A rapid adaptive disaster level map generalization method based on semantic representation
By adaptively adjusting the kernel size through k-means clustering and opening/closing operations, the core disaster area is dynamically extracted and a disaster level map is generated. This solves the problem of low efficiency in disaster information extraction after satellite remote sensing detection, and enables rapid and accurate disaster information expression, supporting disaster emergency response.
Patent Information
- Application Number
- CN202311709866.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-12-13
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2043-12-13
AI Technical Summary
The extraction of disaster information after satellite remote sensing detection faces challenges such as the difficulty in extracting semantic information due to the large and diverse data volume, and the lack of a universal method to adapt to all situations, as different types of disasters exhibit different spatial distribution characteristics.
k-means clustering is used to extract disaster patches. The kernel size is adaptively adjusted by opening and closing operations to dynamically extract the core disaster area. A disaster level map is generated by classifying and synthesizing disaster levels. Finally, patches with an area smaller than the threshold are selected to generate the final disaster map.
It enables rapid and effective disaster information extraction, reduces the computational load of semantic information conversion, improves the efficiency of map semantic expression, and provides rapid and accurate information support for disaster emergency response.
Smart Images

Figure CN117726951B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of disaster management and emergency response, and specifically relates to a rapid synthesis method for adaptive disaster level maps based on semantic expression. Background Technology
[0002] Satellite remote sensing has significant advantages in terms of wide coverage and continuous spatial observation, making it the most effective means of detecting large-scale surface anomalies. However, extracting semantic information from satellite detection results in real time after a disaster event to provide users with disaster information via text broadcasts and to assist relevant departments and affected people with ground-based emergency support faces enormous challenges. These challenges are mainly reflected in the following two aspects. First, the disaster raster data acquired by remote sensing is massive, and the disaster information is too rich; extracting semantic knowledge directly from it without simplification is a huge challenge. Second, different types of disasters are diverse and exhibit different spatial distribution characteristics, making it extremely challenging to adapt a single method to all situations. Summary of the Invention
[0003] Purpose of the invention: In order to overcome the shortcomings of the existing technology, this invention provides a rapid synthesis method for adaptive disaster level maps based on semantic expression, aiming to improve the efficiency of map semantic expression and provide rapid and effective information support for disaster emergency response.
[0004] Technical Solution: To achieve the above objectives, this invention provides a rapid synthesis method for adaptive disaster level maps based on semantic expression, comprising the following steps:
[0005] S1: Extracting disaster patches: Based on the disaster raster map, k-means clustering is used to extract disaster patches to remove the influence of background values;
[0006] S2: Adaptive extraction of disaster core area: Using opening and closing operations, the size of the operation kernel is dynamically adjusted according to the change of the number of patches to achieve adaptive extraction of disaster core area, ensuring that the location of the disaster patches is accurate and the number of patches is small.
[0007] S3: Disaster Level Classification: Using the obtained disaster core area as a mask, extract the raster data of the area to be integrated, and classify the disaster level according to the value of the raster data and the number of categories n, generating a disaster map that is reclassified into n levels;
[0008] S4: Generation and fusion of comprehensive disaster results: Using opening and closing operations, disaster results for each level are generated from high to low level. During this process, the size of the operation kernel is adjusted according to the disaster level, and the disasters of each level are superimposed and fused to finally generate a comprehensive disaster level map.
[0009] S5: Filter patches based on area: On the integrated disaster level map, calculate the area of each patch and delete patches with an area smaller than a set threshold to generate the final disaster map for semantic representation.
[0010] The specific threshold for the area can be set by the user; the default value is row×col×1%.
[0011] Furthermore, the specific process of extracting the affected patches in step S1 is as follows:
[0012] S1-1: Read the maximum value of the raster value, denoted as max. value ;
[0013] S1-2: Read the image in row row and column col as an array of row × col row 1 column, where the values of the array are the values of the corresponding raster cells. Then perform k-means clustering on the array to divide the data into two cluster categories, 0 and 1.
[0014] S1-3: Max value The corresponding category is labeled as the disaster patch and assigned a value of 1, while the other category is assigned a value of 0, resulting in a disaster clustering diagram.
[0015] Furthermore, the order of opening and closing operations in step S2 is as follows: first, the opening operation is performed, that is, the erosion operation is performed first and then the dilation operation is performed to remove small noise in the image and fill small holes in the patches. On this basis, the closing operation is performed to connect adjacent patches and smooth the boundary of the disaster area.
[0016] The computational kernel is circular in shape, with an initial radius of r.
[0017] Furthermore, the process of adaptively extracting the core disaster area in step S2 is as follows:
[0018] S2-1: Set the shape and initial size of the kernel, kernel_size, set the threshold for the number of patches, num, and perform the first opening and closing operation on the disaster clustering graph;
[0019] S2-2: Calculate the number of patches in the first opening and closing operation. If the number of patches is less than or equal to num, the result of the first opening and closing operation is the disaster core area. If the number of patches is greater than num, increase kernel_size, perform the second opening and closing operation, and then calculate the number of patches again. Repeat the adjustment of kernel size until the number of patches is less than or equal to num. The disaster core area extraction is completed, and the final value of the extraction result kernel_size is returned.
[0020] The threshold value `num` for the number of patches is a positive integer greater than or equal to 1. Generally, a smaller `num` value is beneficial for subsequent semantic representation. However, in some cases, such as when disaster patches are widely dispersed and geographically distant, an excessively small `num` value may reduce the processing efficiency of this method. Therefore, in such cases, it is advisable to moderately increase the value of `num` to improve processing efficiency.
[0021] Furthermore, the disaster level classification in step S3 is selected based on the user's needs or data characteristics, and the classification method specifically includes:
[0022] The equal-interval grading method divides the data range into several equal intervals, each with an equal width. This method is suitable for relatively uniform data distribution, but may not be flexible enough when dealing with extreme values.
[0023] The natural breakpoint grading method is suitable for data with obvious distribution patterns and helps to highlight the inherent structure of the data, but it may not be suitable for uniformly distributed data.
[0024] The custom interval hierarchy method allows users to define the interval for each range. This method is suitable for situations where users want to define the interval intervals based on specific needs or domain knowledge.
[0025] The standard deviation grading method allows users to customize the intervals for each range. This method is suitable for situations where users want to define the interval intervals based on specific needs or domain knowledge.
[0026] Quantile grading divides data into equal percentages or quantiles, typically quartiles (25%, 50%, 75%, and 100%).
[0027] Further, the method for generating disaster results in step S4 is as follows: extract the highest level, i.e., level n, and perform an open-close operation on it. Mark the pixels in the operation result as 1 and the remaining pixels as 0 to generate a disaster map of level n. Then, extract the pixels of level n-1 and perform an open-close operation on them. Similarly, mark the pixels in the operation result as 1 and the remaining pixels as 0 to generate a disaster map of level n-1. Process the remaining level n-2 and n-3 pixels according to the above steps until the pixels of level 1 are processed. Finally, generate a total of n disaster maps of different levels, each map corresponding to a specific level of disaster situation.
[0028] Furthermore, the method for superimposing and fusing disasters of different levels in step S4 is as follows: superimposing the generated n disaster maps of different levels, that is, adding the corresponding pixels to obtain the final disaster level map.
[0029] Furthermore, the method for setting the kernel size of the opening and closing operation in step S4 is as follows: when integrating the nth level disaster, the kernel size of the opening and closing operation is the kernel_size returned when extracting the core disaster area. As the level decreases, the kernel size increases, and the increment can be specified by the user, with a default value of 5 pixels.
[0030] Beneficial Effects: Compared with existing technologies, this invention fully utilizes the advantages of map generalization, aiming to efficiently simplify map information. Unlike methods relying on massive neural network models, this invention employs a dynamic adjustment strategy based on image processing and computational kernels. By dynamically adjusting the size of the computational kernels, it achieves adaptive mapping for different disaster levels. By reducing the number of image patches through generalization, the computational load for image-to-semantic information conversion is reduced, shortening the semantic information extraction time. Furthermore, overly complex and fragmented patches are not conducive to the semantic expression of images. This invention achieves the generalization of numerous discrete patches in an image, enabling rapid filtering and expression of the most important and essential disaster information, improving the efficiency of map semantic expression, and providing rapid and effective information support for disaster emergency response. Attached Figure Description
[0031] Figure 1 This is an overall flowchart of the method of the present invention;
[0032] Figure 2 This is the original disaster map of the 2020 Muli fire in an embodiment of the present invention;
[0033] Figure 3 These are schematic diagrams of the core disaster area and disaster level classification diagrams in embodiments of the present invention;
[0034] Figure 4 These are the integrated disaster level map and the final result map used for semantic expression in this embodiment of the invention. Detailed Implementation
[0035] The present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. After reading this invention, any modifications of the invention in various equivalent forms by those skilled in the art will fall within the scope defined by the appended claims.
[0036] like Figure 1 As shown, this invention provides a rapid synthesis method for adaptive disaster level maps based on semantic expression, comprising the following steps:
[0037] S1: Extracting disaster patches: Based on the disaster raster map, k-means clustering is used to extract disaster patches to remove the influence of background values;
[0038] The specific process includes the following steps:
[0039] S1-1: Read the maximum value of the raster value, denoted as max. value ;
[0040] In this embodiment, the disaster raster map is obtained by inversion from remote sensing images, and its characteristic is that the values are continuous;
[0041] S1-2: Read the image in row row and column col as an array of row × col row 1 column, where the values of the array are the values of the corresponding raster cells. Then perform k-means clustering on the array to divide the data into two cluster categories, 0 and 1.
[0042] S1-3: Max value The corresponding category is labeled as the disaster patch and assigned a value of 1, while the other category is assigned a value of 0, resulting in a disaster clustering diagram.
[0043] S2: Adaptive extraction of disaster core area: Using opening and closing operations, the size of the operation kernel is dynamically adjusted according to the change of the number of patches to achieve adaptive extraction of disaster core area, ensuring that the location of the disaster patches is accurate and the number of patches is small.
[0044] The specific process is as follows:
[0045] S2-1: Set the shape and initial size of the kernel, kernel_size, set the threshold for the number of patches, num, and perform the first opening and closing operation on the disaster clustering graph;
[0046] In this embodiment, the processing kernel is circular, its initial size kernel_size is set to 5, and the patch number threshold num is set to 3;
[0047] S2-2: Calculate the number of patches in the first opening and closing operation. If the number of patches is less than or equal to 3, the result of the first opening and closing operation is the disaster core area. If the number of patches is greater than 3, the value of kernel_size needs to be further increased to further aggregate patches and reduce the number of patches. The kernel_size is increased by kernel_size + step (step is the kernel growth step size, which is set to a fixed value of 5 in this embodiment). After increasing, the second opening and closing operation is performed, and the number of patches is calculated again. The kernel size is adjusted repeatedly until the number of patches is less than or equal to 3. The disaster core area is extracted and the final value of the kernel_size is returned.
[0048] S3: Disaster Level Classification: Using the obtained disaster core area as a mask, extract the raster data of the area to be integrated, and classify the disaster level according to the value of the raster data and the number of categories n, generating a disaster map that is reclassified into n levels;
[0049] The disaster level classification method used in this embodiment is equal-interval classification, and the number of categories is set to 5;
[0050] S4: Generation and fusion of comprehensive disaster results: Using opening and closing operations, disaster results for each level are generated from high to low level. During this process, the size of the operation kernel is adjusted according to the disaster level, and the disasters of each level are superimposed and fused to finally generate a comprehensive disaster level map.
[0051] The method for setting the kernel size of the opening and closing operation in the comprehensive process of each level is as follows: When the nth level disaster is comprehensive, the kernel size of the opening and closing operation is the kernel_size returned when the core disaster area is extracted. As the level decreases, the kernel size increases, and the increment can be specified by the user. In this embodiment, the kernel_size grows linearly, and the increment is 5 each time.
[0052] The specific process is as follows:
[0053] S4-1: Extract the highest level, i.e., level n, and perform an opening-closing operation on it. Mark the pixels in the result as 1 and the rest as 0 to generate the level n disaster map. Then, extract the pixels with levels greater than or equal to n-1 and perform opening-closing operations on them. Similarly, mark the pixels in the result as 1 and the rest as 0 to generate the level n-1 disaster map. Process the remaining level n-2 and n-3 pixels in the same way until the level 1 pixels are processed. Finally, generate a total of n disaster maps of different levels, each map corresponding to a specific level of disaster situation.
[0054] S4-2: Overlay the generated n disaster maps of different levels, that is, add the corresponding pixels to obtain the final disaster level map.
[0055] S5: Patch Filtering by Area: Small patches are filtered by area to reduce the number of patches, remove features that have little impact on semantic expression, and improve the computational efficiency of the semantic information extraction process. In this embodiment, the specific threshold for area is row×col×0.01. On the integrated disaster level map, the area of each patch is calculated, and patches with an area smaller than row×col×0.01 are deleted to generate the final disaster map used for semantic expression.
[0056] To verify the effectiveness of the method of the present invention, this embodiment applies the above scheme to a specific disaster: the forest fire that occurred in 2020 at the border of Chutouwan Village, Qiaowa Town, Muli County, Liangshan Prefecture, Sichuan Province, and Xiangjiao Village, Xiangjiao Mongolian Township. The specific disaster situation is as follows: Figure 2As shown, its spatial resolution is 30 meters, and the number of rows and columns are 1346 and 976 respectively, i.e., row = 1346, col = 976. The disaster core area extracted by opening and closing operations in this embodiment is as follows: Figure 3 As shown in (a). In this embodiment, the disaster level is divided into 5 categories, as follows: Figure 3 As shown in (b). Figure 4 (a) is the comprehensive disaster level map. Figure 4 (b) The disaster map generated for semantic representation after deleting small-area patches. The specific environment was Jupyter Notebook and ArcGIS software platform, and Python and related libraries such as NumPy, OpenCV, GDAL, and Matplotlib were used for development experiments.
[0057] To verify the actual effect of the method of the present invention, semantic information expression extraction is performed on the images before and after the synthesis by the method of the present invention. The comparison of the number of patches and semantic extraction time before and after synthesis is shown in the following table:
[0058] Table 1 Comparison of the number of image patches and semantic information extraction time before and after integration.
[0059]
[0060] As can be seen from Table 1, the number of patches in the image after integration by the method of the present invention is much less than that before integration, and the semantic information extraction time of the integrated image is much less than that before integration, thus greatly improving the efficiency of map semantic expression.
Claims
1. A rapid adaptive disaster level map generalization method oriented towards semantic expression, characterized in that, Includes the following steps: S1: Extracting disaster patches: Based on the disaster raster map, k-means clustering is used to extract disaster patches to remove the influence of background values; S2: Adaptive extraction of disaster core area: Using opening and closing operations, the size of the operation kernel is dynamically adjusted by the change in the number of patches to achieve adaptive extraction of disaster core area; S3: Disaster Level Classification: Using the obtained disaster core area as a mask, extract the raster data of the area to be integrated, and classify the disaster level according to the value of the raster data and the number of categories n, generating a disaster map that is reclassified into n levels; S4: Generation and fusion of comprehensive disaster results: Using opening and closing operations, disaster results for each level are generated from high to low level. During this process, the size of the operation kernel is adjusted according to the disaster level, and the disasters of each level are superimposed and fused to finally generate a comprehensive disaster level map. S5: Filter patches based on area: On the integrated disaster level map, calculate the area of each patch and delete patches with an area smaller than a set threshold to generate the final disaster map for semantic representation.
2. The method for rapid generalization of adaptive disaster level maps based on semantic expression as described in claim 1, characterized in that, The specific process for extracting the affected patches in step S1 is as follows: S1-1: Read the maximum value of the raster value, denoted as max. value ; S1-2: Read the image in row row and column col as an array of row × col row 1 column, where the values of the array are the values of the corresponding raster cells. Then perform k-means clustering on the array to divide the data into two cluster categories, 0 and 1. S1-3: Max value The corresponding category is labeled as the disaster patch and assigned a value of 1, while the other category is assigned a value of 0, resulting in a disaster clustering diagram.
3. The rapid synthesis method for adaptive disaster level maps based on semantic expression according to claim 1, characterized in that, The order of opening and closing operations in step S2 is as follows: first, the opening operation is performed, that is, the erosion operation is performed first and then the dilation operation is performed to remove small noise in the image and fill small holes in the patches. On this basis, the closing operation is performed to connect adjacent patches and smooth the boundary of the disaster area. The computational kernel is circular in shape, with an initial radius of r.
4. The method for rapid generalization of adaptive disaster level maps based on semantic expression as described in claim 3, characterized in that, The adaptive extraction of the disaster core area in step S2 is as follows: S2-1: Set the shape and initial size of the kernel, kernel_size, set the threshold for the number of patches, num, and perform the first opening and closing operation on the disaster clustering graph; S2-2: Calculate the number of patches in the first opening and closing operation. If the number of patches is less than or equal to num, the result of the first opening and closing operation is the disaster core area. If the number of patches is greater than num, increase kernel_size, perform the second opening and closing operation, and then calculate the number of patches again. Repeat the adjustment of kernel size until the number of patches is less than or equal to num. The disaster core area extraction is completed, and the final value of the extraction result kernel_size is returned.
5. The method for rapid generalization of adaptive disaster level maps based on semantic expression as described in claim 1, characterized in that, The methods for classifying disaster levels in step S3 include equal interval classification, natural breakpoint classification, custom interval classification, standard deviation classification, and quantile classification.
6. The rapid generalization method for adaptive disaster level maps based on semantic expression according to claim 1, characterized in that, The method for generating disaster results in step S4 is as follows: Extract the highest level, i.e., level n, and perform an open-close operation on it. Mark the pixels in the operation result as 1 and the remaining pixels as 0 to generate a disaster map of level n. Then, extract the pixels of level n-1 and perform an open-close operation on them. Similarly, mark the pixels in the operation result as 1 and the remaining pixels as 0 to generate a disaster map of level n-1. Process the remaining level n-2 and n-3 pixels according to the above steps until all level 1 pixels are processed. Finally, generate a total of n disaster maps of different levels, each map corresponding to a specific level of disaster situation.
7. The rapid synthesis method for adaptive disaster level maps based on semantic expression as described in claim 6, characterized in that, The method for superimposing and fusing disaster maps of different levels in step S4 is as follows: superimpose the generated n disaster maps of different levels, that is, add the corresponding pixels to obtain the final disaster level map.
8. The method for rapid generalization of adaptive disaster level maps based on semantic expression according to claim 6, characterized in that, The method for setting the kernel size of the opening and closing operation in step S4 is as follows: When integrating the nth level disaster, the kernel size of the opening and closing operation is the kernel_size returned when extracting the core disaster area. As the level decreases, the kernel size increases, and the increment can be specified by the user.
Citation Information
Patent Citations
Disaster risk grade classifying method
CN103150474A
Method for flood disaster monitoring and disaster analysis based on vision transformer
US11521379B1
