Urban emergency refuge congestion risk assessment method and system based on image enhancement

Abstract

The invention discloses an urban emergency refuge congestion risk assessment method and system based on image enhancement. The method comprises image acquisition, image enhancement and risk assessment. The invention belongs to the field of image processing, and particularly relates to an urban emergency refuge congestion risk assessment method and system based on image enhancement. According to the scheme, global-local dynamic weights are introduced, and texture information of a dense area and the overall site layout are considered; the fusion of real-time and historical data features is optimized through feature consistency constraint; constructing loss based on the dynamic region weight, and optimizing the urban emergency refuge image; a dynamic area sensing weight is introduced, a crowd dense distribution center is focused, and invalid features of an open area are inhibited; the density perception weight is fused with the multi-scale expansion convolution features, and fine dense region boundary segmentation is achieved; carrying out risk assessment on the real-time urban emergency refuge image based on the dynamic region weight and the historical correlation; and the congestion risk assessment effect is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to an urban emergency shelter crowding risk assessment method and system based on image enhancement. Background Art

[0002] The urban emergency shelter crowding risk assessment method is a comprehensive analysis technology based on real-time and historical data. It is used to identify the crowd distribution status in urban emergency shelters, and through dynamic analysis of crowd density, site conditions and potential risk areas, it assesses the possible crowding risks in the shelters, thereby providing decision support for evacuation and management. However, the general urban emergency shelter crowding risk assessment method has the problem of relying only on real-time images, being significantly affected by lighting and noise, ignoring important details of crowded areas, and thus resulting in poor accuracy of crowding risk assessment results; the general urban emergency shelter crowding risk assessment method has the problem of difficulty in adapting to the characteristic changes of dense and dispersed areas, and capturing the crowd distribution characteristics of the transition from dense to sparse, thus resulting in poor crowding risk assessment results. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides an urban emergency shelter crowding risk assessment method and system based on image enhancement. In view of the problem that the general urban emergency shelter crowding risk assessment method only relies on real-time images, is significantly affected by lighting and noise, ignores important crowded area details, and thus leads to poor accuracy of crowding risk assessment results, this scheme introduces global-local dynamic weights, balances feature extraction priorities, and takes into account texture information in dense areas and overall site layout; optimizes the fusion of real-time and historical data features through feature consistency constraints, and eliminates noise interference; constructs losses based on dynamic area weights to optimize urban emergency shelter images ; Thereby improving the accuracy of crowding risk assessment results; In view of the fact that general urban emergency shelter crowding risk assessment methods are difficult to adapt to the characteristic changes of dense and dispersed areas, and to capture the distribution characteristics of the crowd from dense to sparse transition, which leads to poor crowd risk assessment results, this scheme introduces dynamic area-aware weights, generates weights through a soft attention mechanism, focuses on the center of dense crowd distribution, and suppresses invalid features in open areas; uses density-aware weights to fuse multi-scale dilated convolution features to achieve fine dense area boundary segmentation; and performs risk assessment on real-time urban emergency shelter images based on dynamic area weights and historical correlations; thereby improving the crowding risk assessment effect.

[0004] The technical solution adopted by the present invention is as follows: The method for assessing the risk of crowding in urban emergency shelters based on image enhancement provided by the present invention comprises the following steps:

[0005] Step S1: image acquisition;

[0006] Step S2: image enhancement;

[0007] Step S3: Risk assessment.

[0008] Furthermore, in step S1, the image acquisition is to acquire real-time urban emergency refuge images and corresponding historical urban emergency refuge images, mark high-risk areas, and dynamically weighted fuse the acquired real-time images and historical images to adapt to lighting conditions and noise levels; expressed as: ; ;in, is the input image; is the environmental perception weight; and They are real-time urban emergency shelter images and historical urban emergency shelter images; Brightness is the image brightness; NoiseLevel is the noise level; is the sigmoid activation function; MLP(·) is a multi-layer perceptron.

[0009] Furthermore, in step S2, the image enhancement specifically includes the following steps:

[0010] Step S21: Dynamic feature extraction; specifically: shallow feature extraction, the input image is extracted through the convolution layer shallow features , expressed as: ; Based on global spatial channel attention, the important crowded areas in the image are highlighted and irrelevant background features are suppressed; dynamic global-local weights are generated based on environmental features to control the extraction priority of global features of layout information and local features of details in crowded areas respectively; global and local dynamic weighted representation is: ; ; ; ; Spatial attention weight The calculation is expressed as: ; Get weighted output , expressed as: ; Through non-local feature combination, the global and local features of the crowded area are combined; the feature cascade is expressed as: ; Non-local weighted feature calculation representation: ; ;in, It is a 3×3 convolution kernel; and They are global feature maps and local feature maps respectively; and They are global feature weight and local feature weight respectively; Area is the area of ​​the venue; Density is the crowd density; It is a 1×1 convolution kernel; is global average pooling; It is an element-by-element multiplication operation; It is the feature map after cascading; is the feature cascade operation; W is the non-local feature weight matrix; , and They are feature compression, feature dimensionality reduction and feature transformation; is a matrix multiplication operation; is a normalization operation; is the final non-local feature map;

[0011] Step S22: Feature domain enhancement; specifically: dynamically extract multi-scale context features in crowded areas of the image , introduce dynamic context feature extraction weights , introduce feature consistency constraints; dynamic extraction is expressed as: ; Feature consistency constraints It is expressed as: ; Feature domain denoising, enhancing image clarity and removing artifacts in low-quality areas within the feature domain; Feature fusion, expressed as: ; ; Skip connection optimization, expressed as: ;in, is a downsampling operation; It is a grouped convolution operation; is the square of the L2 norm; It is the fused feature map; is the final optimized feature output; and are the features extracted by the encoder and decoder respectively; is the constraint coefficient; is the constraint weight;

[0012] Step S23: Feature fusion; specifically: generating dynamic saliency weights , expressed as: ;Significant region fusion, expressed as: ; Where SCM(·) is the significant feature module; is the salient feature map after fusion; γ is the scale factor; β is the offset factor; and They are respectively the mapping of non-local features and contextual features after saliency weighting;

[0013] Step S24: loss function design; specifically: define structural similarity , expressed as: ; Introducing dynamic area weights , construct the final loss , expressed as: ; ; where x and They are the real scene image and the generated enhanced image respectively; and They are x and The mean of and They are x and The standard deviation of is the covariance; C1, C2 and C3 are constants; and is the loss weight; and They are the characteristics of high-density areas and low-density areas respectively.

[0014] Furthermore, in step S3, the risk assessment specifically includes the following steps:

[0015] Step S31: Feature extraction: Use dilated convolution to capture the wide range of crowd distribution and global background features in the shelter, and introduce dynamic context-aware weights , expressed as: ; The dynamic features and dilated convolution features are fused and expressed as: ; ;in, is the dilated convolution feature; It is a dilated convolution operation; is the expansion rate; It is the fused multi-feature map;

[0016] Step S32: significant area enhancement; specifically: generating weights through a soft attention mechanism, highlighting key areas, and based on dynamic area perception weights Highlight the dense crowd centers in the shelter; expressed as: ; The attention weight is expressed as: ;in, is the attention weight matrix; is a matrix multiplication operation; is the weighted significant regional feature;

[0017] Step S33: segmentation decoding; specifically: using density-aware weights to control the fusion of multi-scale dilated convolutions, expressed as: ; The decoding module combines multiple convolution operations, expressed as: ; Multi-scale decoding feature fusion, expressed as: ;in, It is a feature map fused with multi-scale dilated convolution; is the density threshold; is the fixed median weight; is the segmentation decoding feature; It is a 7×1 convolution kernel; It is a 11×1 convolution kernel; It is the final fused segmentation feature map; is an upsampling operation;

[0018] Step S34: risk area extraction: identify high-risk areas in the shelter, extract locations and boundaries, and classify them according to the density of the crowd; specifically: use the segmentation results to generate the coordinates of the risk area, expressed as: ;in, is the pixel coordinate set of the high-risk area; (x1, y1) is the pixel position of the image; is the density value; Brightness is the brightness level;

[0019] Step S35: Risk scoring: Calculate the risk score of the image based on the crowd density and area of ​​the dense area , and introduce dynamic regional weights Adjust risk scoring priorities; divide risk scoring ranges, if , it is classified as low risk; , it is classified as medium risk; if , it is classified as high risk; then the urban emergency evacuation crowding risk assessment level is obtained; the risk score is expressed as:

[0020] ; ; ; ; Where N and M are the number of high-risk areas in real-time images and historical images, respectively, and i is the index of the high-risk area; is the area; is the historical risk score; is the i-th high-risk area of ​​the image; and They are historical relevance and real-time relevance; is the characteristic distance between historical data and current data; and is the risk score threshold.

[0021] The urban emergency shelter crowding risk assessment system based on image enhancement provided by the present invention comprises an image acquisition module, an image enhancement module and a risk assessment module;

[0022] The image acquisition module acquires real-time urban emergency refuge images and corresponding historical urban emergency refuge images, and sends the data to the image enhancement module;

[0023] The image enhancement module enhances the clarity and detail retention of the image and highlights the important features of the crowded area through dynamic feature extraction, multi-stage feature domain enhancement and significant area fusion; and sends the data to the risk assessment module;

[0024] The risk assessment module uses multi-scale feature extraction and dynamic attention mechanism to achieve fine segmentation of dense areas and accurate identification of high-risk areas.

[0025] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0026] (1) In view of the problem that general urban emergency shelter crowding risk assessment methods only rely on real-time images, are significantly affected by lighting and noise, and ignore important details of crowded areas, which leads to poor accuracy of crowding risk assessment results, this scheme introduces global-local dynamic weights to balance feature extraction priorities, taking into account the texture information of dense areas and the overall site layout; optimizes the fusion of real-time and historical data features through feature consistency constraints, and eliminates noise interference; constructs losses based on dynamic regional weights to optimize urban emergency shelter images; and thus improves the accuracy of crowding risk assessment results.

[0027] (2) In view of the fact that general urban emergency shelter crowding risk assessment methods are difficult to adapt to the characteristic changes of dense and dispersed areas and capture the distribution characteristics of the crowd from dense to sparse, which leads to poor crowd risk assessment results, this scheme introduces dynamic area-aware weights and generates weights through a soft attention mechanism to focus on the center of dense crowd distribution while suppressing invalid features in open areas. It uses density-aware weights to fuse multi-scale dilated convolution features to achieve fine dense area boundary segmentation. It also performs risk assessment on real-time urban emergency shelter images based on dynamic area weights and historical correlations, thereby improving the crowd risk assessment effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 A schematic diagram of the process of the urban emergency shelter crowding risk assessment method based on image enhancement provided by the present invention;

[0029] Figure 2 A schematic diagram of an urban emergency shelter crowding risk assessment system based on image enhancement provided by the present invention;

[0030] Figure 3 It is a schematic diagram of the process of step S3.

[0031] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0032] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0033] In the description of the present invention, it is necessary to understand that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the referred system or element must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as a limitation on the present invention.

[0034] Example 1, see Figure 1 The present invention provides an urban emergency shelter crowding risk assessment method based on image enhancement, which comprises the following steps:

[0035] Step S1: Image acquisition: acquiring real-time urban emergency refuge images and corresponding historical urban emergency refuge images;

[0036] Step S2: Image enhancement: Through dynamic feature extraction, multi-stage feature domain enhancement and significant area fusion, the clarity and detail retention of the image are enhanced, and the important features of the crowded area are highlighted;

[0037] Step S3: Risk assessment: Use multi-scale feature extraction and dynamic attention mechanism to achieve fine segmentation of dense areas and accurate identification of high-risk areas.

[0038] Example 2, see Figure 1 , this embodiment is based on the above embodiment. In step S1, the image acquisition specifically includes acquiring real-time urban emergency refuge images and corresponding historical urban emergency refuge images, marking high-risk areas, and dynamically weighted fusion of the acquired real-time images and historical images to adapt to lighting conditions and noise levels; it is expressed as: ; ; In low-light and noisy environments, dynamically enhance the role of historical data; in high-quality image scenes, more emphasis is placed on real-time data to improve scene adaptability; Among them, is the input image; is the environmental perception weight; and They are real-time urban emergency shelter images and historical urban emergency shelter images; Brightness is the image brightness; NoiseLevel is the noise level; is the sigmoid activation function; MLP(·) is a multi-layer perceptron.

[0039] Example 3, see Figure 1 This embodiment is based on the above embodiment. In step S2, image enhancement specifically includes the following steps:

[0040] Step S21: Dynamic feature extraction; used to extract dynamic features from the input urban emergency shelter image, including texture and edge information of densely distributed areas, as well as global background features of the site layout in the image and local detail features of crowded areas; specifically: shallow feature extraction, the input image is extracted through the convolution layer to extract shallow features , expressed as: ; Based on global spatial channel attention, the important crowded areas in the image are highlighted and irrelevant background features are suppressed; dynamic global-local weights are generated based on environmental features to control the extraction priority of global features of layout information and local features of details in crowded areas respectively; global and local dynamic weighted representation is: ; ; ; ; Spatial attention weight The calculation is expressed as: ; Get weighted output , expressed as: ; Through non-local feature combination, the global and local features of the crowded area are combined; the feature cascade is expressed as: ; Non-local weighted feature calculation representation: ; ;in, It is a 3×3 convolution kernel; and They are global feature maps and local feature maps respectively; and They are global feature weight and local feature weight respectively; Area is the area of ​​the venue; Density is the crowd density; It is a 1×1 convolution kernel; is global average pooling; It is an element-by-element multiplication operation; It is the feature map after cascading; is the feature cascade operation; W is the non-local feature weight matrix; , and They are feature compression, feature dimensionality reduction and feature transformation; is a matrix multiplication operation; is a normalization operation; is the final non-local feature map;

[0041] Step S22: Feature domain enhancement: Enhance the clarity and robustness of image features through multiple stages, eliminate noise, and retain the contours of people and local details of dense textures; specifically: dynamically extract multi-scale context features in crowded areas of the image , introduce dynamic context feature extraction weights , enhance the feature clarity in noisy and complex scenes, introduce feature consistency constraints, and optimize the fusion quality of historical data and real-time data; dynamic extraction is expressed as: ; Feature consistency constraints It is expressed as: ; Feature domain denoising, enhancing image clarity and removing artifacts in low-quality areas within the feature domain; Feature fusion, expressed as: ; ; Skip connection optimization, expressed as: ;in, is a downsampling operation; It is a grouped convolution operation; is the square of the L2 norm; is the fused feature map; is the final optimized feature output; and are the features extracted by the encoder and decoder respectively; is the constraint coefficient; is the constraint weight;

[0042] Step S23: feature fusion; used to fuse image features of different scales to highlight the saliency of crowded areas; specifically: generate dynamic saliency weights , expressed as: ;Significant region fusion, expressed as: ; Where SCM(·) is the significant feature module; is the salient feature map after fusion; γ is the scale factor; β is the offset factor; and They are respectively the mapping of non-local features and contextual features after saliency weighting;

[0043] Step S24: loss function design; combine noise suppression in low light and occlusion conditions and feature enhancement in the crowd salient area to optimize the image quality; specifically: define structural similarity , expressed as: ; Introducing dynamic area weights , construct the final loss , expressed as: ; ; where x and They are the real scene image and the generated enhanced image respectively; and They are x and The mean of and They are x and The standard deviation of is the covariance; C1, C2 and C3 are constants; and is the loss weight; and They are the characteristics of high-density areas and low-density areas respectively.

[0044] By performing the above operations, in view of the problem that general urban emergency shelter crowding risk assessment methods only rely on real-time images, are significantly affected by lighting and noise, ignore important details of crowded areas, and thus lead to poor accuracy of crowding risk assessment results, this scheme introduces global-local dynamic weights to balance feature extraction priorities, taking into account texture information in dense areas and overall site layout; through feature consistency constraints, optimizes the fusion of real-time and historical data features, and eliminates noise interference; constructs losses based on dynamic regional weights to optimize urban emergency shelter images; and thereby improves the accuracy of crowding risk assessment results.

[0045] Example 4, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. In step S3, risk assessment specifically includes the following steps:

[0046] Step S31: Feature extraction: Use dilated convolution to capture the wide range of crowd distribution and global background features in the shelter, and introduce dynamic context-aware weights , expressed as: ; The dynamic features and dilated convolution features are fused and expressed as: ; ;in, is the dilated convolution feature; It is a dilated convolution operation; is the expansion rate; It is the fused multi-feature map;

[0047] Step S32: salient area enhancement; dynamically focus on the densely distributed center of the crowd in the image, while suppressing irrelevant walls and open areas; specifically: generate weights through the soft attention mechanism, highlight the key areas, and perceive weights based on dynamic areas Highlight the dense crowd centers in the shelters and suppress irrelevant areas; expressed as: ; The attention weight is expressed as: ;in, is the attention weight matrix; is a matrix multiplication operation; is the weighted significant regional feature;

[0048] Step S33: Segmentation decoding: Use the symmetric convolution module to extract the features in the horizontal and vertical directions, realize fine-grained segmentation, refine the segmentation boundaries of dense areas, and identify the dense transition areas between people; specifically: use density-aware weights to control the fusion of multi-scale dilated convolutions, expressed as: ; The decoding module combines multiple convolution operations, expressed as: ; Multi-scale decoding feature fusion, expressed as: ;in, It is a feature map fused with multi-scale dilated convolution; is the density threshold; is the fixed median weight; is the segmentation decoding feature; It is a 7×1 convolution kernel; It is a 11×1 convolution kernel; It is the final fused segmentation feature map; is an upsampling operation;

[0049] Step S34: risk area extraction: identify high-risk areas in the shelter, extract locations and boundaries, and classify them according to the density of the crowd; specifically: use the segmentation results to generate the coordinates of the risk area, expressed as: ;in, is the pixel coordinate set of the high-risk area; (x1, y1) is the pixel position of the image; is the density value; Brightness is the brightness level;

[0050] Step S35: Risk scoring: Calculate the risk score of the image based on the crowd density and area of ​​the dense area , and introduce dynamic regional weights Adjust risk scoring priorities; divide risk scoring ranges, if , it is classified as low risk; , it is classified as medium risk; if , it is classified as high risk; then the urban emergency evacuation crowding risk assessment level is obtained; the risk score is expressed as:

[0051] ; ; ; ; Where N and M are the number of high-risk areas in real-time images and historical images, respectively, and i is the index of the high-risk area; is the area; is the historical risk score; is the i-th high-risk area of ​​the image; and They are historical relevance and real-time relevance; is the characteristic distance between historical data and current data; and is the risk score threshold.

[0052] By performing the above operations, the general urban emergency shelter crowding risk assessment method has the problem that it is difficult to adapt to the characteristic changes of dense and dispersed areas, capture the crowd distribution characteristics of the transition from dense to sparse, and thus lead to poor crowd risk assessment results. This scheme introduces dynamic area-aware weights, generates weights through a soft attention mechanism, focuses on the center of dense crowd distribution, and suppresses invalid features in open areas; uses density-aware weights to fuse multi-scale dilated convolution features to achieve fine dense area boundary segmentation; and performs risk assessment on real-time urban emergency shelter images based on dynamic area weights and historical correlations, thereby improving the crowding risk assessment effect.

[0053] Example 5, see Figure 2 , this embodiment is based on the above embodiment, and the urban emergency shelter crowding risk assessment system based on image enhancement provided by the present invention includes an image acquisition module, an image enhancement module and a risk assessment module;

[0054] The image acquisition module acquires real-time urban emergency refuge images and corresponding historical urban emergency refuge images, and sends the data to the image enhancement module;

[0055] The image enhancement module enhances the clarity and detail retention of the image and highlights the important features of the crowded area through dynamic feature extraction, multi-stage feature domain enhancement and significant area fusion; and sends the data to the risk assessment module;

[0056] The risk assessment module uses multi-scale feature extraction and dynamic attention mechanism to achieve fine segmentation of dense areas and accurate identification of high-risk areas.

[0057] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0058] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0059] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. An urban emergency shelter crowding risk assessment method based on image enhancement, characterized by: The method comprises the following steps: Step S1: Image acquisition: acquiring real-time urban emergency refuge images and corresponding historical urban emergency refuge images; Step S2: image enhancement; Step S3: risk assessment; Step S2 includes step S21: dynamic feature extraction; specifically: shallow feature extraction, the input image is extracted through the convolution layer shallow features , expressed as: ; Based on global spatial channel attention, the important crowded areas in the image are highlighted and irrelevant background features are suppressed; dynamic global-local weights are generated based on environmental features to control the extraction priority of global features of layout information and local features of details in crowded areas respectively; global and local dynamic weighted representation is: ; ; ; ; Spatial attention weight The calculation is expressed as: ; Get weighted output , expressed as: ; Through non-local feature combination, the global and local features of the crowded area are combined; the feature cascade is expressed as: ; Non-local weighted feature calculation representation: ; ;in, It is a 3×3 convolution kernel; and They are global feature maps and local feature maps respectively; and They are global feature weight and local feature weight respectively; Area is the area of ​​the venue; Density is the crowd density; It is a 1×1 convolution kernel; is global average pooling; It is an element-by-element multiplication operation; It is the feature map after cascading; is the feature cascade operation; W is the non-local feature weight matrix; , and They are feature compression, feature dimensionality reduction and feature transformation; is a matrix multiplication operation; is a normalization operation; is the final non-local feature map; is the input image; is the sigmoid activation function.

2. The method for assessing crowding risk of urban emergency shelters based on image enhancement according to claim 1 is characterized in that: In step S2, the image enhancement further comprises the following steps: Step S22: Feature domain enhancement; specifically: dynamically extract multi-scale context features in crowded areas of the image , introduce dynamic context feature extraction weights , introduce feature consistency constraints; dynamic extraction is expressed as: ; Feature consistency constraints It is expressed as: ; Feature domain denoising, enhancing image clarity and removing artifacts in low-quality areas within the feature domain; Feature fusion, expressed as: ; ; Skip connection optimization, expressed as: ;in, is a downsampling operation; It is a grouped convolution operation; is the square of the L2 norm; It is the fused feature map; is the final optimized feature output; and are the features extracted by the encoder and decoder respectively; is the constraint coefficient; is the constraint weight; is the dynamic context feature extraction weight; and They are real-time urban emergency shelter images and historical urban emergency shelter images; is a feature consistency constraint; Step S23: Feature fusion; specifically: generating dynamic saliency weights , expressed as: ;Significant region fusion, expressed as: ; Where SCM(·) is the significant feature module; is the salient feature map after fusion; γ is the scale factor; β is the offset factor; and They are the mapping of non-local features and contextual features after saliency weighting; NoiseLevel is the noise level; Step S24: loss function design.

3. The method for assessing the risk of crowding in urban emergency shelters based on image enhancement according to claim 2 is characterized in that: In step S24, the loss function is specifically designed as follows: Defining structural similarity , expressed as: ; Introducing dynamic area weights , construct the final loss , expressed as: ; ; where x and They are the real scene image and the generated enhanced image respectively; and They are x and The mean of and They are x and The standard deviation of is the covariance; C1, C2 and C3 are constants; and is the loss weight; and They are the characteristics of high-density areas and low-density areas respectively.

4. The method for assessing crowding risk of urban emergency shelters based on image enhancement according to claim 3 is characterized in that: In step S3, the risk assessment specifically includes the following steps: Step S31: Feature extraction: Use dilated convolution to capture the wide range of crowd distribution and global background features in the shelter, and introduce dynamic context-aware weights , expressed as: ; The dynamic features and dilated convolution features are fused and expressed as: ; ;in, is the dilated convolution feature; It is a dilated convolution operation; is the expansion rate; It is the fused multi-feature map; Step S32: significant area enhancement; specifically: generating weights through a soft attention mechanism, highlighting key areas, and based on dynamic area perception weights Highlight the dense crowd centers in the shelter; expressed as: ; The attention weight is expressed as: ;in, is the attention weight matrix; is a matrix multiplication operation; is the weighted significant regional feature; Step S33: segmentation decoding; specifically: using density-aware weights to control the fusion of multi-scale dilated convolutions, expressed as: ; The decoding module combines multiple convolution operations, expressed as: ; Multi-scale decoding feature fusion, expressed as: ;in, It is a feature map fused with multi-scale dilated convolution; is the density threshold; is the fixed median weight; is the segmentation decoding feature; It is a 7×1 convolution kernel; It is a 11×1 convolution kernel; It is the final fused segmentation feature map; is an upsampling operation; Step S34: risk area extraction: identify high-risk areas in the shelter, extract locations and boundaries, and classify them according to the density of the crowd; specifically: use the segmentation results to generate the coordinates of the risk area, expressed as: ;in, is the pixel coordinate set of the high-risk area; (x1, y1) is the pixel position of the image; is the density value; Brightness is the brightness level; Step S35: Risk scoring.

5. The method for assessing crowding risk of urban emergency shelters based on image enhancement according to claim 4 is characterized in that: In step S35, the risk score is specifically: Calculate the risk score of the image based on the crowd density and area in the dense area , and introduce dynamic regional weights Adjust risk scoring priorities; divide risk scoring ranges, if , it is classified as low risk; , it is classified as medium risk; if , it is classified as high risk; then the urban emergency evacuation crowding risk assessment level is obtained; the risk score is expressed as: ; ; ; ; Where N and M are the number of high-risk areas in real-time images and historical images, respectively, and i is the index of the high-risk area; is the area; is the historical risk score; and They are historical relevance and real-time relevance; is the i-th high-risk area of ​​the image; is the characteristic distance between historical data and current data; and is the risk score threshold.

6. The method for assessing crowding risk of urban emergency shelters based on image enhancement according to claim 1 is characterized by: In step S1, the image acquisition specifically includes acquiring real-time urban emergency refuge images and corresponding historical urban emergency refuge images, marking high-risk areas, and dynamically weighted fusion of the acquired real-time images and historical images to adapt to lighting conditions and noise levels; it is expressed as: ; ;in, is the input image; is the environmental perception weight; Brightness is the image brightness; NoiseLevel is the noise level; is the sigmoid activation function; MLP(·) is a multi-layer perceptron.

7. An urban emergency shelter crowding risk assessment system based on image enhancement, used to implement an urban emergency shelter crowding risk assessment method based on image enhancement as claimed in any one of claims 1 to 6, characterized in that: It includes image acquisition module, image enhancement module and risk assessment module; The image acquisition module acquires real-time urban emergency refuge images and corresponding historical urban emergency refuge images, and sends the data to the image enhancement module; The image enhancement module enhances the clarity and detail retention of the image and highlights the important features of the crowded area through dynamic feature extraction, multi-stage feature domain enhancement and significant area fusion; and sends the data to the risk assessment module; The risk assessment module uses multi-scale feature extraction and dynamic attention mechanism to achieve fine segmentation of dense areas and accurate identification of high-risk areas.

Images