Multi-type anomaly on-orbit detection system and method

Abstract

The present application relates to the technical field of anomaly feature recognition, in particular to a multi-type anomaly on-orbit detection system and method, the system comprising: a map collaborative clipping module, a redundant partition identification module, a wave band boundary discrimination module, a sensitive feature generation module and an anomaly structure construction module.In the present application, by extracting the spatial positioning information of remote sensing images and performing image alignment, a pixel overlap distribution matrix is constructed, the adaptive division of map boundaries can be realized based on the frequency concentration area, the accuracy of image area management is improved, combined with main image extraction and boundary matching degree discrimination, the pertinence of anomaly area wave band recognition is effectively enhanced, multi-wave band gray values are combined into three-dimensional vectors, gray vector mapping is generated by coupling spatial position, the fusion expression of spectral and spatial features of thermal anomaly areas is realized, the spatially continuous gray anomaly area is aggregated by connectivity recognition, a stable anomaly structure distribution is constructed, and the recognition accuracy and adaptability in complex change scenarios are improved.

Description

Technical Field

[0001] This invention relates to the field of anomaly feature recognition technology, and in particular to an on-orbit detection system and method for multiple types of anomalies. Background Technology

[0002] The field of anomaly feature recognition technology mainly involves the identification and processing of abrupt, abnormal, or abnormal changes in land features in remote sensing images. Its core includes image data preprocessing (such as radiometric correction and geometric correction), target area localization and compression, classification sample construction, and model- or rule-based image content analysis. High-performance computing platforms combined with fixed algorithm processes enable rapid identification of various surface anomalies. Traditional multi-type anomaly on-orbit detection systems refer to remote sensing image processing devices deployed on satellite platforms, primarily targeting surface anomalies such as fires, landslides, floods, and red tides. These systems employ computing modules such as GPUs or NPUs, utilizing image radiometric correction methods, geometric localization models, CXP interface image parsing, pre-set classification sample libraries, and anomaly detection workflows to complete real-time processing of raw on-board data and multi-type anomaly identification.

[0003] Current technologies for identifying surface anomalies in remote sensing imagery rely on fixed algorithmic processes and pre-defined classification sample libraries. The processing path is primarily static, lacking dynamic adaptability to regional changes. Image processing often involves dividing the entire image into sections, failing to dynamically delineate effective region boundaries based on frequency distribution. This leads to map division errors in areas experiencing urban expansion or geomorphic changes, reducing the accuracy of anomaly identification. Furthermore, the use of a uniform pre-defined sample library for matching during image recognition makes it difficult to provide effective anomaly discrimination when multi-source images have inconsistent band coverage or significant variations in imaging conditions, resulting in boundary misidentification or false positives. In addition, current technologies lack methods for joint feature modeling of grayscale values ​​across multiple bands, preventing the full extraction of spatial correlations between grayscale anomalies across different bands and reducing sensitivity to complex thermal anomaly areas. In multi-class anomaly identification tasks, fixed processes often require repetitive construction of processing chains, lacking the ability to aggregate structural anomalies, thus limiting processing efficiency and stability in areas with complex structures or widely distributed anomalies. For example, when abnormal phenomena such as fires or landslides develop rapidly, image processing capabilities cannot adjust strategies in real time, which can easily lead to delayed recognition or structural misjudgment, affecting the timeliness and accuracy of early warning responses. Summary of the Invention

[0004] To address the technical problems existing in the prior art, embodiments of the present invention provide a multi-type anomaly on-orbit detection system and method. The technical solution is as follows:

[0005] On the one hand, it provides a multi-type anomaly on-orbit detection system, which includes: The map sheet collaborative cropping module acquires multiple remote sensing satellite images of the urban expansion area, extracts spatial positioning information and performs image alignment, counts the number of pixel occurrences, constructs an overlap distribution matrix, identifies frequency concentration areas based on the overlap, extracts boundary coordinates, and forms a map sheet division block coordinate set. The redundant partition identification module divides the map sheet into block coordinate sets, counts the number of times images appear in the map sheet blocks, identifies the most frequent image, combines the number with the map sheet block number, and generates a main image allocation table. The band boundary discrimination module extracts the band image of the main image of the desertification area based on the main image allocation table, and extracts the boundary by combining the abnormal area mask, calculates the spatial offset and overlap between the band and the abnormal boundary, and generates a band combination list. Based on the band combination list, the sensitive feature generation module extracts the corresponding band images in the thermal anomaly area, obtains the gray values ​​of the abnormal pixels, constructs a three-dimensional gray vector and summarizes the coordinates to generate an abnormal gray vector mapping set. The abnormal structure construction module integrates abnormal vectors in the map block based on the abnormal grayscale vector mapping set, identifies spatially clustered pixel sets and determines connectivity, and forms an abnormal spatial distribution structure set.

[0006] As a further embodiment of the present invention, the map sheet division block coordinate set includes boundary coordinates, map sheet block number, and frequency concentration area; the main image allocation table includes image number, map sheet block number, and main image identifier; the band combination list includes spatial offset degree, boundary overlap degree, and band number; the abnormal gray-level vector mapping set includes three-dimensional gray-level vector, pixel coordinates, and corresponding band; and the abnormal spatial distribution structure set includes connected regions, map sheet block index, and spatial clustering features.

[0007] As a further aspect of the present invention, the image collaborative cropping module includes: The positioning and extraction submodule acquires multiple remote sensing satellite image data in the urban expansion area, extracts the spatial positioning information and pixel distribution range of the images, calls the original geographic boundary parameters, performs alignment operations on the geographic pixel positions between images, and generates an image alignment boundary coordinate set. The overlap analysis submodule, based on the image alignment boundary coordinate set, marks the number of times each pixel in the image appears in multiple images, calls the pixel number and marking result information, constructs a pixel overlap distribution matrix, calculates the overlap frequency distribution characteristics of pixels in the region, and generates pixel overlap frequency distribution values. The boundary generation submodule identifies the set of boundary pixels in the frequency concentration area based on the pixel overlap frequency distribution value, extracts the spatial position relationship to construct a continuous boundary line, and generates a map sheet division block coordinate set by combining the distribution trend between boundary points.

[0008] As a further aspect of the present invention, the redundant partition identification module includes: The frequency extraction submodule obtains the coordinate set of the map sheet division blocks, extracts the image number and corresponding pixel distribution in each map sheet block, calculates the number of times the image appears based on the pixel coverage of the image number in the map sheet block, and generates the image appearance count value within the map sheet. The image filtering module filters the number of times an image appears in the map sheet based on the number of times the image appears within the map sheet, identifies the image number with the highest frequency, extracts the corresponding map sheet number, and generates the main image number value of the map sheet. The main image generation submodule calls the main image number value of the map sheet, combines the map sheet number with the corresponding image number, unifies the number field format and arrangement order, and obtains the main image allocation table.

[0009] As a further aspect of the present invention, the band boundary discrimination module includes: The band extraction submodule obtains the main image allocation table, extracts the image number of the corresponding map sheet in the desertified area, obtains the band image data corresponding to the image number, calls the abnormal area mask, extracts the corresponding boundary information in the band image within the mask range, and generates the band boundary position value. The boundary offset submodule extracts abnormal boundary position data based on the band boundary position value, calculates the boundary position offset distance between the two in the map coordinate system, counts the number of overlapping areas under the boundary of the difference band and judges the overlap situation, and obtains the boundary spatial relationship coefficient. The combined sorting submodule calls the boundary space relationship coefficient to jointly sort the offset distance value and boundary overlap amount of the band under the difference map sheet, establishes the band combination sequence according to the sorting position, and obtains the band combination list.

[0010] As a further aspect of the present invention, the sensitive feature generation module includes: The image extraction submodule obtains the band combination list, extracts the image number corresponding to the band combination in the thermal anomaly area, locates the band sequence position of the image number in the original image, extracts the corresponding image data according to the band sequence position, obtains the image pixel value within the mask range of the thermal anomaly area, and generates a regional band pixel matrix. The gray-scale combination submodule extracts the gray-scale value corresponding to each abnormal region pixel in three bands according to the region band pixel matrix, constructs a gray-scale vector according to the three gray-scale values, summarizes the gray-scale vectors and spatial coordinates of all pixels, establishes the correspondence between gray-scale and coordinates, and obtains a three-dimensional gray-scale vector set. The mapping construction submodule calls the three-dimensional grayscale vector set, maps the dimension values ​​and corresponding coordinates in the grayscale vectors to a spatial matrix, integrates multiple sets of coordinate points to form a spatial projection distribution of a continuous region, and obtains an abnormal grayscale vector mapping set.

[0011] As a further aspect of the present invention, the abnormal structure construction module includes: The vector integration submodule extracts the pixel vector information corresponding to the abnormal areas in the map sheet based on the abnormal grayscale vector mapping set, organizes the grayscale vectors and spatial coordinates of the pixels according to the map sheet number, integrates the vector data of the map sheet, and generates the total number of abnormal vectors of the map sheet. The clustering identification submodule calls the total number of map sheet anomaly vectors, calculates the spatial distance value and gray-scale fitting value between pixels, filters the set of pixels that simultaneously meet the requirements of location proximity and gray-scale fitting according to the clustering threshold, extracts the pixel group with spatial clustering characteristics, and obtains the spatial clustering pixel coefficient. The connectivity filtering submodule identifies continuous pixel regions with correlation based on the spatial clustering pixel coefficients, determines whether the number of connections of pixels in the region in the four neighboring directions meets the connectivity requirements, and combines the connected regions of the map sheet blocks to obtain the abnormal spatial distribution structure set.

[0012] As a further aspect of the present invention, the remote sensing satellite image refers to the surface remote sensing image data acquired by an on-orbit remote sensing satellite equipped with a multispectral, panchromatic, or thermal infrared imaging device, in GeoTIFF or HDF standard geocoded image format, and originating from a publicly available remote sensing image database or commercial satellite platform. The spatial positioning information refers to the registration coordinate data in the image used to identify the geographical location of pixels, usually in a geographic coordinate system or a projected coordinate system; The overlapping distribution matrix refers to a two-dimensional matrix constructed by statistically analyzing the number of times pixels appear in the same surface area from multiple remote sensing images. The map sheet block refers to a regular sub-region formed by cropping the original remote sensing image according to spatial boundaries. The main image refers to the remote sensing image with the largest coverage area or the highest frequency of appearance in a single map block. The main image allocation table refers to a table structure that records the mapping relationship between map blocks and corresponding main image numbers.

[0013] As a further aspect of the present invention, the band image refers to image data in a remote sensing image corresponding to the electromagnetic spectrum segment of the target, including blue light, red light, near infrared, and mid-infrared. The abnormal region mask refers to a binary image layer used to identify regions with abnormal features in remote sensing images; The spatial offset refers to the spatial distance between the band boundary contour and the true boundary of the abnormal region. The band combination list refers to a set of multiple band numbers selected based on the degree of matching between the band and the boundary of the abnormal area; The three-dimensional grayscale vector refers to a three-dimensional numerical vector composed of the grayscale values ​​of a single pixel in three sensitive bands. The abnormal gray-level vector mapping set refers to the set that records the three-dimensional gray-level vectors and spatial coordinates corresponding to all pixels in the abnormal region; The abnormal spatial distribution structure set refers to the overall structural coordinate set composed of spatially continuous abnormal regions identified in multiple map blocks.

[0014] On the other hand, a multi-type anomaly on-orbit detection method, which is based on the aforementioned multi-type anomaly on-orbit detection system, includes the following steps: S1: Acquire multiple remote sensing satellite image data within the urban expansion area, extract spatial positioning information and pixel location information from the images, register the images based on time sequence, record the number of times each pixel appears in the difference images, establish a spatial overlap distribution matrix, identify areas with concentrated frequency and extract edge coordinates, and generate a map sheet division block coordinate set. S2: Extract the image number and occurrence frequency within each map block according to the map block coordinate set, identify the image with the highest occurrence frequency in each block, and combine it with the corresponding map block number to generate a main image allocation table; S3: Extract multiple band images of the main image of the desertification area according to the main image allocation table, extract the boundary mask information of the abnormal area, compare the positional offset and overlap of the band boundary and the abnormal boundary, sort them according to the degree of offset and overlap, and generate a band combination list. S4: Call the bands listed in the band combination list to obtain the corresponding image data, extract the multi-band gray values ​​of pixels in the thermal anomaly area, construct a three-dimensional gray vector and record the corresponding coordinates, and generate an anomaly gray vector mapping set. S5: Extract the pixel vectors of abnormal regions within the map block based on the abnormal grayscale vector mapping set, identify the set of pixels with spatial clustering characteristics, construct continuous regions based on connectivity judgment, and generate an abnormal spatial distribution structure set.

[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, by extracting spatial positioning information from remote sensing images and performing image alignment, a pixel overlap distribution matrix is ​​constructed. This enables adaptive division of map boundaries based on frequency concentration areas, improving the accuracy of image region management. Combined with main image extraction and boundary matching degree discrimination, the targeting of anomaly region band identification is effectively enhanced. Multi-band gray values ​​are combined into a three-dimensional vector, coupled with spatial location to generate gray-scale vector mapping, realizing the fusion expression of spectral and spatial features of thermal anomaly regions. By identifying and aggregating spatially continuous gray-scale anomaly regions through connectivity identification, a stable anomaly structure distribution is constructed, improving the identification accuracy and adaptability in complex and changing scenarios. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a system flowchart of the present invention; Figure 2 This is a schematic diagram of the system framework of the present invention; Figure 3 This is a flowchart of the image collaborative cropping module in this invention; Figure 4 This is a flowchart of the redundant partition identification module in this invention; Figure 5 This is a flowchart of the band boundary discrimination module in this invention; Figure 6 This is a flowchart of the sensitive feature generation module in this invention; Figure 7 This is a flowchart of the abnormal structure construction module in this invention; Figure 8 This is a flowchart of the on-orbit detection method for multiple types of anomalies provided in the embodiments of the present invention. Detailed Implementation

[0018] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0019] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0020] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0021] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0022] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0023] This invention provides on-orbit detection systems for multiple types of anomalies, such as... Figure 1-2 The diagram shown illustrates a multi-type anomaly on-orbit detection system. The system includes: The map sheet collaborative cropping module acquires image data from multiple remote sensing satellites in the urban expansion area, extracts the spatial positioning information of the images and performs image alignment operations, marks the occurrence frequency of each pixel in the image and constructs an overlap distribution matrix, identifies frequency concentration areas based on the overlap distribution of pixels in the region, uses the region as the dividing boundary, and extracts the boundary coordinates to form a map sheet dividing block coordinate set; Remote sensing satellite image data refers to surface remote sensing image data acquired by multispectral, panchromatic, or thermal infrared imaging equipment carried by in-orbit remote sensing satellites. The format is GeoTIFF or HDF standard geocoded image format, and it comes from publicly available remote sensing image databases or commercial satellite platforms. It is used for spatial information extraction and surface anomaly identification. Spatial positioning information refers to the registration coordinate data in an image used to identify the geographical location of pixels. It is usually in geographic coordinate system (WGS84) or projected coordinate system (UTM) to ensure the spatial alignment accuracy between different images. The overlap distribution matrix is ​​a two-dimensional matrix constructed by statistically analyzing the number of times pixels appear in the same surface area of ​​multiple remote sensing images. It is used to represent the degree of data overlap and the location of redundant areas between images. Boundary delineation refers to spatial dividing lines set in the map sheet based on the frequency of image overlap or the degree of pixel redundancy, serving as the basis for map sheet block construction and region cropping; The redundant partition identification module extracts the frequency of images in each map block based on the coordinate set of the map block division, identifies the image with the highest frequency in the map block, and combines the corresponding image number with the map block number to form the main image allocation table; A map sheet block refers to a regular sub-region formed by cropping the original remote sensing image according to spatial boundaries. It is the basic spatial unit used for partition recognition and region processing. The master image refers to the remote sensing image with the largest coverage area or the highest frequency of appearance in a single map sheet, which serves as the main data source for regional analysis. The main image allocation table is a table structure that records the mapping relationship between map sheet blocks and corresponding main image numbers. It is used for calling main images and reading band data in the module. The band boundary discrimination module extracts the band images of the main image in the desertified area according to the main image allocation table, and extracts the corresponding boundaries by combining the anomaly area mask. It identifies the spatial offset and boundary overlap between the band boundary and the anomaly boundary, sorts the offset and overlap, and forms a band combination list. Band imagery refers to image data in remote sensing images corresponding to the electromagnetic spectrum of a target, including blue light, red light, near infrared, and mid-infrared. It is the basic data for constructing multidimensional ground feature characteristics. Anomaly mask is a binary image layer used to identify regions with anomalous features in a remote sensing image. 1 represents an anomalous region and 0 represents a background region, which is used to limit the range of feature extraction. Spatial offset refers to the spatial distance between the band boundary profile and the actual boundary of the anomalous area, and is used to measure the consistency between the band and the actual anomalous area. Boundary overlap refers to the percentage of pixels that overlap between the band boundary and the anomalous mask boundary, and is used to determine the overlap capability of the boundary contour. The band combination list refers to a set of multiple band numbers selected based on the degree of matching between the band and the boundary of the abnormal region, which is used for gray-scale combination operations of the abnormal feature vector; The sensitive feature generation module extracts the image data of the corresponding band in the thermal anomaly area based on the band combination list, extracts the gray values ​​of the pixels in the anomaly area in the three bands, combines them into a three-dimensional gray vector, summarizes the vectors and coordinates corresponding to all pixels, and generates an anomaly gray vector mapping set. A three-dimensional grayscale vector is a three-dimensional numerical vector composed of the grayscale values ​​of a single pixel in three sensitive bands, used to represent the reflection characteristics of a pixel in multiple bands. An abnormal grayscale vector mapping set refers to a set that records the three-dimensional grayscale vectors and spatial coordinates corresponding to pixels in all abnormal regions, and is used for the spatial reconstruction of abnormal structures; The abnormal structure construction module integrates the pixel vectors of abnormal regions in the map block according to the abnormal gray-scale vector mapping set, identifies the set of pixels with spatial clustering characteristics, performs connectivity judgment on the clustered regions that meet the spatial continuity requirements, and uses the connected regions in each map block as key abnormal structures to combine them to obtain the abnormal spatial distribution structure set. Spatial clustering features refer to the clustering trend of a group of pixels in spatial distribution, which is used to identify concentrated reaction structures within anomaly areas on the Earth's surface. Spatial continuity refers to the spatial connection state of several pixels in an image, indicating that there is a connection relationship between multiple pixels, either by edges or corners, which serves as a basic condition for the formation of structural blocks. Anomaly spatial distribution structure set refers to the overall structural coordinate set composed of spatially continuous anomalous regions identified in multiple map sheets, and is the core identification result output by the system in orbit.

[0024] The map sheet division block coordinate set includes boundary coordinates, map sheet block number, and frequency concentration area; the main image allocation table includes image number, map sheet block number, and main image identifier; the band combination list includes spatial offset degree, boundary overlap degree, and band number; the abnormal gray-level vector mapping set includes three-dimensional gray-level vector, pixel coordinates, and corresponding band; and the abnormal spatial distribution structure set includes connected regions, map sheet block index, and spatial clustering features.

[0025] Specifically, such as Figure 2 , 3 As shown, the map sheet collaborative cropping module includes: The positioning and extraction submodule acquires multiple remote sensing satellite image data in the urban expansion area, extracts the spatial positioning information and pixel distribution range of the images, calls the original geographic boundary parameters, performs alignment operations on the geographic pixel positions between images, and generates an image alignment boundary coordinate set. First, determine the latitude and longitude boundaries of the urban expansion area. For example, set the target range as 120.10°E to 120.50°E and 30.20°N to 30.80°N. Based on this boundary, download multi-temporal satellite imagery data from publicly available remote sensing data sources. Commonly used data include Landsat-8, Sentinel-2, and GF-2. Each image comes with a metadata file, which needs to be parsed and extracted to include the geographic coordinates of the top left corner of the image, pixel size, image size, projection information, and coordinate system. For example, an image might have the top left corner coordinates as 500,000 x 3,400,000 x 10m, a resolution of 10m, and an image size of 10980 x 10980. For images from multiple sources, a unified projection system is required, often WGS84 or a corresponding UT. The M-partition performs coordinate transformation and establishes the geographic location mapping relationship of pixels under a unified coordinate system through affine transformation. The actual position of each row and column pixel is determined by its starting coordinates and resolution. When there is a resolution difference, for example, one image is 15m resolution and another is 10m, all images need to be resampled to a unified resolution of 10m. Then, based on the unified projection coordinates, the spatial coverage of each image is calculated and a raster structure within the corresponding geographic range is generated. Combining the effective pixel range of each image, a binary mask can be used to represent the effective area, where the effective position is 1 and the invalid position is 0. All images are registered and boundary aligned according to the unified resolution and coordinates to generate a coordinate set containing the overlapping parts of all images, forming a spatial alignment result under a unified reference system.

[0026] The overlap analysis submodule is based on the image alignment boundary coordinate set, marks the number of times each pixel in the image appears in multiple images, calls the pixel number and marking result information, constructs the pixel overlap distribution matrix, calculates the overlap frequency distribution characteristics of pixels in the region, and generates pixel overlap frequency distribution values. Based on the aligned boundary coordinate set, a unified pixel raster structure is constructed, assigning a unique number to each location. For example, if the analysis area is 10km by 10km and the resolution is 10m, the generated raster contains 1000 rows and 1000 columns, totaling 1 million pixels, with numbers ranging from 0 to 999999. For each image, the existence of valid data at the location corresponding to the number is checked, and a corresponding Boolean matrix of validity is constructed. The Boolean values ​​corresponding to the same location number in multiple images are summed to count how many images contain valid pixel values ​​at that location. For example, the number 874521 in... Data is present in all 7 images, so the frequency is 7. After calculation, a frequency matrix is ​​obtained, recording the cumulative number of occurrences of each pixel. The frequency values ​​are then divided into intervals, for example, 1 to 2 times is level 1, 2 to 4 times is level 2, 4 to 6 times is level 3, and 6 times and above is level 4. The frequency value of each pixel is judged and assigned to the corresponding level. At the same time, a frequency level raster matrix is ​​constructed to represent the stability of each pixel's appearance in all images. The level value of each pixel combined with the number information can generate a frequency level and spatial number correspondence table, forming a complete frequency level distribution map.

[0027] The boundary generation submodule identifies the set of boundary pixels in the frequency concentration area based on the pixel overlap frequency distribution value, extracts the spatial position relationship to construct a continuous boundary line, and generates a map sheet division block coordinate set by combining the distribution trend between boundary points. Based on frequency-level images, a frequency threshold is first set, for example, a threshold of 5. All pixel locations with frequencies not less than this threshold are selected from the frequency matrix and marked as high-frequency regions. This region is then formed into a binary matrix to represent which locations have a high repetition rate. Next, edge extraction is performed on this region. Edge detection operators can be used to identify the outer edge locations of consecutive pixels. The spatial coordinates of all edge pixels are then extracted to form a point set. Using adjacency rules, such as 8-neighborhood relationships, the edge points are sequentially connected to construct continuous boundary lines. If there are significant directional changes in the boundary points, a minimum deviation fitting method can be used to approximate the point set as a fold line. Linear or curved structures, such as edge pixels with horizontal coordinates of 100, 105, 110 and vertical coordinates of 200, 202, 205, indicate that the boundary line has an overall upward trend. This can be fitted into a linear boundary structure. The fitting results are arranged in sequence to form a closed polygon boundary. Finally, these boundary coordinates are grouped according to the logic of map sheet division. For example, if the boundary coordinates of a certain area are in the order of (100, 200), (150, 200), (150, 250), (100, 250), then this closed area can be defined as an image map sheet block. Its coordinate set is output as the map sheet boundary structure for region division processing.

[0028] Specifically, such as Figure 2 , 4As shown, the redundant partition identification module includes: The frequency extraction submodule obtains the coordinate set of map sheet division blocks, extracts the image number and corresponding pixel distribution in each map sheet block, calculates the number of times the image appears based on the pixel coverage of the image number in the map sheet block, and generates the image appearance count value within the map sheet. The specific formula for calculating the pixel coverage of an image within a map block based on its image number is as follows: ; Calculate the composite feature value of image pixel coverage, extract the image number and corresponding pixel distribution in each map block, calculate the number of times the image appears based on the pixel coverage of the image number in the map block, and generate the number of times the image appears in the map block; in, Representative map sheet block Image number The pixel coverage of composite feature values, Represents image number In map sheet blocks Inner The pixel value of a pixel. Represents image number In map sheet blocks Inner The pixel weight coefficient of each pixel, Represents image number In map sheet blocks The average value of all pixels within. Represents image number In map sheet blocks The average weight of all cells within the cell. Indicates the image number In map sheet blocks All cell numbers within Summation, Represents image number In map sheet blocks The total number of pixels contained; For a certain map sheet block Monitor and collect the values ​​of all pixels within it and their corresponding weights, and set... Number the image to which this block belongs. The total number of pixels is obtained through actual pixel statistics; the pixel value of each pixel... Data is read from the sensor or output by the image processing module, such as grayscale values ​​or reflectance values; the weights corresponding to each pixel. The weights are calculated using gradient contrast or edge smoothing algorithms, with the weight range falling between [0, 1] (e.g., weights based on pixel gradient functions); then the average value is calculated. ; And calculate the weighted average term: ; Example Flowchart: Assuming in block In the middle, the image is numbered. In terms of detection These pixels correspond to 1000 pixels; these pixels correspond to 1000 pixels. They are respectively: : , ; : , ; : , ; : , ; : , ; Calculate intermediate terms: ; Therefore, the weighted average term is: ; Calculate the average value again: ; ; Then calculate the summation term within the denominator: For each calculate : - : ; - : ; - : ; - : ; - : ; Summation: ; Denominator of the square root: ; Substitute into the formula: ; The result indicates that in this example, the image numbering... In map sheet blocks The weighted average in the image is consistent with its own weighted average, so its composite feature value is zero. This means that the pixel coverage feature of the image in this map block is no different from its weighted average, so the evaluation of its occurrence frequency in this block does not deviate from the benchmark level. The formula's operational logic is based on a combination of pixel-weighted statistics and difference measures. The numerator, by summing the products of each pixel value and its corresponding weight, expresses the overall weighted contribution of a pixel within a local region. Subtracting the average of these products extracts the deviation of local pixels from the global weighted mean; this operation is equivalent to calculating the offset difference in a statistical sense, thus reflecting the unevenness of pixel coverage distribution. The denominator constructs a composite measure of squared difference and absolute difference. The squared difference reflects the dispersion of pixel values ​​relative to the average pixel value, while the absolute difference reflects the offset of weights within the same region. Together, they measure the variation in pixel intensity and weighted distribution. The squared difference amplifies the dispersion difference through quadratic calculation, while the absolute difference preserves linear offset information, making the overall structure of the denominator both sensitive and stable. The square root operation on the outer layer of the denominator restores the dimensions of the composite difference term to the same level as the numerator, ensuring the result value has comparability and dimensionless properties. Taking the absolute value of the whole is used to avoid symbol interference caused by deviation from the direction, so that the result only reflects the comprehensive difference between the pixel weighted coverage intensity and the distribution dispersion.

[0029] The image filtering module filters the image numbers based on the frequency of their occurrence within the map sheet, identifies the image number with the highest frequency, extracts the corresponding map sheet number, and generates the main image number value for the map sheet. Based on the acquired image occurrence data, an image filtering process is performed for each map sheet block. First, all image numbers and their occurrence lists corresponding to the map sheet number are read. The image number with the highest occurrence in the list is selected as the candidate main image number. If multiple images have the same maximum occurrence, such as two images appearing 3 times within the map sheet, additional filtering conditions are required. For example, images with more recent shooting times can be prioritized, or a number sorting logic can be used to prioritize image numbers with earlier numerical values ​​or higher format standardization. Assuming the image numbers are S2A_20220711 and S2B_20220720, the latter is selected as the main image number. After filtering, each map sheet number and its corresponding main image number are organized into a result data pair. The same filtering process is then performed on all map sheet blocks to generate a complete result set containing all map sheet numbers and their corresponding main image numbers, providing a stable number matching structure for main image generation.

[0030] The main image generation submodule calls the map sheet main image number value, combines the map sheet number with the corresponding image number, unifies the number field format and arrangement order, and obtains the main image allocation table; Based on the result set of map sheet number and main image number, a main image allocation table is generated according to set rules. First, each record corresponding to a map sheet and a main image is read, and the two are formatted and combined. For example, a unified field is generated by concatenating them in the form of "map sheet number_image number". If the map sheet number is X001 and the image number is S2B_20220720, the combination is X001_S2B_20220720. To maintain the standardization of the number field, all number strings are uniformly processed, including character replacement, padding, case uniformity, and time field standardization. For example, the number S2B-22-0720 is adjusted to S2B_20220720. After processing, the data is sorted and organized according to the map sheet number from smallest to largest or according to the image time sequence to ensure that the main image allocation table has searchability and consistency. All map sheet and corresponding image number combinations are summarized to form the main image allocation result. Each record represents the main image information corresponding to a map sheet, providing basic data support for subsequent map sheet management and image distribution.

[0031] Specifically, such as Figure 2 , 5 As shown, the band boundary discrimination module includes: The band extraction submodule obtains the main image allocation table, extracts the image number of the corresponding map sheet in the desertified area, obtains the band image data corresponding to the image number, calls the abnormal area mask, extracts the corresponding boundary information in the band image within the mask range, and generates the band boundary position value. After obtaining the master image allocation table, for each map sheet within the desertified area, its corresponding image number is retrieved one by one. Combining this with the map sheet spatial number range, the image ID associated with each map sheet is extracted from the established image index. For example, map sheets X021, X022, and X023 correspond to image numbers S2A_20210815, S2B_20210910, and S2A_20210820, respectively. Then, the remote sensing data service interface is called to obtain all band image data contained in these image numbers. Common bands include the visible red band, near-infrared band, and mid-infrared band. The red band can be used for vegetation identification, near-infrared reflects vegetation vitality, and mid-infrared is used to identify drought levels. After extracting the band data, anomaly mask data for the same area is overlaid. This mask is based on Boolean... The matrix represents the pixel location of the abnormal region, with the position of 1 indicating the region to be extracted. By matching the pixel coordinates, the band data is mapped to the mask region. The abnormal region portion of the image is cropped according to the row and column numbers. An edge detection algorithm is used to identify gray-level change boundaries on the abnormal region band. The Sobel operator can be used to analyze the gradient changes in the horizontal and vertical directions to identify gray-level abrupt change regions as boundary points. For example, on Band 8, when the gray-level difference between adjacent pixels exceeds 20 units, the point is marked as the boundary. All band images are processed, and the coordinates of the boundary points in each abnormal region are recorded to generate a list of corresponding boundary pixel coordinates in each band. This constitutes the band boundary position value data set, which is used for subsequent boundary comparison and offset analysis.

[0032] The boundary offset submodule extracts abnormal boundary position data based on the band boundary position value, calculates the boundary position offset distance between the two in the map sheet coordinate system, counts the number of overlapping areas under the boundary of the difference band and judges the overlap situation, and obtains the boundary spatial relationship coefficient. Based on the band boundary location values, the set of anomalous boundary points in each band is first extracted, and the spatial coordinates of these points are uniformly transformed to the map sheet projection coordinate system. For example, the UTM projection system is uniformly adopted, with a pixel size of 10m×10m. After transformation, the actual position coordinates (X, Y) of each boundary point are obtained. Then, a certain band is selected as the reference band (e.g., Band 4), and the spatial offset of the boundary points of other bands with their corresponding points is calculated one by one. The Euclidean distance formula D=√[(x2-x1)²+(y2-y1)²] is used to calculate the offset value of each corresponding point pair. For example, if the reference point coordinates are (20000, 3310500) and the comparison point is (20010, 3310510), the calculated offset is 14.1m. The average offset values ​​of all boundary points are averaged to form the average boundary offset distance. If the average offset is 12.4m, within the set offset threshold of 15m... Within the band, the registration is marked as valid. Further, the number of overlapping boundary points between different bands is counted, i.e., the number of points in two bands that overlap at pixel positions or within a single pixel. For example, Band 8 has 1980 overlapping points with the reference band, and Band 11 has 1743. These are divided by the total number of boundary points in the reference band to obtain the boundary overlap rate. For example, if the reference boundary points are 3000, the overlap rates are 66% and 58% respectively. A boundary spatial relationship coefficient is then constructed, calculated as: Spatial relationship coefficient = 0.6 × (1 - average offset / 15) + 0.4 × overlap rate. Substituting the average offset of Band 8 (10.2m) and the overlap rate (72%), the spatial relationship coefficient is 0.6 × (1 - 10.2 / 15) + 0.4 × 0.72 ≈ 0.64. Similar calculations are performed for all bands to form the boundary spatial relationship coefficient value for each band, providing a parameter basis for sorting and combination.

[0033] The combined sorting submodule calls the boundary space relationship coefficient to jointly sort the offset distance value and boundary overlap of the band under the difference map sheet, establishes the band combination sequence according to the sorting position, and obtains the band combination list. Based on the boundary spatial relationship coefficient data, the ranking criteria for all bands are organized sequentially, including the average boundary offset distance and the boundary overlap rate. A joint ranking score is calculated as the ranking standard, defined as: Ranking Score = 0.5 × (1 - Average Offset / 15) + 0.5 × Overlap Rate, where the average offset is in meters (m), the maximum threshold is set to 15m, and the overlap rate is the ratio of 0 to 1. A ranking score is generated for each band according to this formula. For example, Band 8 has an average offset of 10.2m and an overlap rate of 72%, so its score is 0.5 × (1 - 10.2 / 15) + 0.5 × 0.72 = 0.645. All bands are scored accordingly. The bands are sorted in descending order, with higher priority bands listed first. The sorting position of each band and its related attribute values ​​are recorded. For example, Band 8 is sorted first, Band 4 is sorted second, and Band 11 is sorted third, forming a band combination sequence. Then, a band combination list is created, recording fields such as map sheet number, band number, sorting position, average offset distance, overlap rate, and score value, as the output list. The band combination list corresponding to map sheet X021 is: Band 8 (score 0.645), Band 4 (score 0.603), and Band 11 (score 0.572). This list can be used for band priority selection and data retrieval control in subsequent analysis stages.

[0034] Specifically, such as Figure 2 , 6 As shown, the sensitive feature generation module includes: The image extraction submodule obtains a list of band combinations, extracts the image numbers corresponding to the band combinations in the thermal anomaly area, locates the band sequence position of the image number in the original image, extracts the corresponding image data based on the band sequence position, obtains the image pixel values ​​within the mask range of the thermal anomaly area, and generates a regional band pixel matrix. After reading the band combination list, the image number and band combination information corresponding to each map sheet are extracted item by item. For example, the corresponding bands for map sheet X033 are Band 8, Band 11, and Band 12, and the image number is S2B_20210710. Then, the sequence number of each band in the remote sensing data is located through the image metadata. After confirming that Band 8 is layer 4, Band 11 is layer 6, and Band 12 is layer 9, the corresponding image area sub-blocks are cropped according to the spatial location of the map sheet. Then, the mask is extracted by matching the thermal anomaly area mask file. For pixel locations with a median value of 1, extract the pixel grayscale values ​​for each of the three bands. For example, if a certain location corresponds to values ​​of 164, 201, and 189 in the three bands, process all the masked pixels one by one, and then establish a two-dimensional grayscale matrix with pixels as rows and bands as columns. Each row represents a pixel in an anomaly region, and each column represents the grayscale values ​​of Band 8, Band 11, and Band 12, respectively. This matrix records the actual image reflection intensity of all pixels in the thermal anomaly region in the specified bands, forming a regional band pixel matrix.

[0035] The gray-scale combination submodule extracts the gray-scale value corresponding to each anomalous region pixel in three bands based on the regional band pixel matrix, constructs a gray-scale vector according to the three gray-scale values, summarizes the gray-scale vectors and spatial coordinates of all pixels, establishes the correspondence between gray-scale and coordinates, and obtains a three-dimensional gray-scale vector set. Based on the regional band pixel matrix, each row of data in the matrix is ​​traversed sequentially. Each row represents the combination of three band gray values ​​of a pixel in an abnormal region. For example, a row with values ​​of 160, 204, and 190 represents the gray value information of the pixel in Band 8, Band 11, and Band 12. This combination of three values ​​is used as a gray value vector. Then, the row and column number of the pixel in the image is read, and its geospatial coordinates are obtained through projection transformation. For example, if the pixel position is in row 103 and column 225, the coordinates obtained after transformation are 2030 meters and 3310450 meters. The position is bound to the gray value vector and recorded to construct a list containing the association between gray values ​​and spatial coordinates. After processing the entire matrix row by row, a three-dimensional gray value vector set is formed. Each record contains three sets of gray values ​​and one set of two-dimensional spatial coordinates. This set has the ability to completely describe the image data in the abnormal region, providing support for subsequent gray value distribution mapping and spatial aggregation processing.

[0036] The mapping construction submodule calls the three-dimensional grayscale vector set, maps the dimension values ​​and corresponding coordinates in the grayscale vectors to a spatial matrix, integrates multiple sets of coordinate points to form the spatial projection distribution of a continuous region, and obtains the abnormal grayscale vector mapping set. Based on a set of three-dimensional grayscale vectors, according to the spatial coordinates of each vector, the corresponding three sets of band grayscale values ​​are filled into a two-dimensional spatial matrix consistent with the original image area. Each coordinate point corresponds to a three-channel grayscale combination. The entire image area forms a three-dimensional matrix containing the grayscale distribution. The size of the matrix is ​​determined by both the image resolution and the image size. For example, in an image area of ​​1000 rows and 1000 columns, each effective pixel position is filled with the corresponding three-channel grayscale value. Positions not marked by the mask are left empty or with zero values. By traversing all grayscale points, a spatially visualized image is formed. The grayscale data distribution structure is further analyzed for connectivity identification and clustering of these coordinate points. The pixel eight-neighborhood connection method is used to aggregate continuous grayscale similar points. The similarity judgment condition is that the grayscale difference of each of the three grayscale channels does not exceed 20. For example, the grayscale differences of Band8, Band11, and Band12 of adjacent pixels are 15, 18, and 19 respectively, which are considered similar and are classified into the same continuous region. After merging all similar points, one or more sets of continuous spatial regions are constructed, and the set of pixel coordinates covered by these regions is output to form an abnormal grayscale vector mapping set.

[0037] Specifically, such as Figure 2 , 7 As shown, the exception structure construction module includes: The vector integration submodule extracts the pixel vector information corresponding to the abnormal areas in the map sheet based on the abnormal grayscale vector mapping set, organizes the grayscale vectors and spatial coordinates of the pixels according to the map sheet number, integrates the vector data of the map sheet, and generates the total number of abnormal vectors of the map sheet. Based on the abnormal grayscale vector mapping set, the pixel data of the thermal anomaly region in each map sheet block is processed one by one. First, the positions with a mask value of 1 within the map sheet number range are identified. The grayscale vector information of these pixels is obtained through the position index. Each vector contains grayscale values ​​of three channels. Combined with spatial coordinates, the data is recorded in a structured manner. All data is uniformly organized based on the map sheet number. For example, if the map sheet number is S2A_20220801, the corresponding thermal anomaly region contains 1034 valid pixels. The grayscale value combinations such as 155, 198, and 176 and their corresponding spatial coordinates such as 2085 and 3310470 are recorded respectively. Then, all map sheets are processed. The abnormal vector data under each map sheet number is summarized. A mapping structure is constructed through the program structure, and a key-value table is established according to the map sheet number. The key is the map sheet number, and the value is a list set containing grayscale vectors and spatial coordinates. Each item is a set of vectors and coordinates. An efficient storage method is used to facilitate subsequent retrieval. At the same time, the total number of abnormal vectors in each map sheet block is counted. The results can be organized as a comparison between the number and the number. For example, the number S2A_20220718 corresponds to 895 vector records, and the number S2B_20220615 corresponds to 1211 records, forming a complete list of the total number of abnormal vectors in each map sheet. This data provides the basic support for subsequent cluster identification and spatial analysis.

[0038] The clustering identification submodule calls the total number of map sheet anomaly vectors, calculates the spatial distance value and gray-scale fitting value between pixels, filters the set of pixels that simultaneously meet the requirements of location proximity and gray-scale fitting based on the clustering threshold, extracts the pixel group with spatial clustering characteristics, and obtains the spatial clustering pixel coefficient. After obtaining the total number of map sheet anomaly vectors, spatial distance analysis and gray-scale similarity calculation are performed on the abnormal pixels within each map sheet number. First, the distance relationship between pixel positions is constructed, and the relative spacing value is calculated by the difference between the horizontal and vertical coordinates. Then, the three-channel gray-scale values ​​of each pair of pixels are compared to calculate the relative matching degree. If the gray-scale difference is less than a certain range in all channels, for example, the gray-scale difference in each channel is within 20, the gray-scale fitting degree can be considered high. To enhance the screening accuracy, a spatial clustering judgment condition is introduced, setting the spatial distance threshold to 5 meters. The gray-scale fitting threshold is implemented according to the above standard, simultaneously satisfying the condition that the distance is less than or equal to 5 meters and the gray-scale fitting is good. Pixel combinations that meet the specified standards are retained as cluster pairs. Relationship links are established between all pixels that meet the conditions. In the constructed link graph, each edge represents a relationship between pixels. All map sheet numbers are processed separately. By traversing the cluster relationship graph, all pixel groups that form spatial cluster structures are identified. The number of members and spatial range of each group are recorded. At the same time, the cluster pixel coefficient of the map sheet block is calculated, which is the proportion of the number of pixels participating in the cluster to the total number of all abnormal pixels. For example, if there are a total of 1034 abnormal vectors in the map sheet, of which 256 form cluster groups, then the cluster pixel coefficient of the map sheet is 24.8%. This index can be used to quantify the cluster intensity within the map sheet.

[0039] The connectivity filtering submodule identifies continuous pixel regions with correlation based on spatial clustering pixel coefficients, determines whether the number of connections of pixels in the region in the four neighboring directions meets the connectivity requirements, and combines the connected regions of map blocks to obtain the abnormal spatial distribution structure set. Based on the clustered pixel set and corresponding spatial coordinate data calculated in the cluster identification stage, connected region identification is further performed. The four-neighbor principle is used to determine the connectivity between pixels. According to the rules, the four adjacent positions (top, bottom, left, right) of each pixel are traversed to determine if there are pixels belonging to the same cluster group whose grayscale values ​​fluctuate within a set range. For example, pixels with a grayscale difference of no more than 15 from the center pixel in each grayscale channel are considered connected. This process is performed on all cluster groups within the map sheet number. A recursive approach is used, starting from any unprocessed pixel, to find all its connected neighboring pixels, forming an independent connected region. After identifying one region, the process continues... Continue searching for the next unprocessed cell, repeating the operation until the connectivity judgment of all aggregated cells is completed. Record each identified connected region as an independent abnormal structural unit, containing the coordinates and grayscale vector information of all participating cells in the region. Output multiple connected regions using the map sheet number as an index. Each map sheet may contain several connected regions. For example, the number S2A_20220801 identifies 5 connected structures, containing 95, 82, 64, 53 and 48 cells respectively, totaling 342 cells forming a spatial connected structure. The remaining cells are scattered and do not form a contiguous area. All the above identification results are summarized to form an abnormal spatial distribution structure set.

[0040] Please see Figure 8 The multi-type anomaly on-orbit detection method is executed based on the aforementioned multi-type anomaly on-orbit detection system and includes the following steps: S1: Acquire multiple remote sensing satellite image data within the urban expansion area, extract spatial positioning information and pixel location information from the images, register the images based on time sequence, record the number of times each pixel appears in the difference images, establish a spatial overlap distribution matrix, identify areas with concentrated frequency and extract edge coordinates, and generate a map sheet division block coordinate set. S2: Extract the image number and occurrence frequency within each map block according to the map block coordinate set, identify the image with the highest occurrence frequency in each block, and combine it with the corresponding map block number to generate the main image allocation table; S3: Extract multiple band images of the main image of the desertification area according to the main image allocation table, extract the boundary mask information of the abnormal area, compare the positional offset and overlap of the band boundary and the abnormal boundary, sort them according to the degree of offset and overlap, and generate a band combination list. S4: Call the bands listed in the band combination list to obtain the corresponding image data, extract the multi-band gray values ​​of pixels in the thermal anomaly area, construct a three-dimensional gray vector and record the corresponding coordinates, and generate an anomaly gray vector mapping set. S5: Extract the pixel vectors of abnormal regions within the map block based on the abnormal grayscale vector mapping set, identify the set of pixels with spatial clustering characteristics, construct continuous regions based on connectivity judgment, and generate an abnormal spatial distribution structure set.

[0041] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A multi-type anomaly on-orbit detection system, characterized in that, The system includes: The map sheet collaborative cropping module acquires multiple remote sensing satellite images of the urban expansion area, extracts spatial positioning information and performs image alignment, counts the number of pixel occurrences, constructs an overlap distribution matrix, identifies frequency concentration areas based on the overlap, extracts boundary coordinates, and forms a map sheet division block coordinate set. The redundant partition identification module divides the map sheet into block coordinate sets, counts the number of times images appear in the map sheet blocks, identifies the most frequent image, combines the number with the map sheet block number, and generates a main image allocation table. The band boundary discrimination module extracts the band image of the main image of the desertification area based on the main image allocation table, and extracts the boundary by combining the abnormal area mask, calculates the spatial offset and overlap between the band and the abnormal boundary, and generates a band combination list. Based on the band combination list, the sensitive feature generation module extracts the corresponding band images in the thermal anomaly area, obtains the gray values ​​of the abnormal pixels, constructs a three-dimensional gray vector and summarizes the coordinates to generate an abnormal gray vector mapping set. The abnormal structure construction module integrates abnormal vectors in the map block based on the abnormal grayscale vector mapping set, identifies spatially clustered pixel sets and determines connectivity, and forms an abnormal spatial distribution structure set. The term "band image" refers to image data corresponding to the electromagnetic spectrum segment of the target in a remote sensing image, including blue light, red light, near-infrared, and mid-infrared. The "abnormal region mask" refers to a binary image layer used to identify regions with abnormal features in the remote sensing image. The "spatial offset" refers to the spatial distance between the band boundary contour and the true boundary of the abnormal region. The "band combination list" refers to a set of multiple band numbers selected based on the degree of matching between the band and the abnormal region boundary. The "three-dimensional grayscale vector" refers to a three-dimensional numerical vector composed of the grayscale values ​​of a single pixel in three sensitive bands. The "abnormal grayscale vector mapping set" refers to a set recording the three-dimensional grayscale vectors and spatial coordinates corresponding to pixels within all abnormal regions. The "abnormal spatial distribution structure set" refers to the overall structural coordinate set composed of spatially continuous abnormal regions identified in multiple map sheets.

2. The multi-type anomaly on-orbit detection system according to claim 1, characterized in that: The map sheet division block coordinate set includes boundary coordinates, map sheet block number, and frequency concentration area; the main image allocation table includes image number, map sheet block number, and main image identifier; the band combination list includes spatial offset degree, boundary overlap degree, and band number; the abnormal gray-level vector mapping set includes three-dimensional gray-level vector, pixel coordinates, and corresponding band; and the abnormal spatial distribution structure set includes connected regions, map sheet block index, and spatial clustering features.

3. The multi-type anomaly on-orbit detection system according to claim 1, characterized in that: The map sheet collaborative cropping module includes: The positioning and extraction submodule acquires multiple remote sensing satellite image data in the urban expansion area, extracts the spatial positioning information and pixel distribution range of the images, calls the original geographic boundary parameters, performs alignment operations on the geographic pixel positions between images, and generates an image alignment boundary coordinate set. The overlap analysis submodule, based on the image alignment boundary coordinate set, marks the number of times each pixel in the image appears in multiple images, calls the pixel number and marking result information, constructs a pixel overlap distribution matrix, calculates the overlap frequency distribution characteristics of pixels in the region, and generates pixel overlap frequency distribution values. The boundary generation submodule identifies the set of boundary pixels in the frequency concentration area based on the pixel overlap frequency distribution value, extracts the spatial position relationship to construct a continuous boundary line, and generates a map sheet division block coordinate set by combining the distribution trend between boundary points.

4. The multi-type anomaly on-orbit detection system according to claim 3, characterized in that: The redundant partition identification module includes: The frequency extraction submodule obtains the coordinate set of the map sheet division blocks, extracts the image number and corresponding pixel distribution in each map sheet block, calculates the number of times the image appears based on the pixel coverage of the image number in the map sheet block, and generates the image appearance count value within the map sheet. The image filtering module filters the number of times an image appears in the map sheet based on the number of times the image appears within the map sheet, identifies the image number with the highest frequency, extracts the corresponding map sheet number, and generates the main image number value of the map sheet. The main image generation submodule calls the main image number value of the map sheet, combines the map sheet number with the corresponding image number, unifies the number field format and arrangement order, and obtains the main image allocation table.

5. The multi-type anomaly on-orbit detection system according to claim 4, characterized in that: The band boundary discrimination module includes: The band extraction submodule obtains the main image allocation table, extracts the image number of the corresponding map sheet in the desertified area, obtains the band image data corresponding to the image number, calls the abnormal area mask, extracts the corresponding boundary information in the band image within the mask range, and generates the band boundary position value. The boundary offset submodule extracts abnormal boundary position data based on the band boundary position value, calculates the boundary position offset distance between the two in the map coordinate system, counts the number of overlapping areas under the boundary of the difference band and judges the overlap situation, and obtains the boundary spatial relationship coefficient. The combined sorting submodule calls the boundary space relationship coefficient to jointly sort the offset distance value and boundary overlap amount of the band under the difference map sheet, establishes the band combination sequence according to the sorting position, and obtains the band combination list.

6. The multi-type anomaly on-orbit detection system according to claim 5, characterized in that: The sensitive feature generation module includes: The image extraction submodule obtains the band combination list, extracts the image number corresponding to the band combination in the thermal anomaly area, locates the band sequence position of the image number in the original image, extracts the corresponding image data according to the band sequence position, obtains the image pixel value within the mask range of the thermal anomaly area, and generates a regional band pixel matrix. The gray-scale combination submodule extracts the gray-scale value corresponding to each abnormal region pixel in three bands according to the region band pixel matrix, constructs a gray-scale vector according to the three gray-scale values, summarizes the gray-scale vectors and spatial coordinates of all pixels, establishes the correspondence between gray-scale and coordinates, and obtains a three-dimensional gray-scale vector set. The mapping construction submodule calls the three-dimensional grayscale vector set, maps the dimension values ​​and corresponding coordinates in the grayscale vectors to a spatial matrix, integrates multiple sets of coordinate points to form a spatial projection distribution of a continuous region, and obtains an abnormal grayscale vector mapping set.

7. The multi-type anomaly on-orbit detection system according to claim 6, characterized in that: The abnormal structure construction module includes: The vector integration submodule extracts the pixel vector information corresponding to the abnormal areas in the map sheet based on the abnormal grayscale vector mapping set, organizes the grayscale vectors and spatial coordinates of the pixels according to the map sheet number, integrates the vector data of the map sheet, and generates the total number of abnormal vectors of the map sheet. The clustering identification submodule calls the total number of map sheet anomaly vectors, calculates the spatial distance value and gray-scale fitting value between pixels, filters the set of pixels that simultaneously meet the requirements of location proximity and gray-scale fitting according to the clustering threshold, extracts the pixel group with spatial clustering characteristics, and obtains the spatial clustering pixel coefficient. The connectivity filtering submodule identifies continuous pixel regions with correlation based on the spatial clustering pixel coefficients, determines whether the number of connections of pixels in the region in the four neighboring directions meets the connectivity requirements, and combines the connected regions of the map sheet blocks to obtain the abnormal spatial distribution structure set.

8. The multi-type anomaly on-orbit detection system according to claim 1, characterized in that: The remote sensing satellite imagery refers to surface remote sensing image data acquired by multispectral, panchromatic, or thermal infrared imaging equipment carried by in-orbit remote sensing satellites. The format is GeoTIFF or HDF standard geocoded image format, and the data originates from publicly available remote sensing image databases or commercial satellite platforms. The spatial positioning information refers to the registration coordinate data in the image used to identify the geographical location of pixels, usually in a geographic coordinate system or a projected coordinate system; The overlapping distribution matrix refers to a two-dimensional matrix constructed by statistically analyzing the number of times pixels appear in the same surface area from multiple remote sensing images. The map sheet block refers to a regular sub-region formed by cropping the original remote sensing image according to spatial boundaries. The main image refers to the remote sensing image with the largest coverage area or the highest frequency of appearance in a single map block. The main image allocation table refers to a table structure that records the mapping relationship between map blocks and corresponding main image numbers.

9. A method for on-orbit detection of multiple types of anomalies, characterized in that, The multi-type anomaly on-orbit detection system according to any one of claims 1-8 includes the following steps: S1: Acquire multiple remote sensing satellite image data within the urban expansion area, extract spatial positioning information and pixel location information from the images, register the images based on time sequence, record the number of times each pixel appears in the difference images, establish a spatial overlap distribution matrix, identify areas with concentrated frequency and extract edge coordinates, and generate a map sheet division block coordinate set. S2: Extract the image number and occurrence frequency within each map block according to the map block coordinate set, identify the image with the highest occurrence frequency in each block, and combine it with the corresponding map block number to generate a main image allocation table; S3: Extract multiple band images of the main image of the desertification area according to the main image allocation table, extract the boundary mask information of the abnormal area, compare the positional offset and overlap of the band boundary and the abnormal boundary, sort them according to the degree of offset and overlap, and generate a band combination list. S4: Call the bands listed in the band combination list to obtain the corresponding image data, extract the multi-band gray values ​​of pixels in the thermal anomaly area, construct a three-dimensional gray vector and record the corresponding coordinates, and generate an anomaly gray vector mapping set. S5: Extract the pixel vectors of abnormal regions within the map block based on the abnormal grayscale vector mapping set, identify the set of pixels with spatial clustering characteristics, construct continuous regions based on connectivity judgment, and generate an abnormal spatial distribution structure set.

Images