemented through programming, such as using Python to read the feature dictionary and extract the distribution information in the regional pixel characteristic data. Then, based on these data, we calculate its statistical distribution parameters, such as mean, variance, skewness, etc., and construct the correlation matrix value. Assume that the characteristic data is in matrix form , the calculation formula of its correlation matrix can be expressed as ,in It is a distribution characteristic calculation function, such as setting the segmentation threshold , partition the pixel values ​​and reconstruct the spectral characteristic matrix. For each segment, the reconstructed parameter matrix is ​​compared pixel by pixel using the dynamic matching method, for example, the pixel characteristic values ​​of different segments are compared with the reference matrix ,in is the matrix to be reconstructed, As a reference matrix, a preliminary characteristic value calculation matrix is ​​generated.

[0146] S412: performing normalization processing on the preliminary characteristic value calculation matrix, adjusting the distribution ratio by calculating the minimum and maximum ranges of the characteristic values, and optimizing the characteristic value distribution weights in combination with the mapping rules of the characteristic value distribution to obtain a normalized parameter matrix;

[0147] First, we need to clarify the scope and goal of normalization. For example, all characteristic values ​​are normalized to the interval [0,1]. The normalization formula can be expressed as ,in and are the minimum and maximum values ​​of the characteristic value matrix respectively. In the specific implementation, the minimum and maximum values ​​are obtained by traversing the matrix, and then the distribution ratio is adjusted. For example, the linear interpolation method is used to optimize the distribution weight of the characteristic value. According to the mapping rule, such as adjusting the normalized value distribution ,in and Adjust the parameters for the distribution, assuming the specific value is , thereby optimizing the distribution weights, and performing the above adjustment process for each characteristic value in the matrix to obtain the normalized parameter matrix.

[0148] S413: Extract characteristic change data from the normalized parameter matrix, perform dynamic analysis based on the fluctuation of distribution parameters, and calculate the standardized change distribution of characteristic values ​​of multiple pixel points using the formula:

[0149] ;

[0150] Generate spectral and spatial characteristic variation matrix;

[0151] in, Represents pixel points and characteristic dimensions The change value of is the first parameter in the normalized parameter matrix The pixel characteristic value, Characteristic component The mean of Characteristic component The standard deviation of is the total number of characteristic components.

[0152] The benefit of this formula is that by calculating the standardized variance of the characteristic components, the influence of the mean and standard deviation on characteristic changes is comprehensively considered, which can significantly enhance the sensitivity of characteristic changes. At the same time, the introduction of dynamic weights makes the results more consistent with the actual change distribution.

[0153] Detailed explanation of the formula and the process of formula calculation and derivation:

[0154] By normalizing the distribution calculation formula Get, assume 、 、 ,but .

[0155] The component values ​​of all pixels are averaged, assuming 、 、 ,but .

[0156] By formula Calculated, assuming 、 、 、 ,but

[0157] ;

[0158] Substitute into the formula:

[0159] ;

[0160] The results show that the pixel and characteristic dimensions The change value of The numerical value shows that the characteristic varies greatly within the distribution range and the characteristic fluctuates significantly. This value is introduced into the spectral and spatial characteristic change matrix for distribution evaluation and further analysis.

[0161] See also Figure 6 , the steps to obtain the image feature classification matrix are as follows:

[0162] S511: Based on the spectral and spatial characteristic change matrix, the structural change matrix in the image is called, and the distribution parameters of multiple partitions in the matrix are compared, and the characteristic comparison and difference calculation in the partition change relationship are used to generate a preliminary classification result matrix;

[0163] By extracting the distribution parameters of multiple partitions one by one, based on the numerical characteristics of the distribution matrix and the relationship between the partition parameters, the characteristics comparison between regions is completed by calculating the mean and standard deviation of the partition changes, and the distribution parameters are normalized and adjusted to the same scale to calculate the differences in the partition parameters. These parameter differences are sorted and combined with the partition characteristic weight distribution to analyze the change range of specific partitions. At the same time, combined with the data reconstruction technology between multiple partitions, the distribution parameters are classified. The change comparison and difference calculation of multiple partitions are completed through the parameter values ​​after matrix classification. By inputting the change parameters of each partition of the matrix into a specific classification model, a preliminary classification result matrix is ​​generated.

[0164] S512: Optimizing the data in the preliminary classification result matrix, adjusting the thresholds of the partition parameters and refining the classification rules, adjusting and classifying the distribution characteristic values ​​of multiple partitions, and obtaining an optimized classification result matrix;

[0165] By normalizing the distribution parameters of each partition in the matrix, the partition characteristic parameters of the matrix are mapped to a fixed characteristic range to eliminate the influence of the distribution parameters. At the same time, the mapped data is analyzed for differences with the target partition data. By calculating the coefficient of variation and difference value distribution of the data in the partition, the data points with significant differences in the partition are extracted, and the weight adjustment rules are used to assign higher weights to the distribution parameters with larger differences. The partition adjustment is performed again to complete the refined classification of the partition parameters. The classification results are compared with the optimized partition data to obtain the optimized classification result matrix.

[0166] S513: Perform characteristic classification operation on the optimized classification result matrix, calculate the classification characteristic value and characteristic weight response value of each partition, and conduct comprehensive analysis in combination with the differences in environmental parameters, using the formula:

[0167] ;

[0168] Calculate the comprehensive classification results of the characteristic values ​​and generate the image characteristic classification matrix;

[0169] in, Indicates the Partition No. The categorical value of the characteristic, For the Partition No. The response value of the parameter, For the The environmental parameter value of the characteristic, is the average value of environmental parameters in all partitions, is the total number of parameters.

[0170] The benefit of the formula is that, by combining the logarithmic ratio of the response value and the environmental parameter, the discrimination and sensitivity of the partition characteristic classification are enhanced, so that the classification results can better reflect the differences in actual distribution characteristics.

[0171] Detailed explanation of the formula and the process of formula calculation and derivation:

[0172] The response value obtained by monitoring the response characteristics of the partition, such as the spectral reflectance value of each partition extracted by the spectral characteristic sensor, assumes that the partition The mid-spectral reflectance data is , then the corresponding Take these values; It is the environmental parameter value quantified by experiment. For example, the light intensity data is obtained by light sensor. The specific value is ; It is calculated by the mean value of the environmental parameters of the partition using the formula:

[0173] ;

[0174] After entering the value:

[0175] ;

[0176] Then bring these parameters into the formula for calculation:

[0177] ;

[0178] Calculate item by item:

[0179] for :

[0180] ;

[0181] for :

[0182] ;

[0183] for :

[0184] ;

[0185] Finally, sum all the terms:

[0186] ;

[0187] This result shows that the partition The classification feature value is , indicating that the partition has a low degree of variation in environmental parameters and response characteristics, which is consistent with the image feature classification results. The numerical results are closely related to the ability of the classification model to distinguish feature differences and can further optimize the accuracy of the classification matrix.

[0188] A remote sensing image processing system based on the fusion of AI and computer vision is used to execute the above-mentioned remote sensing image processing method based on the fusion of AI and computer vision. The system includes:

[0189] The spectral characteristic analysis module analyzes the spectral distribution data based on the input multispectral image data, completes the grouping operation of the spectral channel matrix, extracts the distribution characteristic values ​​between channels, classifies the explicit and implicit parameters, decomposes the classified matrix data, and generates the spectral sparse matrix eigenvalues;

[0190] The spatial feature extraction module is based on the spectral sparse matrix eigenvalues, calls the spatial position matrix, analyzes the cross-distribution characteristics of the spectral channel and the position coordinates, calculates the partition cross-characteristic value, eliminates redundant components, normalizes the spatial position characteristics, extracts the spectral and spatial decoupling characteristics, and generates the spectral and spatial feature decoupling value;

[0191] The feature mapping enhancement module calls the image feature set based on the decoupling value of spectral and spatial features, analyzes the distribution of feature components, screens the distribution area of ​​explicit and implicit feature values, weights the distribution matrix, enhances the explicit and implicit feature components, reconstructs the spectral and spatial feature distribution through mapping, and generates feature dictionary enhancement values;

[0192] The pixel feature reconstruction module uses the feature dictionary enhancement value to call the pixel features in the area, calculate the distribution parameters of the pixels on the spectral matrix, reconstruct the feature values ​​of the pixels and recalculate the distribution parameters, complete the parameter matrix normalization and mapping operation, and generate the spectral and spatial feature change matrix;

[0193] The classification matrix generation module is based on the spectral and spatial characteristic change matrix, calls the structural change matrix, calculates the distribution parameters within the partition, completes the matrix classification operation of the regional distribution parameters, compares the partition distribution relationship in the matrix, extracts the classification parameters of the change relationship, and generates the image characteristic classification matrix.

[0194] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A remote sensing image processing method based on the fusion of AI and computer vision, characterized in that: The following steps are involved: S1: Based on the input multispectral image data, the spectral distribution data of the image is analyzed, the multispectral channel matrix is ​​grouped and numerically operated, the distribution characteristic values ​​and explicit and implicit parameters are extracted, and the matrix decomposition is completed through matrix calculation and classification operations to generate the spectral sparse matrix eigenvalues; The steps for obtaining the eigenvalues ​​of the spectrum sparse matrix are specifically as follows: S111: performing spectral distribution analysis on the input multispectral image data, analyzing the spectral distribution characteristics of multiple channels in the multispectral channel matrix, and generating a multispectral channel spectral characteristic matrix by calculating the relationship between the normalized spectral values ​​and the wavelength distribution among the multiple channels; S112: Calculating a fitting matrix between channels based on the multi-spectral channel spectral characteristic matrix, performing standardization processing on the matrix, combining the explicit and implicit parameters of the multiple channels, and adjusting the weights of the matrix component values ​​according to the weight distribution rule of the characteristic parameters to generate an explicit and implicit weighted characteristic matrix; S113: Performing singular value decomposition on the explicit and implicit weighted characteristic matrix, and combining the decomposed characteristic value matrix to screen the judgment relationship between the proportion of multiple characteristic values ​​and the threshold, using the formula: ； Calculate and generate spectral sparse matrix eigenvalues; in, Representative The sparse eigenvalues ​​of the channel, Represents the channel in the explicit and implicit weighted characteristic matrix With channel The component value of Represents the channel in the explicit and implicit weighted characteristic matrix Corresponding to The square of the characteristic value, represents the explicit and implicit weight components, Represents the number of channels, represents the total number of characteristic values; S2: Based on the spectral sparse matrix eigenvalues, the spatial position matrix of the image is called to analyze the cross-distribution characteristics of the spectral channels and the position coordinates, calculate the characteristic values ​​within the partition, remove the redundant components and complete the normalization and feature extraction operations to generate the spectral and spatial feature decoupling values; S3: Based on the spectral and spatial feature decoupling values, call the image feature set, analyze the multidimensional feature components, perform partition screening on the explicit and implicit feature values, perform weighting and enhancement operations on the distribution matrix of the feature components, reconstruct the extracted features through feature mapping, and generate feature dictionary enhancement values; S4: Based on the feature dictionary enhancement value, call the regional pixel characteristics of each pixel point, reconstruct the distribution parameters and calculate the characteristic values ​​of the pixel points and the spectral matrix, normalize and map the parameter matrix, calculate the characteristic change distribution, and generate the spectral and spatial characteristic change matrix; S5: Based on the spectral and spatial characteristic change matrix, the structural change matrix in the image is called, the distribution parameters of multiple partitions in the matrix are classified and compared with the matrix, the classification results are extracted by calculating the regional change relationship, the characteristic classification operation is completed, and the image characteristic classification matrix is ​​generated; The spectral sparse matrix eigenvalues ​​include distribution characteristic values, explicit and implicit parameters, matrix grouping results, and matrix operation results. The spectral and spatial feature decoupling values ​​are specifically spectral partition characteristic values, spatial position matrix cross-distribution characteristic values, normalized characteristic values, and de-redundant characteristic values. The feature dictionary enhancement values ​​include characteristic component distribution matrix, explicit and implicit characteristic partition values, characteristic mapping reconstruction values, and weighted enhancement characteristic values. The spectral and spatial characteristic change matrix specifically refers to pixel point area distribution parameters, characteristic value change matrix, normalized parameter matrix, and characteristic distribution mapping matrix. The image characteristic classification matrix includes multi-partition distribution parameters, structural change matrix, regional change relationship classification results, and characteristic classification comparison results.

2. The remote sensing image processing method based on the fusion of AI and computer vision according to claim 1, characterized in that: The steps for obtaining the spectral and spatial feature decoupling values ​​are specifically as follows: S211: Based on the spectral sparse matrix eigenvalues, calling the spatial position matrix of the image, analyzing the cross-distribution characteristics of the spectral channels and the position coordinates, calculating the spatial position characteristic value corresponding to each spectral channel, and generating a spectral space cross-distribution matrix; S212: Eliminating redundant components in the distribution matrix based on the partition characteristic values ​​of the spectral spatial cross-distribution matrix, and determining and adjusting normalization weights based on the frequency distribution characteristics of the characteristic values ​​and a normalization standard to generate a characteristic normalized distribution matrix; S213: Perform cross-validation and parameter extraction on the characteristic components in the characteristic normalized distribution matrix, construct an explicit and implicit weight matrix for the multi-partition characteristic components, and calculate the normalized decoupling ratio of the components using the formula: ； Generate decoupled values ​​of spectral and spatial features; in, Represents spectral channels and spatial coordinates The decoupled eigenvalue of Spectral channels in the representative characteristic normalized distribution matrix In quantity The value on Represents the components in the explicit and implicit weight matrix In spatial coordinates The weight value on is the characteristic adjustment coefficient, used to correct the ratio of deviation, Normalization adjustment parameters are used to control the numerical balance during the denominator normalization process. is the deviation correction value of the characteristic component, is the number of spectral components, is the number of coordinate components, Spectral channels in the representative characteristic normalized distribution matrix In spatial components The characteristic value on .

3. The remote sensing image processing method based on the fusion of AI and computer vision according to claim 2, characterized in that: The steps for obtaining the feature dictionary enhancement value are specifically as follows: S311: Based on the spectral and spatial feature decoupling values, call the associated image feature set, analyze the data structure and correlation degree in the multi-dimensional feature components, and generate a pre-processed explicit and implicit feature value matrix through mapping analysis and explicit and implicit comparison of the feature components; S312: performing partition screening on the pre-processed explicit and implicit characteristic value matrix, based on the frequency distribution characteristics of the characteristic values, comparing the standard deviation and mean of the characteristic values ​​in the partition, and combining the validity judgment of the characteristic values ​​to eliminate abnormal values, thereby generating a filtered characteristic partition matrix; S313: Perform weighting and enhancement operations on the filtered feature partition matrix, and calculate the dynamic weight coefficients of the feature components using the formula: ； Calculate the feature map to reconstruct the extracted features and generate feature dictionary enhancement values; in, Representative feature partition and characteristic dimensions The enhancement value, Characteristic component The weight of Partitioning features Middle The value of the component, Characteristic component The mean of To avoid small positive numbers with zero denominator, is the total number of characteristic components.

4. The remote sensing image processing method based on the fusion of AI and computer vision according to claim 3, characterized in that: The steps for obtaining the spectrum and spatial characteristic change matrix are specifically as follows: S411: Based on the feature dictionary enhancement value, calling the regional pixel characteristic data associated with each pixel point, parsing the multidimensional distribution information of the regional pixel characteristic data, extracting the statistical distribution parameters of the regional pixel characteristic data and calculating the correlation matrix value, reconstructing each dimension of the spectral matrix segment by segment using the distribution parameters, dynamically matching and comparing the pixel values ​​according to the reconstructed parameter matrix, and generating a preliminary characteristic value calculation matrix; S412: performing normalization processing on the preliminary characteristic value calculation matrix, adjusting the distribution ratio by calculating the minimum and maximum ranges of the characteristic values, optimizing the characteristic value distribution weights in combination with the mapping rules of the characteristic value distribution, and obtaining a normalized parameter matrix; S413: Extract characteristic change data from the normalized parameter matrix, perform dynamic analysis based on the fluctuation of distribution parameters, and calculate the standardized change distribution of characteristic values ​​of multiple pixel points using the formula: ； Generate spectral and spatial characteristic variation matrix; in, Represents pixel points and characteristic dimensions The change value of is the first parameter in the normalized parameter matrix The pixel characteristic value, Characteristic component The mean of Characteristic component The standard deviation of is the total number of characteristic components.

5. The remote sensing image processing method based on the fusion of AI and computer vision according to claim 4, characterized in that: The steps for obtaining the image characteristic classification matrix are specifically as follows: S511: Based on the spectral and spatial characteristic change matrix, the structural change matrix in the image is called, and a preliminary classification result matrix is ​​generated by comparing the distribution parameters of multiple partitions in the matrix and using characteristic comparison and difference calculation in the partition change relationship; S512: Optimizing the data in the preliminary classification result matrix, adjusting the thresholds of the partition parameters and refining the classification rules, adjusting and classifying the distribution characteristic values ​​of multiple partitions, and obtaining an optimized classification result matrix; S513: Performing a characteristic classification operation on the optimized classification result matrix, calculating the classification characteristic value and characteristic weight response value of each partition, and performing a comprehensive analysis in combination with the environmental parameter differences, using the formula: ； Calculate the comprehensive classification results of the characteristic values ​​and generate the image characteristic classification matrix; in, Indicates the Partition No. The categorical value of the characteristic, For the Partition No. The response value of the parameter, For the The environmental parameter value of the characteristic, is the average value of environmental parameters in all partitions, is the total number of parameters.

6. A remote sensing image processing system based on the fusion of AI and computer vision, characterized by: The remote sensing image processing method based on the fusion of AI and computer vision according to any one of claims 1 to 5, wherein the system comprises: The spectral characteristic analysis module analyzes the spectral distribution data based on the input multispectral image data, completes the grouping operation of the spectral channel matrix, extracts the distribution characteristic values ​​between channels, classifies the explicit and implicit parameters, decomposes the classified matrix data, and generates the spectral sparse matrix eigenvalues; The spatial feature extraction module calls the spatial position matrix based on the spectral sparse matrix eigenvalues, analyzes the cross-distribution characteristics of the spectral channels and position coordinates, calculates the partition cross-characteristic values, eliminates redundant components, normalizes the spatial position characteristics, extracts the spectral and spatial decoupling characteristics, and generates the spectral and spatial feature decoupling values; The feature mapping enhancement module calls the image feature set based on the spectral and spatial feature decoupling value, analyzes the characteristic component distribution, screens the distribution area of ​​the explicit and implicit characteristic values, weights the distribution matrix, enhances the explicit and implicit characteristic components, reconstructs the spectral and spatial characteristic distribution through mapping, and generates a feature dictionary enhancement value; The pixel characteristic reconstruction module calls the pixel characteristics in the region based on the feature dictionary enhancement value, calculates the distribution parameters of the pixels on the spectral matrix, reconstructs the characteristic values ​​of the pixels and recalculates the distribution parameters, completes the parameter matrix normalization and mapping operation, and generates the spectral and spatial characteristic change matrix; The classification matrix generation module calls the structural change matrix based on the spectral and spatial characteristic change matrix, calculates the distribution parameters within the partition, completes the matrix classification operation of the regional distribution parameters, compares the partition distribution relationship in the matrix, extracts the classification parameters of the change relationship, and generates the image characteristic classification matrix.

Images