Multi-dimensional data processing system and method for public security management

Abstract

The invention relates to the field of data processing, and particularly discloses a multi-dimensional data processing system and method for public security management, and the method comprises the steps: firstly obtaining a public security area monitoring video collected by a camera and public security management feedback text information uploaded to a cloud platform by a user, and then carrying out the deep learning technology, and feature extraction and correlation analysis are carried out on the public security information and the public security information, and finally a classification result is obtained through a classifier to judge whether a public security management early warning is sent, so that the accuracy and timeliness of public security management are improved, and the modernization and intelligentization of a public security management system are promoted.

Description

Technical Field

[0001] This application relates to the field of data processing, and more specifically, to a multi-dimensional data processing system and method for public security management. Background Technology

[0002] Public security management, through legal and administrative means and cooperation from all sectors of society, aims to maintain social order, prevent crime and other illegal activities, and thus ensure social safety and stability. Its core objectives are to regulate people's behavior, prevent crime, promptly handle emergencies, protect people's lives and property, and ensure the normal operation of society.

[0003] Current security management methods typically rely on surveillance systems and patrol teams. In public places such as residential areas, streets, shopping malls, and parking lots, surveillance systems monitor the situation in real time. If the system detects suspicious behavior, relevant departments will dispatch patrol personnel to conduct on-site inspections. However, this approach may lead to uneven distribution of police resources and potential response delays.

[0004] Therefore, a multi-dimensional data processing system and method for public security management is desired. Summary of the Invention

[0005] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide a multi-dimensional data processing system and method for public security management. It first acquires surveillance video of a public security area captured by cameras and public security management feedback text information uploaded by users to a cloud platform. Then, it utilizes deep learning technology to extract features and perform correlation analysis on both. Finally, it uses a classifier to obtain classification results to determine whether to issue a public security management warning, thereby improving the accuracy and timeliness of public security management and promoting the modernization and intelligence of the public security management system.

[0006] According to one aspect of this application, a multi-dimensional data processing system for public security management is provided, comprising:

[0007] The security management data acquisition module is used to acquire security area surveillance videos collected by cameras and security management feedback text information uploaded by users to the cloud platform.

[0008] The public security management data extraction module is used to extract the public security area monitoring attention feature vector and the public security management feedback text information feature vector from the public security area monitoring video captured by the camera and the public security management feedback text information uploaded by the user to the cloud platform.

[0009] The public security management early warning judgment module is used to determine whether to issue a public security management early warning based on the public security area monitoring attention feature vector and the public security management feedback text information feature vector.

[0010] According to another aspect of this application, a multi-dimensional data processing method for public security management is provided, comprising:

[0011] Acquire surveillance video of the security area collected by cameras and security management feedback text information uploaded by users to the cloud platform;

[0012] Extract the security area monitoring attention feature vector and the security management feedback text information feature vector from the security area surveillance video captured by the camera and the security management feedback text information uploaded by the user to the cloud platform;

[0013] Based on the security area monitoring attention feature vector and the security management feedback text information feature vector, it is determined whether to issue a security management warning.

[0014] Compared with existing technologies, this application provides a multi-dimensional data processing system and method for public security management. It first acquires surveillance video of public security areas collected by cameras and public security management feedback text information uploaded by users to the cloud platform. Then, it uses deep learning technology to extract features and perform correlation analysis on the two. Finally, it uses a classifier to obtain classification results to determine whether to issue a public security management warning, thereby improving the accuracy and timeliness of public security management and promoting the modernization and intelligence of the public security management system. Attached Figure Description

[0015] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0016] Figure 1 This is a block diagram of a multi-dimensional data processing system for public security management according to an embodiment of this application.

[0017] Figure 2 This is a block diagram of the security management data extraction module in a multi-dimensional data processing system for security management according to an embodiment of this application.

[0018] Figure 3 This is a block diagram of a security management early warning judgment module in a multi-dimensional data processing system for security management according to an embodiment of this application.

[0019] Figure 4 This is a flowchart of a multi-dimensional data processing method for public security management according to an embodiment of this application. Detailed Implementation

[0020] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0021] Figure 1 This is a block diagram of a multi-dimensional data processing system for public security management according to an embodiment of this application. Figure 1 As shown, a multi-dimensional data processing system 100 for public security management according to an embodiment of this application includes: a public security management data acquisition module 110, used to acquire public security area surveillance video captured by a camera and public security management feedback text information uploaded by users to a cloud platform; a public security management data extraction module 120, used to extract public security area monitoring attention feature vectors and public security management feedback text information feature vectors from the public security area surveillance video captured by the camera and the public security management feedback text information uploaded by users to the cloud platform; and a public security management early warning judgment module 130, used to determine whether to issue a public security management early warning based on the public security area monitoring attention feature vectors and the public security management feedback text information feature vectors.

[0022] In the aforementioned multi-dimensional data processing system 100 for public security management, the public security management data acquisition module 110 is used to acquire surveillance videos of public security areas collected by cameras and public security management feedback text information uploaded by users to the cloud platform. It should be understood that the surveillance videos collected by cameras can be integrated with the cloud platform through the video surveillance system. Typically, cameras transmit real-time video streams to a local server or directly connect to the cloud platform, transmitting data via internet protocols (such as RTSP, RTMP, etc.). The cloud platform is responsible for storing, managing, and analyzing this video data, and typically has functions such as video stream transcoding, storage compression, and intelligent monitoring analysis. Through real-time feedback from video surveillance, relevant personnel can instantly obtain dynamic information about the public security area, providing effective support for public security management. Secondly, the public security management feedback text information uploaded by users to the cloud platform is usually collected through mobile applications or web platforms. Users can fill out feedback forms on designated public security management platforms or input text through an app. This information can include reports of public security problems, descriptions of abnormal events, or feedback on security risks within the area. After this text information is transmitted to the cloud platform via the network, the system stores, classifies, and processes it. To improve information processing efficiency, the cloud platform can integrate Natural Language Processing (NLP) technology to analyze user feedback text, identify potential problems, and automatically assign them to the appropriate management departments or personnel for processing. Video footage captured by cameras and user-uploaded feedback text information are aggregated, stored, and analyzed through the cloud platform to form an integrated security management system. This system enables real-time monitoring of security dynamics and improves incident response efficiency.

[0023] Specifically, public security management, through legal and administrative means and collaborative efforts with all sectors of society, aims to maintain social order, prevent crime and other illegal activities, and ensure social safety and stability. Its main objectives are to regulate citizen behavior, prevent crime, respond quickly to emergencies, protect the safety of people's lives and property, and ensure the normal operation of society. Current public security management methods typically rely on surveillance systems and patrol forces. In public areas such as residential areas, streets, shopping malls, and parking lots, surveillance equipment monitors around the clock, detecting suspicious activities in real time. When the surveillance system detects abnormal behavior, relevant departments dispatch patrol personnel to the scene for verification. However, this traditional approach may lead to an uneven distribution of police resources, and in some cases, the response speed of patrol teams may be slow, resulting in a certain delay. Therefore, in the technical solution of this application, by acquiring surveillance video of public security areas collected by cameras and public security management feedback text information uploaded by users to the cloud platform, and combining this with deep learning technology to determine whether a public security management warning needs to be issued, the accuracy and response speed of public security management can be improved, thereby promoting the modernization and intelligentization of the public security management system.

[0024] In the aforementioned multi-dimensional data processing system 100 for public security management, the public security management data extraction module 120 is used to extract public security area monitoring attention feature vectors and public security management feedback text information feature vectors from the public security area surveillance video captured by cameras and the public security management feedback text information uploaded by users to the cloud platform. It should be understood that by extracting these feature vectors, the public security management system can more intelligently analyze surveillance videos and user feedback text, accurately identify potential security threats, optimize resource allocation and emergency response, and improve the efficiency and effectiveness of regional public security management.

[0025] Figure 2 This is a block diagram of a security management data extraction module in a multi-dimensional data processing system for security management according to an embodiment of this application. Figure 2 As shown, in a specific embodiment of this application, the public security management data extraction module 120 includes: a public security area surveillance video feature extraction unit 121, used to extract features from the public security area surveillance video collected by the camera to obtain the public security area surveillance attention feature vector; and a public security management feedback text information feature extraction unit 122, used to extract features from the public security management feedback text information uploaded by the user to the cloud platform to obtain the public security management feedback text information feature vector.

[0026] It is understandable that feature extraction from surveillance videos of public security areas captured by cameras aims to transform the raw video data into a high-dimensional vector representation that is easy for machine learning and pattern recognition—namely, the attention feature vector for public security area surveillance. Given that camera-captured video data is typically continuous dynamic images containing a large amount of redundant information and background noise, directly extracting useful information from these videos is difficult and inefficient for manual analysis. Through feature extraction, the system can transform key information in the video (such as abnormal behavior, moving objects, crowd gatherings, etc.) into structured feature vectors, facilitating computer processing and analysis. These feature vectors contribute to intelligent anomaly detection, behavior analysis, and security early warning.

[0027] Furthermore, considering that user-uploaded feedback texts often contain various forms of expression and information, including reports of incidents, anomalies, and suspected security risks, directly analyzing and processing these texts manually is not only time-consuming and labor-intensive but also inefficient. Therefore, feature extraction can convert the text content into a structured vector form, enabling the system to automatically analyze and understand the key information implicit in the text. In addition, feature extraction helps eliminate redundant information, extracting content valuable for public security management, and providing a foundation for further classification, analysis, and response.

[0028] In a specific embodiment of this application, the security area surveillance video feature extraction unit 121 includes: extracting key frames from the security area surveillance video captured by the camera to obtain multiple security area surveillance key frames; and performing feature encoding on the multiple security area surveillance key frames to obtain the security area surveillance attention feature vector.

[0029] It is understandable that surveillance videos are typically composed of a series of continuous frames; however, most of these frames may be static or show little change. Therefore, during analysis, it is necessary to select frames that represent key moments. The purpose of extracting keyframes is to reduce redundant information, save storage space, and improve computational efficiency. Keyframes usually reveal important moments in the video, such as the instant when abnormal behavior occurs, or the critical moments when people enter or leave the monitored area. Therefore, keyframe extraction can effectively reduce the computational burden of the system and ensure more efficient and accurate security incident analysis. In the technical solution of this application, keyframes are extracted from security area surveillance videos captured by cameras. The aim is to select the most representative or informative image frames from the continuous frames of the video. These keyframes can effectively reflect important events or changes in the video. This process not only helps reduce redundant data but also significantly improves the efficiency of subsequent analysis, retrieval, and event recognition. Especially in security management scenarios, keyframe extraction helps the system focus on moments with potential threats or abnormal behavior, while avoiding processing irrelevant static scenes. Keyframes can be identified by comparing the image differences between adjacent frames. When the pixel difference between two adjacent frames in a video exceeds a set threshold, it indicates a significant change in the video, potentially suggesting an important event or activity. These frames are selected as keyframes. This method is simple and intuitive, effectively capturing dynamic changes. Through these methods, the system can extract the most informative keyframes from large amounts of video data, thereby improving the efficiency of security monitoring and supporting timely detection and response to security incidents.

[0030] Furthermore, keyframes in surveillance footage are crucial image frames extracted from video surveillance streams. These frames typically contain key information about security incidents, such as suspicious individuals or unusual behaviors. However, image information in surveillance videos is often highly complex and redundant, and directly analyzing these images can lead to inefficient processing. Therefore, feature encoding is necessary. Feature encoding involves applying deep learning models (such as Convolutional Neural Networks, CNNs) to extract effective features from keyframes, identifying objects, activities, and their spatial locations within the image. These features include not only local details but may also include global scene information. Through encoding, the system can transform each keyframe into a low-dimensional vector representation, effectively compressing the complex information of the image and extracting the most discriminative features.

[0031] In one specific embodiment of this application, feature encoding is performed on the multiple security area monitoring keyframes to obtain the security area monitoring attention feature vector, including: passing the multiple security area monitoring keyframes through an attention-based security area monitoring feature extractor to obtain a security area monitoring attention feature map; and performing max pooling on the security area monitoring attention feature map to obtain the security area monitoring attention feature vector.

[0032] It is understandable that key frames in surveillance videos may contain a large amount of redundant information or background content, which is not meaningful for event analysis. Traditional feature extraction methods may fail to effectively distinguish between important and background areas, resulting in inaccurate information extraction. Attention mechanisms can simulate the selective attention process of human vision. By assigning different weights to different areas, they focus on the areas most likely to contain abnormal behavior or events. In surveillance scenarios, key areas may include human activity, unusual lingering, or sudden events. Attention mechanisms can improve the efficiency of extracting and analyzing this key information. In the technical solution of this application, multiple key frames of security area surveillance are processed by an attention-based feature extractor. The aim is to automatically filter out the most important areas or features from a large number of surveillance images, thereby improving the efficiency of detecting and responding to key events. In security management, surveillance videos may contain a large amount of static or background information, while what truly needs attention is often the dynamic changes or abnormal behavior in specific areas. By introducing attention mechanisms, the system can intelligently focus on key information areas in video frames, thereby enhancing the ability to extract important features and improving the overall performance of the model. This method significantly enhances the intelligent analysis capabilities of surveillance systems in complex environments, enabling security management systems to more accurately identify abnormal events and respond promptly. Specifically, the convolutional coding portion of the attention-based security area surveillance feature extractor performs deep convolutional coding on multiple security area surveillance keyframes to obtain an initial convolutional feature map; the initial convolutional feature map is input into the spatial attention portion of the attention-based security area surveillance feature extractor to obtain a spatial attention map; the spatial attention map is then processed through a Softmax activation function to obtain a spatial attention feature map; and the positional multiplication of the spatial attention feature map and the initial convolutional feature map is calculated to obtain the security area surveillance attention feature map. More specifically, inputting the initial convolutional feature map into the spatial attention part of the security area monitoring feature extractor based on the attention mechanism to obtain a spatial attention map includes: performing average pooling and max pooling along the channel dimension on the initial convolutional feature map to obtain an average feature matrix and a maximum feature matrix; concatenating the average feature matrix and the maximum feature matrix and adjusting the channels to obtain a channel feature matrix; and using the convolutional layer of the spatial attention feature map to perform convolutional encoding on the channel feature matrix to obtain a spatial attention map.

[0033] Furthermore, the purpose of max pooling the attention feature map of the security area surveillance is to simplify the spatial information of the feature map, extract the most representative features, and thus obtain a more compact and discriminative feature vector. This process helps to extract high-level information of key areas from complex surveillance images and significantly reduces computational complexity. In security surveillance, the attention feature map typically shows the importance of different areas in a video frame, some of which (such as densely populated areas or places where abnormal behavior occurs) may require more attention. Max pooling effectively preserves the most salient features of these key areas by selecting the maximum value within a specific region, while removing unimportant details and background noise. The steps of max pooling include dividing the attention feature map into multiple small sub-regions, selecting the maximum value within each sub-region, and then concatenating or summing these maximum values ​​into a new feature vector. This feature vector represents the most important spatial features of the input image and is the most discriminative feature representation extracted from the original image. Through max pooling, the system can retain the most critical information while reducing computational load, which helps improve the accuracy of subsequent tasks such as behavior recognition, anomaly detection, and event classification.

[0034] In a specific embodiment of this application, the security management feedback text information feature extraction unit 122 includes: segmenting the security management feedback text information uploaded by the user to the cloud platform to obtain a security management feedback text information word sequence; and passing the security management feedback text information word sequence through a security management feedback text information convolutional coding model to obtain the security management feedback text information feature vector.

[0035] It is understandable that public security management feedback text information often contains descriptions, events, and contexts related to public security. Word segmentation transforms the text from disordered character forms into a meaningful sequence of words. In this way, the system can better understand the content of the text and extract key information about the event based on the relationships between words. In the technical solution of this application, word segmentation is performed on public security management feedback text information uploaded by users to the cloud platform. The aim is to transform the raw text data into machine-understandable word units, thereby facilitating further analysis and processing. Text data is usually continuous natural language, but computers cannot directly understand its semantics when processing this data. Therefore, it is necessary to segment the text into smaller semantic units, usually words or phrases. In this way, word segmentation can break down continuous characters in public security management feedback text information into independent words to extract keywords and semantic features, aiding subsequent analysis and judgment. Specifically, word segmentation methods are generally divided into two main categories: rule-based methods and statistical methods. Rule-based methods rely on pre-defined dictionaries and rules to segment by matching words in the text. This method is simple and intuitive, suitable for well-structured language, but may not be accurate enough for some new words or technical terms. Statistical methods utilize machine learning techniques to learn probabilistic relationships between words from large amounts of labeled corpora, enabling word segmentation without a dictionary. Modern deep learning methods, such as Long Short-Term Memory (LSTM) networks and Transformer models, can also be used for word segmentation, and can segment text more accurately based on contextual information. Once word segmentation is complete, the resulting sequence of words in the security management feedback text can be used as input for subsequent Natural Language Processing (NLP) tasks, such as sentiment analysis, keyword extraction, and topic modeling. These steps provide a foundation for further security management early warning and decision-making, helping the system to more accurately understand user feedback and respond promptly. In summary, word segmentation can transform complex natural language information into a machine-processable format, providing accurate and timely feedback information support for security management systems.

[0036] Furthermore, the word sequences of public security management feedback text are processed using a convolutional coding model to extract effective semantic information from the original text and convert it into a low-dimensional feature vector, facilitating subsequent analysis and decision-making. Public security management feedback text typically contains a large amount of key information, such as event descriptions, time, location, and relevant personnel. However, this information usually exists in natural language, with a complex structure that is difficult to process directly by computers. Therefore, encoding the text using a convolutional coding model can effectively extract the latent semantic features of the text, improving the model's ability to understand the text content. Convolutional Neural Networks (CNNs), when processing text, can utilize their local receptive field characteristics to capture important local features from word sequences in the text. For example, through convolutional operations, the model can identify key phrases or terms related to public security management (such as "abnormal behavior," "emergency events," etc.), which typically carry important semantic information. Traditional text processing methods such as the bag-of-words model or TF-IDF methods cannot fully capture the contextual relationships in the text, while convolutional coding models can capture local dependencies between words through a sliding window, thereby improving the representational ability of text features. Specifically, firstly, the security management feedback text information is converted into word vector representations, with common methods including Word2Vec, GloVe, or BERT. Each word is mapped to a high-dimensional vector representing its semantic features. Next, these word vector sequences are processed using a convolutional neural network. The convolution operation performs local convolution on the word sequence through a sliding window, extracting features from each local region (e.g., features of certain keyword combinations). The output of the convolutional layer is typically multiple feature maps, each representing different local semantic information. Then, pooling layers (such as max pooling or average pooling) are used to extract the important information from each feature map, resulting in a fixed-length feature vector. The pooling layer reduces the dimensionality of the feature map, retaining the most salient features and removing redundant information. Finally, through these processes, the resulting feature vector can fully represent the core information of the original security management feedback text. In this way, the feature vector extracted by the convolutional coding model is not only more compact in dimensionality but also more focused on the key semantics of the text, improving the processing efficiency and accuracy of the security management system. Specifically, the embedding layer of the convolutional coding model for security management feedback text information is used to map each security management feedback word in the sequence of security management feedback text information words into a word embedding vector to obtain a sequence of security management feedback word embedding vectors; the BERT model based on the converter of the convolutional coding model for security management feedback text information is used to perform global contextual semantic encoding on the sequence of security management feedback word embedding vectors to obtain multiple security management feedback feature vectors; and the multiple security management feedback feature vectors are concatenated to obtain the security management feedback text information feature vector.

[0037] In the aforementioned multi-dimensional data processing system 100 for public security management, the public security management early warning judgment module 130 is used to determine whether to issue a public security management early warning based on the public security area monitoring attention feature vector and the public security management feedback text information feature vector. It should be understood that determining whether to issue a public security management early warning based on the public security area monitoring attention feature vector and the public security management feedback text information feature vector aims to combine visual data and text data to comprehensively analyze the public security situation in a multimodal manner, thereby determining whether there are potential security threats or abnormal situations. This judgment process not only relies on surveillance images in video but also identifies potential risk events by analyzing information in management feedback text, providing decision-makers with accurate early warning information.

[0038] Figure 3 This is a block diagram of a security management early warning judgment module in a multi-dimensional data processing system for security management according to an embodiment of this application. Figure 3 As shown, in a specific embodiment of this application, the security management early warning judgment module 130 includes: a security management information feature fusion unit 131, used to fuse the security area monitoring attention feature vector and the security management feedback text information feature vector to obtain a security management early warning judgment feature vector; a security management information feature optimization unit 132, used to perform single estimation bias correction based on the target parameter space on the security management early warning judgment feature vector to obtain an optimized security management early warning judgment feature vector; and a security management early warning classification judgment unit 133, used to pass the optimized security management early warning judgment feature vector through a classifier to obtain a classification result, the classification result being used to determine whether to issue a security management early warning.

[0039] It is understandable that the purpose of fusing the attention feature vector of security area monitoring and the feature vector of security management feedback text information is to comprehensively utilize multimodal information and improve the accuracy and comprehensiveness of the security management early warning system. The attention feature vector of security area monitoring mainly extracts visual features that may indicate abnormal behavior, such as crowd gatherings or emergencies, by analyzing key areas in video surveillance. The feature vector of security management feedback text information reflects textual information from on-site management personnel, witnesses, or other channels. This text typically contains key content such as event descriptions, personnel activities, and abnormal behavior. By fusing these two feature vectors, the system can combine visual and semantic data to comprehensively determine whether there are potential security threats, thereby issuing more accurate warnings. The methods for fusing these two feature vectors typically include feature concatenation, weighted summation, or adaptively adjusting the importance of different modalities through an attention mechanism. First, features are extracted from the monitoring video and feedback text separately, resulting in two high-dimensional feature vectors. Then, the two feature vectors are merged into a more information-dense vector using a concatenation method, or different weights are assigned according to the importance of the modality using a weighted summation method. The fused feature vector is then input into the decision model for early warning judgment. Furthermore, the attention mechanism can dynamically adjust the weights of monitoring and textual information based on different contexts, allowing the system to prioritize inputs of a particular modality in specific situations. For example, when there are obvious anomalies in video surveillance footage, the system may rely more heavily on video features. Ultimately, the fused security management early warning judgment feature vector can provide more comprehensive contextual analysis, enabling the early warning system to more accurately determine whether to issue an early warning signal based on integrated information, helping relevant personnel to respond to potential risks in a timely manner.

[0040] Specifically, considering that video data captured by cameras, after keyframe extraction and attention mechanisms, yields visual features, these features focus on objects, scenes, and dynamic changes in the image, typically reflecting environmental and event characteristics within a security area, such as unusual activity and human behavior. User-uploaded security management feedback text information, however, is expressed through language, containing descriptions of events related to the security situation, emotional tendencies, or feedback. Text data reflects understanding, evaluation, or suggestions regarding security issues, exhibiting strong subjectivity and semantic complexity. When fusing this information from different modalities, despite processing visual and text data using different feature extraction methods, their respective feature vectors still differ significantly. Visual features are typically high-dimensional image data, mainly containing low- and mid-level features such as pixels, textures, objects, and region relationships, while text features contain grammatical, semantic, and contextual information, reflecting the complex logic behind language. This heterogeneity makes it difficult for visual and text features to directly complement or reinforce each other during fusion, and it's challenging to simultaneously capture the main characteristics of both. Especially after extracting visual and textual features separately through attention mechanisms and convolutional coding models, the inherent structural differences between the two mean that the fused feature vector may not fully retain the unique information of each data type. Furthermore, information in video data is temporal and closely related to environmental changes, while textual data is more static and language-based. While simple fusion of the two can yield some comprehensive features, the lack of an effective cross-modal information mapping mechanism means that the fused security management early warning feature vector often fails to strengthen the intrinsic connections between different data points. This weak cross-modal information connectivity makes it difficult for the model to accurately capture and utilize the potential correlation between video and text, thus affecting the accuracy of the final early warning judgment. Therefore, in the technical solution of this application, an optimized security management early warning feature vector is obtained by performing a single estimation bias correction based on the target parameter space on the security management early warning feature vector.

[0041] The process of correcting the single estimation bias of the security management early warning judgment feature vector based on the target parameter space to obtain an optimized security management early warning judgment feature vector includes: extracting the target parameter matrix of the classifier; performing node-based decomposition of the target parameter matrix in units of row vectors to obtain a set of target parameter node encoding vectors; using each target parameter node encoding vector in the set of target parameter node encoding vectors as a walk topology space, and applying topological space constraints to the security management early warning judgment feature vector to obtain a set of constrained security management early warning judgment feature vectors; and calculating the positional mean vector of the set of constrained security management early warning judgment feature vectors to obtain the optimized security management early warning judgment feature vector.

[0042] Specifically, using each target parameter node encoding vector in the set of target parameter node encoding vectors as the walk topology space, topological space constraints are applied to the security management early warning judgment feature vector to obtain a set of constrained security management early warning judgment feature vectors. This includes: multiplying the security management early warning judgment feature vector and the transpose of the target parameter node encoding vector, and calculating the natural exponential function value of the multiplication result to obtain the weighted exponential response weight; calculating the Euclidean distance between the security management early warning judgment feature vector and the target parameter node encoding vector to obtain the node encoding distance value; performing a dot product between the node encoding distance value and the security management early warning judgment feature vector, and calculating the natural exponential function value for each feature value of the dot product vector to obtain the security management distance guided exponential feature vector; and multiplying the weighted exponential response weight and the security management distance guided exponential feature vector to obtain the constrained security management early warning judgment feature vector.

[0043] The optimized security management early warning judgment feature vector is obtained by performing a single estimation bias correction based on the target parameter space on the security management early warning judgment feature vector, which is expressed by the following optimization formula:

[0044]

[0045]

[0046]

[0047] in, Represents the objective parameter matrix. The first, second, and third nodes of the set of target parameter node encoding vectors are represented. The, the Each target parameter node encoding vector Represents the transpose of a vector. This represents the feature vector for early warning judgment in public security management. Represents matrix multiplication. This indicates dot product by position. Represents the computation of vectors sum vector The Euclidean distance between them The set of constrained security management early warning judgment feature vectors is represented by the first... A constrained feature vector for public security management early warning judgment. This represents the total number of feature vectors in the set of constrained public security management early warning judgment. This represents the optimized feature vector for public security management early warning judgment;

[0048] In the technical solution of this application, a single estimation bias correction is performed on the feature vector for public security management early warning judgment based on the target parameter space. This process first extracts key parameters for decision-making from a pre-trained classifier. These parameters form a matrix in a high-dimensional space, where each row represents a weight or influencing factor in a different dimension. The target parameter matrix reveals the location and shape of the model's decision boundary, thereby inferring which input features are most critical to the prediction result.

[0049] Next, the target parameter matrix is ​​decomposed into node-based vectors to obtain a set of node-encoded vectors for the target parameters. Here, each row vector is treated as a node in graph theory, meaning that each set of parameters is now considered an entity with potential connectivity. This transformation allows the application of methods from graph theory and network science to explore the interactions between features. The node-encoded vectors not only carry information about the original parameters but also implicitly contain knowledge about the overall system topology. Node-based decomposition further reveals the inherent connection patterns or structures of the data, enabling the optimized feature vectors to better adapt to new task requirements.

[0050] Then, using each target parameter node encoding vector in the set of target parameter node encoding vectors as the walking topology space, topology space constraints are applied to the security management early warning judgment feature vectors to obtain the set of constrained security management early warning judgment feature vectors. "Walking" using the topology space defined by the node encoding vectors actually simulates an exploratory process, aiming to find feature transformations that best preserve the characteristics of the original data structure. Each step determines the next position based on the probability distribution of the current state. The topology space constraints ensure that the feature representation retains certain invariants even in different contexts. Simultaneously, it can promote cross-domain transfer learning because it emphasizes the general relationships between features rather than domain-specific details. In this way, the feature space is reconstructed, making the optimized feature vectors more compact and possessing better generalization ability.

[0051] Finally, the location-based mean vector of the set of constrained security management early warning judgment feature vectors is calculated to obtain the optimized security management early warning judgment feature vector. Calculating the mean vector is a statistical aggregation method used to synthesize the optimal solution from multiple perspectives. The idea behind this step is to reduce the bias caused by a single estimate by fusing information provided by different sample points. The averaging process is equivalent to performing a soft vote, enhancing the expressiveness of common features and making the optimized feature vector more stable and reliable.

[0052] Furthermore, the optimized security management early warning judgment feature vector is obtained by fusing different types of data (such as surveillance footage and text information). It contains potential anomaly or threat information, providing rich input features for the classifier. By using the classifier, the system can automatically determine whether the current security situation meets the criteria for issuing an early warning based on these feature vectors. The core function of the classifier is to automatically distinguish different categories of input data based on known patterns in the training data. In a security management system, the classification task for early warnings is usually a binary classification problem, i.e., determining whether the current situation belongs to a "normal" or "abnormal" state. The classifier can transform the optimized security management early warning judgment feature vector, after feature extraction and fusion, into a specific classification result (e.g., outputting "normal" or "abnormal"). Through this judgment, the system can decide whether to issue an early warning signal and take timely countermeasures. Specifically, in a specific embodiment of this application, firstly, the optimized security management early warning judgment feature vector is used as the input to the classifier. This feature vector typically contains key semantic information extracted from surveillance videos and management feedback text, and has been transformed into a high-dimensional vector representation. Next, a suitable classifier is selected for prediction. Common classifiers include Support Vector Machines (SVM), Random Forests, Decision Trees, or Deep Neural Networks (DNNs). These classifiers learn the feature distributions of different categories (such as "normal" and "abnormal") through training on a large amount of historical data, and can determine which category the input data belongs to based on the new feature vectors. Specifically, the classifier compares the input feature vector with the decision boundary learned during training, and outputs a category label according to preset rules (e.g., threshold judgment, distance measurement, etc.). If the classifier outputs a label of "abnormal" or "high risk," it indicates that there may be potential security risks in the current situation, and the system will issue a security management warning; if the output label is "normal," it indicates that the current situation is within the safe range, and no warning is needed. Finally, through the classifier's determination, the system can automatically make a decision on whether to issue a warning based on real-time monitoring and feedback information. This automated warning mechanism can greatly improve response speed, reduce the bias of human judgment, ensure timely handling of potential security problems, and effectively improve the efficiency and accuracy of public safety management.

[0053] In summary, this application embodiment first acquires security area surveillance video collected by cameras and security management feedback text information uploaded by users to the cloud platform. Then, it uses deep learning technology to extract features and perform correlation analysis on the two. Finally, it uses a classifier to obtain classification results to determine whether to issue a security management warning, thereby improving the accuracy and timeliness of security management and promoting the modernization and intelligence of the security management system.

[0054] As described above, the multi-dimensional data processing system 100 for public security management according to the embodiments of this application can be implemented in various terminal devices. In one example, the multi-dimensional data processing system 100 for public security management can be integrated into the terminal device as a software module and / or a hardware module. For example, the multi-dimensional data processing system 100 for public security management can be a software module in the operating system of the terminal device, or it can be an application developed for the terminal device; of course, the multi-dimensional data processing system 100 for public security management can also be one of many hardware modules of the terminal device.

[0055] Alternatively, in another example, the multi-dimensional data processing system 100 for security management and the terminal device can also be separate devices, and the multi-dimensional data processing system 100 for security management can be connected to the terminal device via wired and / or wireless networks, and transmit interactive information in accordance with an agreed data format.

[0056] Figure 4 This is a flowchart of a multi-dimensional data processing method for public security management according to an embodiment of this application. Figure 4 As shown, the multi-dimensional data processing method for public security management according to an embodiment of this application includes: S110, acquiring security area surveillance video captured by a camera and security management feedback text information uploaded by a user to a cloud platform; S120, extracting a security area surveillance attention feature vector and a security management feedback text information feature vector from the security area surveillance video captured by the camera and the security management feedback text information uploaded by the user to the cloud platform; S130, determining whether to issue a public security management warning based on the security area surveillance attention feature vector and the security management feedback text information feature vector.

[0057] Here, those skilled in the art will understand that the specific operations of each step in the above-described multi-dimensional data processing method for public security management have been referenced above. Figures 1 to 3 The description of the multi-dimensional data processing system used for public security management is detailed here, and therefore, its repeated description will be omitted.

Claims

1. A multi-dimensional data processing system for public security management, characterized in that, include: The security management data acquisition module is used to acquire security area surveillance videos collected by cameras and security management feedback text information uploaded by users to the cloud platform. The public security management data extraction module is used to extract the public security area monitoring attention feature vector and the public security management feedback text information feature vector from the public security area monitoring video captured by the camera and the public security management feedback text information uploaded by the user to the cloud platform. The public security management early warning judgment module is used to determine whether to issue a public security management early warning based on the public security area monitoring attention feature vector and the public security management feedback text information feature vector.

2. The multi-dimensional data processing system for public security management according to claim 1, characterized in that, The security management data extraction module includes: The security area surveillance video feature extraction unit is used to extract features from the security area surveillance video captured by the camera to obtain the security area surveillance attention feature vector; The security management feedback text information feature extraction unit is used to extract features from the security management feedback text information uploaded by the user to the cloud platform to obtain the security management feedback text information feature vector.

3. The multi-dimensional data processing system for public security management according to claim 2, characterized in that, The security area surveillance video feature extraction unit includes: Keyframes are extracted from the security area surveillance video captured by the camera to obtain multiple security area surveillance keyframes. Feature encoding is performed on the multiple security area monitoring keyframes to obtain the security area monitoring attention feature vector.

4. The multi-dimensional data processing system for public security management according to claim 3, characterized in that, Feature encoding is performed on the multiple security area surveillance keyframes to obtain the security area surveillance attention feature vector, including: The multiple security area monitoring keyframes are processed by an attention-based security area monitoring feature extractor to obtain a security area monitoring attention feature map. The security area monitoring attention feature map is subjected to max pooling to obtain the security area monitoring attention feature vector.

5. The multi-dimensional data processing system for public security management according to claim 4, characterized in that, The security management feedback text information feature extraction unit includes: The security management feedback text information uploaded by the user to the cloud platform is segmented into words to obtain the security management feedback text information word sequence; The sequence of words in the security management feedback text is passed through a security management feedback text convolutional coding model to obtain the feature vector of the security management feedback text.

6. The multi-dimensional data processing system for public security management according to claim 5, characterized in that, The security management early warning judgment module includes: The public security management information feature fusion unit is used to fuse the public security area monitoring attention feature vector and the public security management feedback text information feature vector to obtain the public security management early warning judgment feature vector. The public security management information feature optimization unit is used to perform single estimation deviation correction based on the target parameter space on the public security management early warning judgment feature vector to obtain an optimized public security management early warning judgment feature vector. The public security management early warning classification and judgment unit is used to pass the optimized public security management early warning judgment feature vector through a classifier to obtain a classification result, and the classification result is used to determine whether to issue a public security management early warning.

7. The multi-dimensional data processing system for public security management according to claim 6, characterized in that, The security management information feature optimization unit includes: Extract the target parameter matrix of the classifier; The target parameter matrix is ​​decomposed into a node-based structure using row vectors to obtain a set of target parameter node encoding vectors; Using each target parameter node encoding vector in the set of target parameter node encoding vectors as the walking topology space, topology space constraints are applied to the security management early warning judgment feature vectors to obtain a set of constrained security management early warning judgment feature vectors; and The optimized security management early warning judgment feature vector is obtained by calculating the positional mean vector of the set of constrained security management early warning judgment feature vectors.

8. The multi-dimensional data processing system for public security management according to claim 7, characterized in that, Using each target parameter node encoding vector in the set of target parameter node encoding vectors as the walking topology space, topology space constraints are applied to the security management early warning judgment feature vectors to obtain a set of constrained security management early warning judgment feature vectors, including: After multiplying the security management early warning judgment feature vector and the transpose of the target parameter node encoding vector, the natural exponential function value of the multiplication result is calculated to obtain the weighted exponential response weight. Calculate the Euclidean distance between the security management early warning judgment feature vector and the target parameter node encoding vector to obtain the node encoding distance value; The node encoding distance value and the security management early warning judgment feature vector are multiplied by a dot, and the natural exponential function value is calculated for each feature value of the vector after the dot product to obtain the security management distance guided exponential feature vector; The weighted index response weight and the security management distance guiding index feature vector are multiplied together to obtain the constrained security management early warning judgment feature vector.

9. A multi-dimensional data processing method for public security management, characterized in that, include: Acquire surveillance video of the security area collected by cameras and security management feedback text information uploaded by users to the cloud platform; Extract the security area monitoring attention feature vector and the security management feedback text information feature vector from the security area surveillance video captured by the camera and the security management feedback text information uploaded by the user to the cloud platform; Based on the security area monitoring attention feature vector and the security management feedback text information feature vector, it is determined whether to issue a security management warning.

10. The multi-dimensional data processing method for public security management according to claim 9, characterized in that, The security area monitoring attention feature vector and the security management feedback text information feature vector are extracted from the security area surveillance video captured by the camera and the security management feedback text information uploaded by the user to the cloud platform, including: Feature extraction is performed on the security area surveillance video captured by the camera to obtain the security area surveillance attention feature vector; Feature extraction is performed on the security management feedback text information uploaded by the user to the cloud platform to obtain the feature vector of the security management feedback text information.

Images