Search

CN-121999495-A - High-resolution satellite data semantic analysis and intelligent recognition system and method

CN121999495ACN 121999495 ACN121999495 ACN 121999495ACN-121999495-A

Abstract

The invention relates to the field of high-resolution remote sensing image analysis based on artificial intelligence, and discloses a high-resolution satellite data semantic analysis and intelligent recognition system and method, comprising the steps of carrying out texture complexity, edge density and significance response analysis on a high-resolution satellite image, generating a multi-scale self-adaptive grid unit and establishing a spatial index; the method comprises the steps of loading corresponding multi-temporal images to generate a multi-scale space-time feature sequence, determining a neighborhood grid set according to the multi-scale space-time feature sequence and a space index to generate a space-time correlation matrix and predict semantic change trend of a main grid, carrying out scale alignment and local disturbance suppression on the multi-temporal features to construct a cross-space-time stability reference frame, fusing the cross-space-time stability reference frame, the semantic change trend, the space-time correlation matrix and preliminary semantic judgment of a current frame to form a comprehensive confidence coefficient matrix, adjusting key feature weights and matching parameters, and outputting a grid semantic recognition result after correction. The invention has the advantage of improving the recognition precision.

Inventors

  • WANG JINLIANG
  • LI YANG
  • GUAN JINSHENG

Assignees

  • 湖南创信伟立科技股份有限公司

Dates

Publication Date
20260508
Application Date
20260119

Claims (10)

  1. 1. The high-resolution satellite data semantic analysis and intelligent recognition method is characterized by comprising the following steps of: performing texture complexity, edge density and significance response analysis on the high-resolution satellite image, performing dynamic scale division according to the regional difference, generating a multi-scale self-adaptive grid unit and establishing a spatial index; based on the multi-scale self-adaptive grids and the corresponding multi-time phase images loaded by the spatial indexes, performing multi-scale convolution coding, structure description, gradient sequence extraction and brightness difference calculation on each grid to generate a multi-scale space-time characteristic sequence; Determining a neighborhood grid set according to the multi-scale space-time feature sequence and the space index, constructing a space-time coupling relation among grids, generating a space-time correlation matrix, predicting the semantic change trend of a main grid, performing scale alignment and local disturbance suppression on the multi-time phase features, and constructing a cross-space-time stable reference frame; Fusing the cross-space-time stable reference frame, the semantic change trend, the space-time correlation matrix and the preliminary semantic judgment of the current frame, and calculating the structural consistency, the time sequence consistency and the neighborhood correlation confidence coefficient to form a comprehensive confidence coefficient matrix; and carrying out difference comparison on the basis of the comprehensive confidence coefficient matrix and the historical recognition record, adjusting the key feature weight and the matching parameter, and outputting a corrected grid semantic recognition result.
  2. 2. The method for semantic analysis and intelligent recognition of high-resolution satellite data according to claim 1, wherein the process of generating the multi-scale adaptive grid unit and establishing the spatial index is as follows: Performing initial blocking on the image according to the spatial position and the resolution, and calculating texture entropy, direction gradient statistics and edge sharpness index for each blocking; carrying out quantitative evaluation on the importance of the segmented region by combining the saliency response map, and dividing the image into different scale levels according to the space detail density, the self-similarity and the structural complexity in the region; And generating a multi-scale self-adaptive grid unit by adopting a self-adaptive grid generation strategy aiming at each scale level, recording grid numbers, geographic ranges, level identifications and neighborhood relations, and establishing a spatial index.
  3. 3. The method for semantic analysis and intelligent recognition of high-resolution satellite data according to claim 2, wherein the loading of the corresponding multi-temporal images based on the multi-scale adaptive grid and the spatial index comprises the following steps: extracting image blocks corresponding to grid units of different scales from a multi-temporal image dataset according to grid numbers and geographic ranges recorded by the spatial index; performing geometric correction, brightness standardization and cloud and fog region elimination on the image; and organizing and caching the image blocks according to the scale level of the multi-scale self-adaptive grid.
  4. 4. A method for semantic analysis and intelligent recognition of high-resolution satellite data according to claim 3, wherein the process of generating the multi-scale spatio-temporal feature sequence is: Based on multi-temporal image blocks corresponding to each scale grid, performing convolutional coding, structural texture description, direction gradient sequence extraction and local brightness differential analysis on the image blocks in each scale grid unit to obtain a primary feature set of each time node; the primary feature sets of the same grid unit at different time nodes are ordered according to time stamps, and time synchronization, cross-scale alignment, abnormal feature marking and feature sparsification processing are executed; and constructing a multi-scale space-time characteristic sequence according to the characteristic stability, the integrity and the time density.
  5. 5. The method for semantic analysis and intelligent recognition of high-resolution satellite data according to claim 4, wherein the process of generating a space-time correlation matrix and predicting semantic change trend of the main grid is as follows: determining a space neighborhood set of the main grid according to the space index, and constructing a corresponding time sequence neighborhood set according to the time stamp; Performing joint analysis on the multi-scale space-time characteristic sequences of the main grid and the neighborhood, and calculating the characteristic similarity, the directional gradient consistency and the local variation coupling degree; and generating a space-time correlation matrix according to the coupling degree, and predicting the semantic change trend of the main grid according to the space-time correlation matrix.
  6. 6. The method for semantic analysis and intelligent recognition of high-resolution satellite data according to claim 5, wherein the process of constructing a space-time-stable reference frame is as follows: Based on the space-time correlation matrix and the predicted semantic change trend of the main grid, performing scale alignment, local noise disturbance suppression, structure residual filtering and neighborhood consistency enhancement processing on the multi-temporal feature sequences of each grid; Performing joint analysis on the processed space-time characteristics by utilizing the space adjacency, the time continuity and the cross-scale dependency relationship represented by the space-time correlation matrix, extracting a structure stable region, a continuous change mode and a key change response, and performing weighted inhibition on abnormal disturbance characteristics; And integrating the extracted stable structure and the variation pattern according to the grid number, the spatial position and the scale level to generate a space-time stable reference frame.
  7. 7. The method for semantic analysis and intelligent recognition of high-resolution satellite data according to claim 6, wherein the process of fusing the cross-space-time stable reference frame, the semantic change trend, the space-time correlation matrix and the preliminary semantic judgment of the current frame is as follows: Performing grid-by-grid comparison on the preliminary semantic judgment of the current frame and the structural characteristics of the cross-space stable reference frame, and calculating the space consistency, the local texture matching degree and the cross-scale response difference of each grid; carrying out neighborhood compensation and cross-time phase weighting correction on grids with structural deviation or local abnormality by using neighborhood coupling information provided by a space-time correlation matrix; and carrying out time sequence correction on a drift region which is possibly generated in the preliminary judgment according to the predicted semantic change trend comparison history and the neighborhood information, and generating a preliminary fusion estimated value.
  8. 8. The method for semantic analysis and intelligent recognition of high-resolution satellite data according to claim 7, wherein the process of forming the integrated confidence matrix comprises the following steps: Respectively calculating the confidence coefficient of structural consistency for the fusion semantic estimation result of each grid, wherein the confidence coefficient comprises the stability of internal characteristics of the grids, the texture matching degree and the trans-scale characteristic residual error; Calculating the confidence coefficient of the time sequence consistency based on the time sequence comparison history multi-temporal characteristics and the predicted semantic change trend, wherein the confidence coefficient comprises characteristic drift amount, trend matching degree and abnormal change marks; Calculating neighborhood relevance confidence coefficient including neighborhood similarity, local coupling degree and neighborhood deviation correction coefficient by using the space-time relevance matrix and neighborhood grid characteristic information; and fusing the structure consistency confidence, the time sequence consistency confidence and the neighborhood relevance confidence according to a preset weighting strategy to generate a comprehensive confidence matrix.
  9. 9. The method for semantic analysis and intelligent recognition of high-resolution satellite data according to claim 8, wherein the process of outputting the corrected grid semantic recognition result is as follows: comparing the comprehensive confidence coefficient matrix with the history identification record grid by grid, and calculating the confidence coefficient deviation of each grid in the structure, time and neighborhood dimensions; the method comprises the steps of adjusting key feature weights, including texture features, gradient directions and brightness responses, of grids with deviation exceeding a preset threshold according to confidence contribution and historical weights; and recalculating the grid semantic classification probability according to the adjusted feature weights and grid matching parameters, carrying out neighborhood interpolation or weighted fusion processing on the uncertain grid, and forming a final output from the corrected grid semantic judgment result.
  10. 10. A high resolution satellite data semantic parsing and intelligent recognition system, applied to the method according to any one of claims 1-9, comprising: the self-adaptive grid module is used for analyzing textures, edges and saliency of the satellite images and executing dynamic scale division to generate a multi-scale self-adaptive grid and a spatial index; The space-time characteristic module is used for loading the multi-temporal images and extracting multi-scale space-time characteristics of convolution codes, structure descriptions, gradient sequences and brightness differences on each grid; The association modeling module is used for determining neighborhood grids according to the space-time characteristics and the spatial indexes, constructing space-time coupling relations among grids, predicting semantic change trend and generating a space-time stable reference frame; The confidence fusion module is used for fusing the stable reference frame, the change trend and the incidence matrix with the current frame judgment, and calculating the comprehensive confidence coefficient of the structural consistency, the time sequence consistency and the neighborhood relevance; And the self-adaptive correction module is used for comparing the comprehensive confidence coefficient with the history recognition record, adjusting the feature weight and the matching parameter and generating a corrected semantic recognition result.

Description

High-resolution satellite data semantic analysis and intelligent recognition system and method Technical Field The invention relates to the field of high-resolution remote sensing image analysis based on artificial intelligence, in particular to a high-resolution satellite data semantic analysis and intelligent recognition system and method. Background In a temporary construction scene of an urban road, targets such as construction fence, material stacking and small machinery generally occupy only few pixels in a high-resolution satellite image, the targets are weak in texture and unstable in outline, and are easy to confuse with shadows, light reflection areas, road damage and the like, so that reliable features of a model are difficult to extract. Therefore, it is necessary to design a high-resolution satellite data semantic analysis and intelligent recognition system and method for improving recognition accuracy. Disclosure of Invention Aiming at the defects of the prior art, the invention provides a high-resolution satellite data semantic analysis and intelligent recognition system and method, which have the advantage of improving recognition precision and solve the problems in the background art. In order to achieve the aim of improving the identification precision, the invention provides the following technical scheme that the high-resolution satellite data semantic analysis and intelligent identification method comprises the following steps: performing texture complexity, edge density and significance response analysis on the high-resolution satellite image, performing dynamic scale division according to the regional difference, generating a multi-scale self-adaptive grid unit and establishing a spatial index; Based on the multi-scale self-adaptive grids and the spatial indexes, loading corresponding multi-temporal images, and performing multi-scale convolution coding, structure description, gradient sequence extraction and brightness difference calculation on each grid to generate a multi-scale space-time characteristic sequence; Determining a neighborhood grid set according to the multi-scale space-time feature sequence and the space index, constructing a space-time coupling relation among grids, generating a space-time correlation matrix, predicting the semantic change trend of a main grid, performing scale alignment and local disturbance suppression on the multi-time phase features, and constructing a cross-space-time stable reference frame; Fusing the cross-space-time stable reference frame, the semantic change trend, the space-time correlation matrix and the preliminary semantic judgment of the current frame, and calculating the structural consistency, the time sequence consistency and the neighborhood correlation confidence coefficient to form a comprehensive confidence coefficient matrix; and carrying out difference comparison on the basis of the comprehensive confidence coefficient matrix and the historical recognition record, adjusting the key feature weight and the matching parameter, and outputting a corrected grid semantic recognition result. Preferably, the process of generating the multi-scale adaptive grid unit and establishing the spatial index is as follows: Performing initial blocking on the image according to the spatial position and the resolution, and calculating texture entropy, direction gradient statistics and edge sharpness index for each blocking; carrying out quantitative evaluation on the importance of the segmented region by combining the saliency response map, and dividing the image into different scale levels according to the space detail density, the self-similarity and the structural complexity in the region; And generating a multi-scale self-adaptive grid unit by adopting a self-adaptive grid generation strategy aiming at each scale level, recording grid numbers, geographic ranges, level identifications and neighborhood relations, and establishing a spatial index. Preferably, the loading of the corresponding multi-temporal images based on the multi-scale adaptive grid and the spatial index comprises the following steps: extracting image blocks corresponding to grid units of different scales from a multi-temporal image dataset according to grid numbers and geographic ranges recorded by the spatial index; performing geometric correction, brightness standardization and cloud and fog region elimination on the image; and organizing and caching the image blocks according to the scale level of the multi-scale self-adaptive grid. Preferably, the process of generating the multi-scale spatio-temporal feature sequence is: Based on multi-temporal image blocks corresponding to each scale grid, performing convolutional coding, structural texture description, direction gradient sequence extraction and local brightness differential analysis on the image blocks in each scale grid unit to obtain a primary feature set of each time node; the primary feature sets of the same grid unit a