Urban management event detection method, device, equipment, medium and program product

Abstract

The invention discloses a city management event detection method, apparatus and device, a medium and a program product. The method comprises the steps of obtaining target image data of a target city management event; performing feature extraction and convolution processing on the target image data based on a phantom bottleneck module to obtain explicit multiplexing feature data; performing convolution processing on the explicit multiplexing feature data based on a re-parameterization bottleneck module to obtain implicit multiplexing feature data; and performing multi-scale feature prediction on the implicit multiplexing feature data by using a multi-scale attention mechanism in the improved YOLOv8s detection model, and predicting a classification result of the target city management event based on the fused features. According to the embodiment of the invention, the implicit multiplexing data is adopted to extract the multi-scale feature data, and different from a conventional weight calculation process, the detection result is obtained through multi-scale feature detection, so that the data processing efficiency can be improved, the calculation cost can be reduced, and the detection precision of the city management event can be improved.

Description

Technical Field

[0001] This application belongs to the field of computer vision technology, and in particular relates to a method, device, equipment, medium and program product for detecting urban management events. Background Technology

[0002] The YOLOv8 real-time object detection algorithm's network architecture mainly consists of three parts: the backbone network, the neck network, and the head network. YOLOv8 can quickly identify and locate objects in images or videos, providing various configurations and technical strategies for real-time detection technology to meet different application needs.

[0003] Currently, when applying the YOLOv8 model to complex environments, the computational load for complex data increases with the amount of data being detected, leading to problems of high model weights and computational complexity. In the real-time detection of urban behavioral events using the YOLOv8 model, a large number of urban management behavioral events place higher demands on the model's weights. Reducing the model weights and computation accordingly lowers hardware requirements, leading to a qualitative change from a quantitative one. Optimizing the detection model can effectively alleviate computational pressure and reduce urban management costs. However, the low proportion of effective feature weights severely affects detection accuracy, resulting in low accuracy of detection results. Summary of the Invention

[0004] This application provides a method, apparatus, equipment, medium, and program product for detecting urban management incidents, which can improve data processing efficiency, reduce computing costs, and improve the detection accuracy of urban management incidents.

[0005] In a first aspect, embodiments of this application provide a method for detecting urban management events, comprising: acquiring target image data of the target urban management event; performing feature extraction and convolution processing on the target image data based on the phantom bottleneck module in the improved YOLOv8s detection model to obtain explicit reused feature data; performing convolution processing on the explicit reused feature data based on the reparameterization bottleneck module in the improved YOLOv8s detection model to obtain implicit reused feature data, wherein the reparameterization bottleneck module is obtained by structural reparameterization of the phantom bottleneck module; performing multi-scale feature fusion on the implicit reused feature data based on the multi-scale attention mechanism in the improved YOLOv8s detection model, and predicting the classification result of the target urban management event based on the fused features.

[0006] In one possible implementation, after outputting the detection result of the target city management event, the method further includes: determining the event target problem corresponding to the target city management event based on the detection result; statistically analyzing the event target problems corresponding to multiple city management events to determine the target city and marking the event target problem.

[0007] In one possible implementation, acquiring target image data of a target urban management event includes: collecting image information and video information of the target urban management event; extracting corresponding image data from images in the video information at set time intervals; filtering and enhancing the image data and image information to obtain enhanced image data; and performing data annotation processing on the enhanced image data to obtain target image data.

[0008] In one possible implementation, the phantom bottleneck module in the improved YOLOv8s detection model performs feature extraction and convolution processing on the target image data to obtain explicitly reused feature data. This includes: using a first Ghost module stacked in the phantom bottleneck module to perform convolution and activation processing on the target image data, and performing head fusion processing on the processed target image data to generate corresponding first feature map data; using a second Ghost module stacked in the phantom bottleneck module to perform layer-by-layer convolution processing on the first feature map data, and performing head fusion processing on the processed first feature map data to generate second feature map data; and concatenating the first feature map data and the second feature map data to generate explicitly reused feature data.

[0009] In one possible implementation, the step of performing convolution processing on the explicit reused feature data based on the reparameterization bottleneck module in the improved YOLOv8s detection model to obtain implicit reused feature data includes: performing convolution and batch normalization processing on the explicit reused feature data based on the first Rep module in the reparameterization bottleneck module, and performing neck fusion and activation processing on the processed explicit reused feature data to obtain nonlinear feature data; and performing convolution and batch normalization processing on the nonlinear feature data using the second Rep module in the reparameterization bottleneck module, and performing neck fusion processing on the processed nonlinear feature data to obtain implicit reused feature data.

[0010] In one possible implementation, the step of fusing implicitly reused feature data using the multi-scale attention mechanism in the improved YOLOv8s detection model, and predicting the classification result of the target urban management event based on the fused features, includes: performing layer-by-layer convolution processing on the implicitly reused feature data using the multi-scale convolution module in the improved YOLOv8s detection model to obtain local feature data; performing convolution processing on the local feature data using a depth strip convolution module with a target number of features to obtain multi-scale feature data; performing weighted fusion processing on the multi-scale feature data using an attention module, and predicting the classification result of the target urban management event based on the fused features.

[0011] In one possible implementation, before obtaining the implicit reuse feature data, the method further includes: performing convolution processing on the regular sampling points in the target image data to obtain regular feature map data of the center point position of the regular sampling points; performing position offset on the irregular sampling points in the target image data to obtain modified coordinate information; and performing interpolation and resampling on the modified coordinate information to obtain feature map data of the corresponding coordinate positions after variable kernel convolution processing.

[0012] Secondly, embodiments of this application provide a detection device for urban management events. The device includes: an acquisition module for acquiring target image data of a target urban management event; an explicit reuse module for performing feature extraction and convolution processing on the target image data based on the phantom bottleneck module in the improved YOLOv8s detection model to obtain explicit reuse feature data; an implicit reuse module for performing convolution processing on the explicit reuse feature data based on the reparameterization bottleneck module in the improved YOLOv8s detection model to obtain implicit reuse feature data, wherein the reparameterization bottleneck module is obtained by structural reparameterization of the phantom bottleneck module; and a monitoring module for performing multi-scale feature fusion on the implicit reuse feature data based on the multi-scale attention mechanism in the improved YOLOv8s detection model, and predicting the classification result of the target urban management event based on the fused features.

[0013] Thirdly, embodiments of this application provide a detection device for urban management events, the device comprising: a processor and a memory storing computer program instructions; the processor, when executing the computer program instructions, implements the urban management event detection method as described in any one of the first aspects.

[0014] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer program instructions, which, when executed by a processor, implement the method for detecting urban management events as described in any one of the first aspects.

[0015] Fifthly, embodiments of this application provide a computer program product, wherein instructions in the computer program product, when executed by a processor of an electronic device, cause the electronic device to perform the urban management event detection method as described in any one of the first aspects.

[0016] The urban management event detection method, apparatus, device, medium, and program product of this application embodiment acquires target image data; performs feature extraction and convolution processing on the target image data to obtain explicit reused feature data; performs convolution processing on the explicit reused feature data to obtain implicit reused feature data; performs multi-scale feature prediction on the implicit reused feature data, and outputs the detection result of the target urban management event. It can extract multi-scale feature data using implicit reused data, which differs from the conventional weight calculation process. Obtaining the detection result through multi-scale feature detection can improve data processing efficiency, reduce computational costs, and improve the detection accuracy of urban management events. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating a method for detecting urban management incidents provided in an embodiment of this application; Figure 2 This is a flowchart illustrating another method for detecting urban management events provided in an embodiment of this application; Figure 3 This is a schematic diagram of a process for determining target image data provided in an embodiment of this application; Figure 4 This is a schematic diagram of a process for determining explicit reuse feature data provided in an embodiment of this application; Figure 5 This is a flowchart illustrating a process for determining implicit reuse feature data, provided in an embodiment of this application. Figure 6 This is a schematic diagram of another process for determining implicit reuse feature data provided in an embodiment of this application; Figure 7 This is a schematic diagram of a process for determining a test result provided in an embodiment of this application; Figure 8 This is a schematic diagram of another process for determining the detection result provided in an embodiment of this application; Figure 9 This is a schematic diagram of a process for determining feature map data provided in an embodiment of this application; Figure 10 This is a schematic diagram of another process for determining feature map data provided in an embodiment of this application; Figure 11 This is a flowchart illustrating a method for detecting urban management incidents provided in an embodiment of this application; Figure 12 This is a flowchart illustrating a method for detecting urban management incidents provided in an embodiment of this application; Figure 13 This is a schematic diagram of the structure of a detection device for urban management incidents provided in an embodiment of this application; Figure 14 This is a schematic diagram of the structure of a detection device for urban management events provided in an embodiment of this application. Detailed Implementation

[0019] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.

[0020] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0021] The YOLOv8 model is the eighth generation of the YOLO series, representing the latest advancements in deep learning and computer vision. It is widely recognized as a leading model in its class due to its speed and accuracy. The YOLOv8 network architecture primarily consists of three parts: the backbone, the neck, and the head. The backbone uses CSPDarknet53, the neck uses a Path Aggregation Network (PANet), a bidirectional fusion network that operates from top to bottom and bottom to top, achieving a lightweight design while maintaining accuracy. The head uses a decoupled head, learning target location and category information through different network branches and then fusing them.

[0022] The YOLOv8 model incorporates a C2f module in its backbone to enhance gradient flow and improve model performance. The Neck section of the YOLOv8 module fuses feature maps of different scales using bottom-up and top-down paths, enhancing the model's multi-scale detection capabilities. The Head section of the YOLOv8 model is responsible for converting the feature maps extracted from the Backbone and Neck into the final detection results. The regression and prediction branches are separated, and the detection method is changed from Anchor-based to Anchor-free, simplifying the model structure. However, the current YOLOv8 model has high weights and computational cost, resulting in low processing accuracy. This reduces the weight ratio of effective features when applied to the detection of urban management events.

[0023] To address the aforementioned technical problems, embodiments of this application provide a method, apparatus, device, medium, and program product for detecting urban management events. The method involves acquiring target image data; performing feature extraction and convolution processing on the target image data to obtain explicit reused feature data; performing convolution processing on the explicit reused feature data to obtain implicit reused feature data; performing multi-scale feature prediction on the implicit reused feature data; and predicting the classification result of the target urban management event based on the fused features. This method utilizes implicit reused data to extract multi-scale feature data, differing from the conventional weight calculation process. By obtaining detection results through multi-scale feature detection, it improves data processing efficiency, reduces computational costs, and enhances the detection accuracy of urban management events. The method for detecting urban management events provided in this application embodiment is described below.

[0024] Figure 1 This is a flowchart illustrating a method for detecting urban management incidents provided in an embodiment of this application. Figure 1 As shown, the specific steps of the urban management incident detection method include: S101~S105.

[0025] S101. Obtain target image data of the target city management event.

[0026] Target city management events refer to a collective term for a selection of detected events within a chosen city. Target image data can be understood as image or video information related to the detection of target city management events.

[0027] Furthermore, with the increasing demands for urban grassroots governance, the tasks of urban management are becoming increasingly heavy. Urban management is a complex yet crucial task, involving all aspects of the city, including environmental sanitation, traffic management, urban planning, and community development. These tasks may seem trivial, but they are key to the city's operation, impacting citizens' quality of life and the city's image. A large number of urban management events place higher demands on the weights of the detection model. Reducing the model weights and computational complexity correspondingly lowers hardware requirements; quantitative changes lead to qualitative changes, and optimizing the detection model can effectively alleviate computational pressure and reduce urban management costs. To analyze and detect management events in a selected target city, an improved YOLOv8s detection model is used for predictive analysis of management events. First, data collection is required, including collecting image and video information related to management events in the target city. The collected image and video information is then preprocessed to obtain target image data, providing a data basis for the next step of applying the improved YOLOv8s detection model for feature extraction and predictive analysis of management events.

[0028] S102. Based on the phantom bottleneck module in the improved YOLOv8s detection model, feature extraction and convolution processing are performed on the target image data to obtain explicit reuse feature data.

[0029] The Ghost Bottleneck module is a core module in GhostNet, which achieves lightweight feature extraction by stacking Ghost modules.

[0030] Based on the collected target image data of the target city management events, the conventional Bottleneck module was first improved to obtain the Ghost Bottleneck module. The Ghost Bottleneck module was then used to extract features and perform convolution processing on the target image data. After data processing, explicitly reusable feature data was obtained, achieving feature reuse and reducing feature redundancy.

[0031] S103. The reparameterization bottleneck module in the improved YOLOv8s detection model performs convolution processing on the explicit reuse feature data to obtain implicit reuse feature data. The reparameterization bottleneck module is obtained by structural reparameterization of the phantom bottleneck module.

[0032] While convolution and fusion operations using the Ghost Bottleneck module do not increase parameter usage or floating-point computation, they require complex memory copying, incurring significant computational costs for the hardware. Furthermore, a reparameterized bottleneck module, Rep Bottleneck, is created by reparameterizing the Ghost Bottleneck module. This Rep Bottleneck module then performs further convolution on the obtained explicit reused feature data to calculate implicit reused feature data. This implicit reuse reduces parameter usage and floating-point computation, improving detection speed.

[0033] S104. Based on the multi-scale attention mechanism in the improved YOLOv8s detection model, perform multi-scale feature fusion on implicit reuse feature data, and predict the classification results of target urban management events based on the fused features.

[0034] The multi-size attention mechanism mentioned here can be understood as including multi-size convolution and attention modules, used to dynamically weight and adjust the features of the input image data.

[0035] In the process of urban management, in order to carry out urban management work more effectively, there are high requirements for detection accuracy. For example, in the case of street vendor economy in urban management, the high density of night market stalls and the certain degree of mutual occlusion make actual detection difficult and the detection accuracy is low. Therefore, an attention mechanism is added to dynamically adjust the features of the input image to improve the detection accuracy.

[0036] Based on the obtained implicit reused feature data, in order to further improve the detection accuracy of image feature data, a multi-scale attention mechanism processing process is added. Through the multi-scale attention mechanism, multi-scale feature prediction of implicit reused feature data is performed, and the image feature data is further dynamically weighted and adjusted, making the image features more prominent, thereby improving the accuracy of the detection results. This achieves the technical effect of improving data processing efficiency and improving the detection accuracy of urban management events.

[0037] The urban management event detection method of this application involves acquiring target image data; performing feature extraction and convolution processing on the target image data to obtain explicit reused feature data; performing convolution processing on the explicit reused feature data to obtain implicit reused feature data; performing multi-scale feature prediction on the implicit reused feature data; and predicting the classification result of the target urban management event based on the fused features. This method utilizes implicit reused data to extract multi-scale feature data, which differs from the conventional weight calculation process. By obtaining detection results through multi-scale feature detection, it improves data processing efficiency, reduces computational costs, and enhances the detection accuracy of urban management events.

[0038] Figure 2 This is a flowchart illustrating another method for detecting urban management events provided in an embodiment of this application. Figure 2 As shown, the specific steps of the urban management incident detection method include: S201~S202.

[0039] S201. Determine the target issue of the target city management event based on the detection results.

[0040] S202. Statistically analyze the target issues corresponding to multiple urban management events, determine the target cities, and mark the target issues of the events.

[0041] After extracting features from target image data in target city management events using the improved YOLOv8s detection model, various problems leading to these events are identified based on the different detection results. Key factors within these problems are then identified, and this detection result is used as the event target problem causing the management issue—that is, the most significant problem in the target city management event is predicted. Using the same technical methods, event target problems corresponding to the same management events in different cities are detected. Based on the detected key problems, all event target problems detected across multiple city management events are statistically analyzed to determine the target city and label the event target problems.

[0042] For example, regarding the street vendor economy in cities, based on the test results, problems such as illegal street vending, random occupation of roads, improper placement of garbage bins, and overflowing garbage should be reported and backed up in a timely manner, so that law enforcement officers can deal with them promptly. The problems in different areas should be statistically analyzed, and areas with multiple problems and frequent problems should be marked as key areas, making law enforcement work more targeted.

[0043] This application embodiment uses an improved YOLOv8s detection model to extract features from target image data in target city management events, detects key management problems that cause key management problems in the target city, and then marks and manages key management problems by statistically analyzing the management status of the same management problems in multiple cities, thereby achieving the detection and governance effect of city management events.

[0044] Figure 3 This is a schematic diagram of a process for determining target image data provided in an embodiment of this application. Figure 3 This refers to the process of acquiring target image data in step S101. Figure 3 Is Figure 1 The description is based on the corresponding embodiments. Figure 3 The provided diagram illustrates that the process of determining the target image data specifically includes: S301~S304.

[0045] S301. Collect image and video information of target city management events.

[0046] The system utilizes image acquisition devices such as cameras and on-site cameras to capture images of target urban management events and collect image and video data of different urban management events.

[0047] S302. Extract the corresponding image data from the images in the video information according to the set time interval.

[0048] To further normalize the data, the collected video data was organized. For example, Adobe Premiere software was used to extract video data, and KMPlayer software was used to extract the first frame of each second of the video data as sample data, thereby converting the video data into image data.

[0049] S303. The image data and image information are filtered and image enhancement processes are performed to obtain image enhancement data.

[0050] S304. Perform data annotation processing on the image enhancement data to obtain the target image data.

[0051] After the acquired image data was organized, it underwent preprocessing. First, the image data was filtered based on criteria such as sharpness and completeness. Then, the filtered image data was divided into training, validation, and test sets in a 7:2:1 ratio to facilitate subsequent training and validation of the improved YOLOv8s detection model. To improve the network's generalization performance, data augmentation operations were performed on the training set images, including random brightness enhancement, horizontal flipping, and adding Gaussian noise. The Labelme data annotation tool was used for data annotation. Each image generated a JSON file based on bounding box and behavior annotations. The generated JSON files were then converted into TXT files required by the improved YOLOv8s model using Python, yielding the target image data.

[0052] This application embodiment preprocesses the collected image and video data, performs image data filtering, data enhancement, and data annotation to obtain key image information reflecting urban management events, providing a basis for the next step of detecting management events in the image data.

[0053] Figure 4 This is a schematic diagram of a process for determining explicit reuse feature data provided in an embodiment of this application. Figure 4 This refers to the process of determining explicit reuse feature data in step S102. Figure 4 Is Figure 1 The description is based on the corresponding embodiments. Figure 4 The provided diagram illustrates the specific steps in determining the explicit reuse feature data, including steps S401 to S403.

[0054] S401. The first Ghost module stacked in the Phantom Bottleneck module is used to perform convolution and activation processing on the target image data, and the processed target image data is subjected to head fusion processing to generate the corresponding first feature map data.

[0055] S402. The first feature map data is processed by layer-by-layer convolution using the second Ghost module stacked in the Phantom Bottleneck module, and the processed first feature map data is then processed by head fusion to generate the second feature map data.

[0056] S403. The first feature map data and the second feature map data are concatenated to generate explicit reuse feature data.

[0057] The Ghost Bottleneck module primarily consists of two stacked Ghost modules. Based on the obtained target image data, the first Ghost module in the Ghost Bottleneck module performs traditional convolution and activation operations, and then performs head fusion processing on the processed target image data to generate a first feature map with a smaller number of channels. Then, the second Ghost module in the Ghost Bottleneck module performs depthwise convolution operations on the newly generated first feature map data, and performs head fusion processing on the processed first feature map data to generate a second feature map. Finally, the generated second feature map data and the first feature map data obtained in the first step are concatenated to form the final explicit reused feature data.

[0058] This application embodiment improves the C2f module in the improved YOLOv8s detection model by using the first Ghost module of the Ghost Bottleneck module for convolution processing to obtain the first feature map data. Then, the second Ghost module performs convolution processing again to obtain the second feature map data. By concatenating the two feature map data, explicit reuse feature data is obtained, realizing feature reuse, reducing feature redundancy, reducing the number of parameters and floating-point calculations, and achieving network lightweighting while ensuring network performance.

[0059] Figure 5 This is a schematic diagram of a process for determining implicit reuse feature data provided in an embodiment of this application. Figure 5 This refers to the process of determining implicit reuse feature data in step S104. Figure 5 Is Figure 1 The description is based on the corresponding embodiments. Figure 5 The provided diagram illustrates the process of determining implicitly reused feature data, which specifically includes S501~S502.

[0060] S501. Based on the first Rep module in the reparameterization bottleneck module, the explicit reuse feature data is subjected to convolution and batch normalization processing, and the processed explicit reuse feature data is subjected to neck fusion and activation processing to obtain nonlinear feature data.

[0061] S502. The second Rep module in the reparameterization bottleneck module is used to perform convolution and batch normalization on the nonlinear feature data, and the neck fusion processing is performed on the processed nonlinear feature data to obtain implicit reuse feature data.

[0062] While the Ghost Bottleneck module reduces feature redundancy through explicit feature reuse, it requires complex memory copying. To further optimize the Ghost Bottleneck module, structural reparameterization is employed, resulting in the reparameterized Rep Bottleneck module. The first Rep module, with its stacked layers, performs convolution and batch normalization on the explicitly reused feature data calculated by the Ghost Bottleneck module, along with neck fusion and activation processing, to obtain nonlinear feature data. The second Rep module then performs convolution and batch normalization on this nonlinear feature data, along with neck fusion, to obtain implicitly reused feature data. This further reduces the module's weight and improves detection speed.

[0063] In one possible example scenario, Figure 6 This is a schematic diagram illustrating another process for determining implicit reuse feature data provided in an embodiment of this application. According to... Figure 6 The provided diagram illustrates how, to address the issues of high weights and computational complexity, an improved YOLOv8s detection model is used as the baseline network. The Bottleneck structure of the C2f module is modified. To address the high computational cost caused by the numerous convolutional operations in the Bottleneck module, the advantages of the Ghost module are leveraged to achieve feature reuse and reduce feature redundancy, resulting in the Ghost Bottleneck module. The Ghost Bottleneck module uses a head-fusion concat operation instead of a neck-fusion add operation. Specifically, conv performs the convolutional processing, ReLU introduces non-linearity to the neural network, enabling it to learn activation functions for complex features, dconv allows the convolutional kernel to dynamically adjust its sampling position, thus more flexibly adapting to objects of different scales and shapes, significantly improving the model's ability to capture complex visual features, and add is used for neck-fusion convolution processing. The Ghost Bottleneck consists of two stacked Ghost modules. The Ghost module operates in two steps. The first step is to generate a feature map with a small number of channels through traditional convolution operations. The second step is to perform layer-by-layer convolution operations on the newly generated feature map and then concatenate the generated feature map with the feature map obtained in the first step to form the final output.

[0064] Although the head fusion concat operation does not increase parameter usage or floating-point computation, it requires complex memory copying, imposing a significant computational cost on the hardware. Therefore, the Ghost Bottleneck module is further optimized through structural reparameterization, resulting in the reparameterized bottleneck Rep Bottleneck module. By adding batch normalization (BN) to the branches of the first and second Ghost modules, non-linearity is introduced into the training process, and ReLU is shifted to the back, better satisfying the rules of structural reparameterization. Borrowing from MobileNetV2, the second Ghost module no longer uses the ReLU activation function, replacing the less efficient head fusion concat operation with a neck fusion add operation, achieving implicit feature reuse. This reduces parameter usage and floating-point computation while maintaining network performance, further lightweighting the system and improving detection speed.

[0065] This application embodiment obtains explicit multiplexed feature data by using the Ghost Bottleneck module, and then obtains implicit multiplexed feature data by using the derived Rep Bottleneck module, thereby achieving implicit data reuse and lightweighting, and thus improving the technical effect of detection accuracy.

[0066] Figure 7 This is a schematic diagram of a process for determining a test result provided in an embodiment of this application. Figure 7 This refers to the process of determining the test result in step S105. Figure 7 Is Figure 1 The description is based on the corresponding embodiments. Figure 7 The provided diagram illustrates the process for determining the test results, specifically including steps S701 to S703.

[0067] S701. The implicit reused feature data is processed by layer-by-layer convolution using the multi-scale convolution module in the improved YOLOv8s detection model to obtain local feature data.

[0068] S702. Local feature data is processed by convolution of the target number of depth strip convolution modules to obtain multi-scale feature data.

[0069] S703. Use the attention module to perform weighted fusion processing on the multi-scale feature data, and predict the classification result of the target urban management event based on the fused features.

[0070] The multi-scale convolutional attention module consists of a multi-scale convolution module and an attention module. The multi-scale convolution module performs layer-by-layer convolutions on the input implicitly reused feature data to aggregate local information. Then, it performs convolution operations on the implicitly reused feature data using multiple multi-branch depthwise strip convolutions of different sizes to obtain feature data at different scales. The attention module performs channel attention-weighted fusion of the obtained feature data at different scales through pointwise convolutions to obtain the final feature map data. Based on the fused features, it predicts the classification results of the target urban management events.

[0071] The attention weights, obtained by channel attention weighting of feature data at different scales, are represented by Equation 1: Equation 1 in, This represents the input multi-scale feature data. Indicates attention weights, This represents the i-th depth stripe branch. This indicates layer-by-layer convolution.

[0072] Based on the obtained main weights, further fusion is performed to obtain the output of the multi-scale convolutional attention module, which is represented by Equation 2: Equation 2 in, This represents element-wise matrix multiplication. This represents the output of the multi-scale convolutional attention module, i.e., the detection result of the target urban management event.

[0073] The improved YOLOv8s detection model is optimized by adding a multi-scale convolution attention module, which dynamically adjusts the features of the input image to improve detection accuracy.

[0074] In one possible example scenario, Figure 8 This is a schematic diagram of another process for determining detection results provided in an embodiment of this application. According to... Figure 8The provided diagram illustrates how dwconv computes layer-by-layer convolutions. The multi-scale convolution module within the multi-scale convolution attention module performs 5×5 layer-by-layer convolutions on the input feature map data, aggregating local information. Then, it performs convolution operations on the feature map data using three depth strip convolutions of different sizes to obtain contextual features at different scales. Next, the attention module performs channel attention weighting on the obtained feature map data through 1×1 pointwise convolutions, adaptively weighting and fusing feature data at different scales. Finally, the input to the multi-scale convolution attention module is weighted to obtain the final detection result. The multi-branch convolution process uses depth strip convolutions of different sizes instead of depthwise convolutions. For example, inspired by SegNeXt, the kernel size of each branch can be set to 7, 11, and 21 respectively. Since strip convolutions better meet the requirements of lightweight design, what would have required a 7×7 kernel 2D convolution can now be achieved with only a 1×7 and a 7×1 strip convolution, thus reducing the number of parameters.

[0075] This application embodiment adds a multi-scale convolutional attention module to the implicit reuse feature data to dynamically weight and fuse the implicit reuse feature data of the input image, thereby reducing the number of parameters while improving the detection accuracy.

[0076] Figure 9 This is a schematic diagram of a process for determining feature map data provided in an embodiment of this application. Figure 9 This refers to the process of determining the feature map data in step S105. Figure 9 Is Figure 4 Further explanation of the control method corresponding to the previous embodiments. According to... Figure 9 The provided schematic diagram shows that the process of determining the feature map data specifically includes: S901~S902.

[0077] S901. Perform convolution processing on the regular sampling points in the target image data to obtain regular feature map data of the center point position of the regular sampling points.

[0078] S902. The irregular sampling points in the target image data are offset to obtain the modified coordinate information. The modified coordinate information is interpolated and resampled to obtain the feature map data of the corresponding coordinate positions after variable kernel convolution processing.

[0079] Convolutional operations have achieved significant results in deep learning, but traditional convolutional operations have two inherent drawbacks. First, traditional convolutional operations are limited by a sliding window, only able to acquire information within that window and unable to access information from other locations. Furthermore, the window shape is fixed, which does not align with the trend towards adaptive learning. Second, the kernel size of ordinary convolution is k×k. As k increases, the number of parameters increases quadratically, resulting in wasted network overhead due to fixed parameters and sampling shapes. To address these issues, we explore variable kernel convolution (Convolutional Kernel with Arbitrary Sampled Shapes and Arbitrary Number of Parameters). Variable kernel convolution no longer uses fixed parameters and sampling shapes, allowing for an arbitrary number of parameters and arbitrary sampling shapes, providing more options for balancing network overhead and performance. Variable kernel convolution consists of two parts: defining the initial sampling position and the variable convolution operation.

[0080] First, regular sampling point shapes are generated. Then, the remaining irregular sampling points are added in, and convolution calculations are performed using the initial coordinate codes to form the final sampling shape. Convolution processing is performed on the regular sampling points in the target image data to obtain regular feature map data of the center point positions of the regular sampling points. Specifically, the convolution operation using the initial coordinate codes of the convolution kernel is represented by Equation 3: Equation 3 Where P0 represents the initial original coordinate point, and Pn represents the initial coordinates of the sampling point with P0 as the origin for the convolution operation. This represents the weight value corresponding to the relative point Pn.

[0081] Next, the irregular sampling points in the target image data are offset to obtain modified coordinate information. This modified coordinate information is then interpolated and resampled to update the data. Equation 4 is obtained: Equation 4 in, Indicates corresponding points The position offset, Pn represents the initial coordinates of the sampling points with P0 as the origin for the convolution operation. This represents the weight value corresponding to the relative point Pn.

[0082] By processing regular sampling points and using variable kernel convolution for irregular sampling points, convolution processing can be performed on data with any number of parameters and any sampling shape, providing more options for the trade-off between network overhead and performance.

[0083] In one possible example scenario, Figure 10 This is a schematic diagram of another process for determining feature map data provided in an embodiment of this application. According to Figure 10 The provided diagram illustrates that in urban management incident handling, a large number of incidents place higher demands on the weights of the detection model, while also requiring higher detection speeds to meet performance requirements. Using variable kernel convolutions instead of standard convolutions yields more effective feature information, providing more options for balancing network overhead and performance. Figure 10 As shown, the initial position is first defined, and regular-shaped sampling points and irregular sampling points are found. Then, the number of regular sampling points is determined by rounding the square root of the parameters and used for filling. Finally, the remaining sampling points are sampled sequentially from left to right around the next row of regular sampling point shapes. Standard convolution kernels for regular sampling points use the center point as the origin, but convolutions for irregular sampling points often lack a center point. To be applicable to convolutions of all scales, the top-left corner is set as the origin, as irregular sampling coordinates cannot match convolution operations of the corresponding size. Variable kernel convolution obtains modified coordinates through position offset, performs reconstruction processing through Reshap, and obtains features at the corresponding position through interpolation and resampling. The position offset is learned and calculated from the image through ordinary convolution, thus enabling convolution processing with any number of parameters and data of any sampling shape. For example, taking a variable kernel convolution with a kernel size of 7 (N=7), the offset is added to the original coordinates to obtain the modified coordinates, and finally, features at the corresponding position are obtained through interpolation and sampling.

[0084] In one possible example scenario, Figure 11 This is a flowchart illustrating a method for detecting urban management incidents provided in an embodiment of this application. Figure 11As shown in the diagram, Rep-C2f represents the Rep-C2f module based on the C2f module, used for core feature extraction to obtain implicitly reused feature data. MSCA represents the Multi-Scale Convolution Attention (MSCA) module, used for multi-scale convolution processing and attention convolution processing. AK8Conv represents the Convolutional Kernel with Arbitrary SampledShapes and Arbitrary Number of Parameters (AK8conv), used for processing regular sampling points and irregular sampling points, enabling convolution processing on data with arbitrary numbers of parameters and arbitrary sampling shapes. The backbone module in the YOLOv8s model is the core part of the network. Its main function is to extract multi-scale features of the image through a series of convolutional and deconvolutional layers, providing basic feature representations for subsequent object detection tasks. The prediction head module in the YOLOv8s model is the core component responsible for the final object detection task. Its role is to directly output prediction results, including the target's location coordinates, class probability, and confidence score, using features extracted from the backbone and multi-scale information fused with the neck. The neck module integrates the different scale features output from the backbone module into richer semantic information through multi-scale feature fusion. The asymmetric convolutional block (ACBS) module is used to efficiently extract and enhance image features through specific combinations of convolutional operations. The cross-stage partial fusion (C2f) module divides the input features into multiple processing paths, avoiding the gradient vanishing problem in deep network training and making the model more convergent. The upsample module mainly upsamples low-resolution feature maps to restore their spatial resolution for fusion with higher-resolution feature maps, thereby improving the model's ability to detect multi-scale objects. The split module is a core component of the C2f feature fusion module. Its main function is to enhance the interaction and fusion capabilities of features from different stages through feature map segmentation and parallel processing, improving the model's performance in multi-scale object detection. The SILU activation function enhances the nonlinear expressive power of the model by introducing a gating mechanism to optimize the activation process of the feature map, thereby improving the model's ability to extract complex features. The CBS module performs nonlinear transformation and standardization on the input feature map, extracts local features through convolution, and accelerates training and stabilizes gradients through batch normalization. The CLS module is a component for image classification tasks; its core function is to determine the category of the input image and output a confidence score for each category.In the improved YOLOv8s detection model, MaxPool2d primarily reduces the spatial size of the feature map through downsampling, thereby reducing computational cost and enhancing the model's robustness to changes in target scale. Its core principle is to use the maximum value within the pooling window as the output, effectively preserving salient features while suppressing noise.

[0085] according to Figure 11 As shown, the backbone network of the improved YOLOv8s detection model performs conventional convolution and feature extraction operations through the standard asymmetric convolutional block (ACBS) module and the cross-stage partial fusion (C2f) module. The ACBS module includes variable kernel convolution (AK8Conv), normalization (BN), and the activation function SILU. Variable kernel convolution is applied to handle data with arbitrary numbers of parameters and sampling shapes for different numbers of sampling points. The C2f module is further improved by stacking Ghost modules to obtain a Ghost bottleneck, reducing the number of parameters through feature reuse and avoiding feature map redundancy. Further structural reparameterization yields the Rep Bottleneck module, i.e., the Rep-C2f module, achieving implicit feature reuse and improving detection speed. Finally, the fast spatial pyramid pooling (SPPF) module is used, whose core function is to efficiently fuse multi-scale contextual information during feature extraction. Then, a novel multi-scale convolutional attention (MSCA) module is added, and layer-by-layer convolutional local fusion processing is performed in the neck network. Multi-scale feature extraction is then achieved through a multi-branch structure in the prediction head network, yielding features at different scales. Pointwise convolutional operations are used to better adapt to channel attention, while more cost-effective depthwise convolutional operations are employed to reduce computational costs. Furthermore, variable kernel convolutions with arbitrary sampling shapes and parameters are used to handle multi-task convolutional operations, achieving efficient feature extraction through irregular convolutional operations and improving network performance.

[0086] In one possible example scenario, Figure 12 This is a flowchart illustrating a method for detecting urban management incidents provided in an embodiment of this application. According to... Figure 12The provided diagram divides the urban management incident detection method into three parts. The first part is the data acquisition and preprocessing module, which uses image acquisition devices such as cameras and on-site cameras to capture video and image data of urban management incidents, and then performs image data filtering, data augmentation, and data annotation. The second part is the urban management violation incident detection model, used to build the urban management model. It mainly utilizes the structural reparameterization of the backbone network to improve the C2f module, realize implicit feature reuse, achieve model lightweighting, and improve detection speed. Then, a multi-scale convolutional attention mechanism is added to capture multi-scale feature maps and strengthen the weight ratio of effective features. Finally, variable kernel convolution is used instead of standard convolution to obtain more effective feature information, providing more options for the trade-off between network overhead and performance. This improves the feature extraction operation of the improved YOLOv8s detection model for different urban management incidents, improving detection accuracy. The third part is the urban management incident detection system. Based on the detection results, it promptly reports and backs up problems such as illegal street vending, random occupation of roads, improper placement of garbage bins, and overflowing garbage, facilitating timely handling. It also statistically analyzes the problems in different areas and highlights areas with multiple problems and frequent occurrences, making law enforcement work more targeted.

[0087] This application embodiment performs variable kernel convolution on irregular sampling points, initializes coordinate transformation using an initial sampling coordinate algorithm with reference to the origin coordinates, obtains modified coordinates through position offset, and obtains features at the corresponding positions through interpolation and resampling, thereby obtaining more effective feature information. This makes the feature map data in the backbone network of the improved YOLOv8s detection model more universal and provides more options for the trade-off between network overhead and performance.

[0088] Figure 13 This is a schematic diagram of the structure of a detection device for urban management incidents provided in an embodiment of this application. According to... Figure 13 The provided diagram shows that the structure of the urban management incident detection device 100 specifically includes: The acquisition module 1301 is used to acquire target image data of the target city management event; The explicit multiplexing module 1302 is used to extract features and perform convolution processing on target image data based on the Ghost Bottleneck module in the improved YOLOv8s detection model to obtain explicit multiplexing feature data. Implicit reuse module 1303 is used to perform convolution processing on explicit reuse feature data based on the reparameterization bottleneck module in the improved YOLOv8s detection model to obtain implicit reuse feature data. The reparameterization bottleneck module is obtained by structural reparameterization of the phantom bottleneck module. The monitoring module 1304 is used to perform multi-scale feature fusion on implicit reused feature data based on the multi-scale attention mechanism in the improved YOLOv8s detection model, and predict the classification results of target urban management events based on the fused features.

[0089] The urban management incident detection device 100 provided in this embodiment can be as follows: Figure 13 The urban management event detection device 100 shown can realize Figures 1-12 For a detailed description of the technical effectiveness of the urban management incident detection method shown, please refer to [link / reference]. Figures 1-12 The corresponding explanation is concise and will not be elaborated upon here.

[0090] Figure 14 This is a schematic diagram of the structure of a detection device for urban management incidents provided in an embodiment of this application. The detection device for urban management incidents may include a processor 1401 and a memory 1402 storing computer program instructions.

[0091] Specifically, the processor 1401 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0092] Memory 1402 may include mass storage for data or instructions. For example, and not limitingly, memory 1402 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 1402 may include removable or non-removable (or fixed) media. Where appropriate, memory 1402 may be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, memory 1402 is non-volatile solid-state memory.

[0093] Memory may include read-only memory (ROM), random access memory (RAM), disk storage media devices, optical storage media devices, flash memory devices, and electrical, optical, or other physical / tangible memory storage devices. Therefore, typically, memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the method according to the first aspect of this application.

[0094] The processor 1401 reads and executes computer program instructions stored in the memory 1402 to implement any of the urban management event detection methods in the above embodiments.

[0095] In one example, the urban management incident detection device may further include a communication interface 1403 and a bus 1410. For example, Figure 14 As shown, the processor 1401, memory 1402, and communication interface 1403 are connected through bus 1410 and complete communication with each other.

[0096] The communication interface 1403 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.

[0097] Bus 1410 includes hardware, software, or both, that couples components of an online data traffic metering device together. For example, and not limitingly, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 1410 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, any suitable bus or interconnect is contemplated herein.

[0098] Furthermore, in conjunction with the urban management event detection methods described in the above embodiments, this application embodiment can provide a computer storage medium for implementation. This computer storage medium stores computer program instructions; when these computer program instructions are executed by a processor, they implement any of the urban management event detection methods described in the above embodiments.

[0099] This application also provides a computer program product, including a computer program that, when executed by a processor, implements any of the urban management event detection methods described in the above embodiments.

[0100] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0101] The functional blocks shown in the above-described structural diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0102] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0103] The aspects of this application have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by dedicated hardware performing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0104] The above description is merely a specific implementation of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.

Claims

1. A method for detecting urban management incidents, characterized in that, include: Acquire target image data of the target city management event; The target image data is subjected to feature extraction and convolution processing based on the phantom bottleneck module in the improved YOLOv8s detection model to obtain explicit reuse feature data. The explicit reused feature data is convolved based on the reparameterization bottleneck module in the improved YOLOv8s detection model to obtain implicit reused feature data. The reparameterization bottleneck module is obtained by structural reparameterization of the phantom bottleneck module. The implicit reused feature data is fused using the multi-scale attention mechanism in the improved YOLOv8s detection model, and the classification result of the target urban management event is predicted based on the fused features.

2. The method according to claim 1, characterized in that, After outputting the detection results of the target city management event, the method further includes: Based on the detection results, determine the target issue corresponding to the target urban management event; The event target issues corresponding to multiple urban management events are statistically analyzed to determine the target cities and to mark the event target issues.

3. The method according to claim 1, characterized in that, The acquisition of target image data for the target city management event includes: Collect image and video information of target city management events; The images in the video information are extracted at set time intervals to obtain the corresponding image data; The image data and image information are filtered and image enhancement processes are performed to obtain image enhancement data; The image enhancement data is then processed with data annotation to obtain the target image data.

4. The method according to claim 1, characterized in that, The phantom bottleneck module in the improved YOLOv8s detection model performs feature extraction and convolution processing on the target image data to obtain explicitly reused feature data, including: The target image data is convolved and activated using the first Ghost module stacked in the phantom bottleneck module, and the processed target image data is then subjected to head fusion processing to generate the corresponding first feature map data. The first feature map data is processed by layer-by-layer convolution using the second Ghost module stacked in the phantom bottleneck module, and the processed first feature map data is then fused with a header to generate the second feature map data. The first feature map data and the second feature map data are concatenated to generate explicit reused feature data.

5. The method according to claim 1, characterized in that, The reparameterization bottleneck module in the improved YOLOv8s detection model performs convolution processing on the explicit reuse feature data to obtain implicit reuse feature data, including: Based on the first Rep module in the reparameterization bottleneck module, the explicit reuse feature data is subjected to convolution and batch normalization processing, and the processed explicit reuse feature data is subjected to neck fusion and activation processing to obtain nonlinear feature data. The nonlinear feature data is convolved and batch normalized using the second Rep module in the reparameterization bottleneck module, and the processed nonlinear feature data is then subjected to neck fusion processing to obtain implicit reuse feature data.

6. The method according to claim 1, characterized in that, The process of fusing implicitly reused feature data using the multi-scale attention mechanism in the improved YOLOv8s detection model, and predicting the classification result of the target urban management event based on the fused features, includes: The implicit reused feature data is processed by layer-by-layer convolution using the multi-scale convolution module in the improved YOLOv8s detection model to obtain local feature data. The local feature data is convolved by a target number of deep strip convolution modules to obtain multi-scale feature data. The multi-scale feature data is weighted and fused using an attention module, and the classification result of the target urban management event is predicted based on the fused features.

7. The method according to claim 4, characterized in that, Before obtaining the implicit reuse feature data, the method further includes: Convolution processing is performed on the regular sampling points in the target image data to obtain regular feature map data of the center point position of the regular sampling points; The irregular sampling points in the target image data are offset to obtain modified coordinate information. The modified coordinate information is then interpolated and resampled to obtain feature map data of the corresponding coordinate positions after variable kernel convolution processing.

8. A detection device for urban management incidents, characterized in that, The device includes: The acquisition module is used to acquire target image data of the target city management event; An explicit reuse module is used to perform feature extraction and convolution processing on the target image data based on the phantom bottleneck module in the improved YOLOv8s detection model to obtain explicit reuse feature data. An implicit reuse module is used to perform convolution processing on the explicit reuse feature data based on the reparameterization bottleneck module in the improved YOLOv8s detection model to obtain implicit reuse feature data. The reparameterization bottleneck module is obtained by structural reparameterization of the phantom bottleneck module. The monitoring module is used to perform multi-scale feature fusion on the implicit reused feature data based on the multi-scale attention mechanism in the improved YOLOv8s detection model, and predict the classification result of the target urban management event based on the fused features.

9. A detection device for urban management incidents, characterized in that, The device includes: a processor and a memory storing computer program instructions; When the processor executes the computer program instructions, it implements the method for detecting urban management events as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer program instructions, which, when executed by a processor, implement the method for detecting urban management events as described in any one of claims 1-7.

11. A computer program product, characterized in that, When the instructions in the computer program product are executed by the processor of the electronic device, the electronic device performs the urban management event detection method as described in any one of claims 1-7.

Images