Data processing method and related equipment based on edge AI computing and storage integrated chip

Abstract

The present invention relates to a data processing method based on an edge AI computing and storage integrated chip, comprising the following steps: inputting image data to be processed into a preset edge AI computing and storage integrated chip for data classification to obtain sub-image data to be processed; identifying the sub-image data to be processed to determine whether there is an abnormal situation in the sub-image data to be processed, and if so, marking the sub-image data to be processed based on the abnormal situation to obtain a marked abnormal situation image; performing image compression processing on the marked abnormal situation image to obtain abnormal image compressed data; performing abnormal level classification on the abnormal image compressed data to obtain an abnormal classification result; prioritizing the abnormal situation based on the abnormal classification result to obtain an abnormal event priority sequence table, which solves the technical problem that the computing power and storage space of traditional chips are relatively limited, which limits the processing capacity of complex data sets.

Description

Technical Field

[0001] The present invention relates to the field of edge AI computing and storage chip technology, and in particular to a data processing method and related equipment based on an edge AI computing and storage integrated chip. Background Art

[0002] In today's rapidly developing intelligent era, the widespread use of Internet of Things (IoT) devices and the continuous advancement of artificial intelligence (AI) technology have given rise to the need for efficient data processing methods. Although the traditional cloud computing model provides powerful computing power for large-scale data processing, it has exposed a series of shortcomings when faced with massive real-time data generated by edge devices. First of all, since cloud processing requires a large amount of raw data to be uploaded to a remote server for analysis, this not only increases the pressure on network bandwidth, but also brings significant latency problems. For some application scenarios that require instant response, such as autonomous driving and smart security monitoring, this delay is unacceptable. In addition, frequent data transmission has also caused users to worry about privacy and security, especially when sensitive information is involved.

[0003] In response to the above challenges, a data processing method based on edge AI computing and storage integrated chips has emerged. This method aims to achieve localized data processing by integrating advanced AI algorithms and high-performance computing resources in the hardware on the edge side, thereby effectively alleviating the pressure brought by cloud services. However, in actual application, existing edge computing solutions still have some problems that need to be solved. On the one hand, although edge devices can reduce the amount of data transmission and reduce response time, their computing power and storage space are relatively limited, which limits the deployment of complex models and the processing capabilities of large data sets. On the other hand, how to ensure the security and accuracy of edge-side data processing, especially in the identification and labeling of abnormal situations, has become a key research topic, because any misjudgment or missed detection may lead to serious consequences.

[0004] Therefore, this study proposes an innovative data processing method based on an edge AI computing and storage integrated chip, focusing on improving the efficiency of image data processing while ensuring data security and privacy protection. This method not only realizes local classification and anomaly detection of image data, but also further optimizes the management and feedback mechanism of abnormal events by introducing lossless compression technology and priority sorting models. Specifically, by accurately labeling and compressing abnormal situations, the data volume can be effectively reduced, which is convenient for subsequent efficient transmission; and the abnormal level division and priority sorting ensure that the alarm signal can be promptly conveyed to the relevant responsible personnel according to the degree of urgency, thereby improving the speed and accuracy of emergency response. This series of improvement measures not only meets the needs of application scenarios with high real-time requirements, but also provides new ideas and technical support for building a more intelligent and reliable edge computing system. Summary of the invention

[0005] The main purpose of the present invention is to provide a data processing method and related equipment based on an edge AI computing and storage integrated chip, which solves the technical problem that traditional chips have relatively limited computing power and storage space, limiting the processing capabilities of complex data sets.

[0006] To achieve the above objectives, the present invention provides a data processing method based on an edge AI computing and storage integrated chip, comprising the following steps: The image data to be processed is input into the preset edge AI computing and storage integrated chip for data classification to obtain the sub-image data to be processed; The sub-image data to be processed is identified by a preset identification algorithm to determine whether there is an abnormality in the sub-image data to be processed; if so, the sub-image data to be processed is marked based on the abnormality to obtain an image marked with abnormality; wherein the abnormality has at least one and is different from each other; Performing image compression processing on the annotated abnormal situation image by using a preset lossless compression algorithm to obtain abnormal image compressed data; The abnormal image compressed data is sent to a remote control device, and the abnormal image compressed data is classified into abnormal levels by the remote control device to obtain an abnormal classification result; A preset priority sorting model is adopted to prioritize the abnormal situations based on the abnormal classification results to obtain an abnormal event priority sequence list, and an alarm signal is automatically generated based on the abnormal event priority sequence list. The alarm signal is distributed to the mobile devices or workstations of relevant responsible personnel through a preset communication protocol for early warning prompts.

[0007] Furthermore, the step of inputting the image data to be processed into a preset edge AI computing and storage integrated chip for data classification to obtain sub-image data to be processed includes: Inputting the image data to be processed into a preset edge AI computing and storage integrated chip to obtain input data, and performing data preprocessing on the input data to obtain preprocessed data; Performing spectral analysis on the preprocessed data to obtain spectral feature data, performing boundary detection on the spectral feature data using an adaptive threshold segmentation algorithm to obtain boundary features, and performing topological analysis on the boundary features to obtain topological feature data; Performing dimensionality reduction processing on topological feature data through kernel principal component analysis to obtain dimensionality reduction features, and performing manifold learning on the dimensionality reduction features to obtain manifold feature data; A deep belief network is used to classify the manifold feature data to obtain sub-image data to be processed.

[0008] Further, the sub-image data to be processed is identified by a preset identification algorithm to determine whether there is an abnormality in the sub-image data to be processed, and if so, the sub-image data to be processed is marked based on the abnormality to obtain an image marked with abnormality, including: Using a preset convolutional neural network to extract features from the sub-image data to be processed, to obtain a sub-image feature vector set; wherein each feature vector in the sub-image feature vector set corresponds to a key area in the sub-image data to be processed; Inputting the sub-image feature vector set into a preset recognition algorithm for recognition to determine whether there is an abnormality in the sub-image data to be processed; If so, classifying and labeling the sub-image feature vector set based on the abnormal situation to distinguish between a normal sub-image feature vector set and an abnormal sub-image feature vector set; Using a preset target detection algorithm, the sub-image data to be processed is region-located based on the abnormal sub-image feature vector set to obtain an abnormal region frame; Based on the abnormal region frame, the sub-image data to be processed is classified at the pixel level to obtain a binary segmentation map of the abnormal region and the normal region; wherein the abnormal region is represented by a specific color or grayscale value in the sub-image data to be processed; Performing morphological processing on the binary segmentation image to obtain an optimized abnormal region boundary; Through a preset text generation network, descriptive annotation text is automatically generated for the optimized abnormal region boundary, and a text description is given to the abnormal region to obtain an annotated abnormal situation image.

[0009] Furthermore, the image compression processing is performed on the annotated abnormal situation image by using a preset lossless compression algorithm to obtain abnormal image compressed data, including: Using an image edge detection algorithm to extract edges of the annotated abnormal situation image to obtain edge image data, and performing pixel value statistics on the edge image data to obtain a pixel value frequency distribution table; Performing Huffman tree construction on the pixel value frequency distribution table through a Huffman tree construction algorithm to obtain a Huffman coding tree, and performing coding table generation on the Huffman coding tree to obtain a Huffman coding table; Performing Huffman coding on the annotated abnormal situation image based on the Huffman coding table to obtain a Huffman coding data stream; Performing run-length encoding on the Huffman-encoded data stream to obtain RLE-encoded data; Segmenting the RLE encoded data into data blocks using a preset lossless compression algorithm to obtain segmented RLE encoded data; The segmented RLE encoded data is compressed using a preset lossless compression algorithm to obtain abnormal image compressed data.

[0010] Furthermore, the abnormal level classification of the abnormal image compression data by the remote control device to obtain the abnormal classification result includes: Extracting features from the abnormal image compression data using an image feature extraction algorithm in the remote control device to obtain an abnormal feature data set, and sorting the abnormal feature data set to obtain an ordered feature sequence; Recalling a historical ordered feature sequence from a database, performing quartile calculation on the historical ordered feature sequence through a quartile calculator to obtain a historical quartile distribution, and performing outlier detection on the historical quartile distribution to obtain a historical outlier range; Binning the abnormal feature data set based on the historical abnormal value range to obtain a feature binning result, and performing boundary optimization on the feature binning result to obtain an optimized binning boundary; A hierarchical clustering algorithm is used to perform cluster analysis on the optimized bin boundaries to obtain abnormal level clustering results, and density estimation is performed on the abnormal level clustering results to obtain level density distribution; The grade density distribution is scored and calculated by a multi-index comprehensive scorer to obtain an abnormal grade score, and the abnormal grade score is graded and mapped to obtain an abnormal grade label; The abnormal image compression data is graded based on the abnormal grade label to obtain the graded abnormal data, and the graded abnormal data is encapsulated to obtain the abnormal classification result.

[0011] Furthermore, the preset priority sorting model is used to prioritize the abnormal situations based on the abnormal classification results to obtain an abnormal event priority sequence table, including: An abnormality level analyzer is used to numerically quantize the abnormality classification result to obtain an abnormality level quantization vector; Performing timeliness calculation on the abnormality level quantization vector through a time decay function to obtain a timeliness abnormality feature, and performing an influence range analysis on the timeliness abnormality feature to obtain an influence range vector; Assigning weights to the influence range vector based on the hierarchical analysis method to obtain a weighted influence range vector; Using a preset priority sorting model, the abnormal situations are prioritized based on the weight impact range vector to obtain an initial priority sequence, and conflict detection and elimination are performed on the initial priority sequence to obtain an initial priority sequence after elimination; The initial priority sequence after elimination is subjected to constraint optimization through preset priority constraint conditions to obtain an optimized priority sequence; The optimized priority sequence is formatted to obtain an abnormal event priority sequence table.

[0012] Furthermore, the automatic generation of an alarm signal based on the abnormal event priority sequence table and the distribution of the alarm signal to the mobile device or workstation of the relevant responsible personnel for early warning through a preset communication protocol include: Performing semantic analysis on the abnormal event priority sequence table through a natural language generator to obtain an event semantic vector, and performing context association analysis on the event semantic vector to obtain associated semantic data; wherein the associated semantic data includes event description, impact degree and disposal suggestion; Performing information integration on the associated semantic data to obtain rich media alarm content, and automatically generating an alarm signal based on the rich media alarm content; Performing distributed transaction processing on the alarm distribution task sequence to obtain an alarm distribution task sequence, and performing priority scheduling analysis on the alarm distribution task sequence based on the abnormal event priority sequence table to obtain a task scheduling plan; Dynamically allocating resources to the alarm distribution task sequence based on the task scheduling scheme to obtain a resource allocation strategy table, and optimizing the network topology of the alarm distribution task sequence based on the resource allocation strategy table to obtain an optimal transmission path; Through a preset communication protocol, the alarm distribution task sequence is distributed to the mobile devices or workstations of relevant responsible personnel based on the optimal transmission path for early warning prompts.

[0013] The present invention also provides a data processing device based on an edge AI computing and storage integrated chip, comprising: A classification module is used to input the image data to be processed into a preset edge AI computing and storage integrated chip for data classification to obtain sub-image data to be processed; an identification module, used to identify the sub-image data to be processed by a preset identification algorithm to determine whether there is an abnormality in the sub-image data to be processed, and if so, to mark the sub-image data to be processed based on the abnormality to obtain an image marked with abnormality; wherein the abnormality has at least one and is different from each other; A compression module, used for performing image compression processing on the annotated abnormal situation image by using a preset lossless compression algorithm to obtain abnormal image compressed data; A classification module, used for sending the abnormal image compressed data to a remote control device, and performing abnormal level classification on the abnormal image compressed data through the remote control device to obtain an abnormal classification result; A sorting module is used to adopt a preset priority sorting model to prioritize the abnormal situations based on the abnormal classification results, obtain an abnormal event priority sequence table, and automatically generate an alarm signal based on the abnormal event priority sequence table, and distribute the alarm signal to the mobile devices or workstations of relevant responsible personnel through a preset communication protocol for early warning prompts.

[0014] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0015] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned methods are implemented.

[0016] The data processing method based on the edge AI computing and storage integrated chip provided by the present invention comprises the following steps: inputting the image data to be processed into a preset edge AI computing and storage integrated chip for data classification to obtain sub-image data to be processed; identifying the sub-image data to be processed by a preset recognition algorithm to determine whether there is an abnormality in the sub-image data to be processed, and if so, marking the sub-image data to be processed based on the abnormality to obtain a marked abnormality image; wherein the abnormality has at least one and is different from each other; performing image compression processing on the marked abnormality image by a preset lossless compression algorithm to obtain abnormal image compressed data; The abnormal image compressed data is sent to the remote control device, and the abnormal image compressed data is classified into abnormal levels by the remote control device to obtain the abnormal classification result; the preset priority sorting model is used to prioritize the abnormal situation based on the abnormal classification result to obtain the abnormal event priority sequence table, and the alarm signal is automatically generated based on the abnormal event priority sequence table, and the alarm signal is distributed to the mobile device or workstation of the relevant responsible personnel through the preset communication protocol for early warning prompts; through the above technical means, the technical problem that the traditional chip computing power and storage space are relatively limited, which limits the processing ability of complex data sets, is solved. By identifying and marking different types of abnormal situations, the system can adapt to more complex usage scenarios, and even in the case of poor network conditions or limited resources, the normal operation of basic functions can be guaranteed. In addition, the preset communication protocol ensures that the alarm signal can be accurately transmitted to the designated recipient, enhancing the technical effect of the robustness of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a schematic diagram of the steps of a data processing method based on an edge AI computing and storage integrated chip in one embodiment of the present invention; Figure 2 This is a structural block diagram of a data processing device based on an edge AI computing and storage integrated chip in one embodiment of the present invention; Figure 3 It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0018] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0020] like Figure 1 As shown, Figure 1 This is a schematic diagram of the steps of a data processing method based on an edge AI computing and storage integrated chip in one embodiment of the present invention; In one embodiment of the present invention, a data processing method based on an edge AI computing and storage integrated chip is provided, comprising the following steps: Step S1: input the image data to be processed into a preset edge AI computing and storage integrated chip for data classification to obtain sub-image data to be processed.

[0021] Specifically, when implementing the step of "inputting the image data to be processed into the preset edge AI computing and storage integrated chip for data classification to obtain the sub-image data to be processed", it is first necessary to understand that the "image data to be processed" here refers to the original image information collected from various front-end devices such as sensors or cameras. These image data usually contain a lot of redundant or irrelevant background information, and our goal is to filter out the data fragments that really need to be paid attention to, namely the "sub-image data to be processed". To achieve this goal, we use a specially designed "edge AI computing and storage integrated chip". This chip integrates high-performance computing units and large-capacity storage resources, and can perform complex AI algorithms and data analysis tasks at a location close to the data source, the so-called "edge side". When the image data is transmitted to this chip, it immediately starts the built-in data classification mechanism. This classification is not a simple segmentation, but a comprehensive analysis of the input image based on deep learning models or other advanced machine learning algorithms. For example, in the application scenario of intelligent traffic monitoring, the system can automatically identify and separate key elements such as roads, vehicles, and pedestrians, while ignoring non-relevant areas such as the sky and buildings. In this way, not only can the amount of data to be processed later be effectively reduced and efficiency improved, but also each extracted sub-image can be ensured to be a carrier of information with potential value. Once the preliminary data classification is completed, the next step is to generate the "sub-image data to be processed". At this stage, the algorithm will further refine the results of previous work, such as determining the specific boundaries of each object of interest and cutting it into independent small image blocks. To give a specific example, if a camera captures an abnormal activity (such as an illegal intrusion) in an intelligent security monitoring system, the chip will quickly locate the exact location of the event and then accurately intercept the person or object involved as sub-image data. In this way, the subsequent anomaly detection step can focus on analyzing these important information that have been screened, thereby greatly improving the response speed and accuracy of the entire system. In addition, such a design also meets the strict requirements of modern IoT applications for immediacy and efficiency, ensuring reliable performance even under poor network conditions.

[0022] Step S2, identifying the sub-image data to be processed by a preset recognition algorithm to determine whether there is any abnormality in the sub-image data to be processed. If so, marking the sub-image data to be processed based on the abnormality to obtain a marked abnormality image; wherein there is at least one abnormality that is different from each other.

[0023] Specifically, after the initial classification of image data is achieved, the next key step is to "identify the sub-image data to be processed through a preset recognition algorithm to determine whether there are any abnormalities in the sub-image data to be processed". This process involves using specially designed algorithms to analyze and interpret the previously extracted sub-image data. These algorithms can be neural network models based on deep learning or other types of machine learning methods, which are pre-trained to recognize specific types of targets or behavior patterns. When the sub-image data is input into the edge AI computing and storage integrated chip, the system will start the recognition algorithm to perform a detailed analysis of the content in the image. For example, in the application scenario of intelligent traffic monitoring, the recognition algorithm can detect whether the behavior of moving objects such as pedestrians and vehicles conforms to the expected rules. If a pedestrian is found to suddenly break into a motor vehicle lane or a vehicle is traveling at a speed exceeding the prescribed limit, these behaviors will be regarded as "abnormal situations". Each abnormal situation is unique and has at least one feature that allows it to be distinguished from normal situations. In order to ensure accuracy and reliability, the recognition algorithm may combine multiple technical means, such as target tracking, action recognition, and environmental perception, in order to more fully understand the dynamic changes in the scene. Once the abnormal situation is determined, the system will enter the next step - "marking the sub-image data to be processed based on the abnormal situation to obtain a marked abnormal situation image". In this process, the system will add corresponding marking information to the sub-image data containing the abnormal situation. This kind of marking is not just a simple marking of the location of the abnormal area, but also a detailed record of the specific type of the abnormality and the context in which it occurs. Continuing with the above-mentioned intelligent traffic monitoring as an example, if a speeding car is detected, not only the location of the car will be marked in the image, but also its license plate number, driving direction, and the degree of exceeding the speed limit. This will not only help the subsequent further analysis, but also provide clear and clear evidence support for the relevant departments. In addition, since different abnormal situations may appear in the same frame of the image at the same time, the system must have the ability to handle multiple abnormalities. For example, at the same time, there may be pedestrians running red lights, and there may also be vehicles changing lanes illegally. At this time, the recognition algorithm needs to accurately distinguish and mark each abnormal situation separately to ensure that each abnormality can receive appropriate attention and processing. Such a mechanism not only improves the flexibility and adaptability of the system, but also provides more robust support for complex and changeable practical application scenarios, ensuring that various abnormal events can be captured and responded to in a timely manner even in busy urban traffic environments, thereby effectively maintaining public safety and social order.

[0024] Step S3, performing image compression processing on the annotated abnormal situation image using a preset lossless compression algorithm to obtain abnormal image compressed data.

[0025] Specifically, after the identification and annotation of abnormal situations are completed, the next important step is to "compress the annotated abnormal situation images by using a preset lossless compression algorithm to obtain abnormal image compressed data". This process is designed to ensure that no key details are lost during the transmission and storage of image information, and to effectively reduce the amount of data, thereby optimizing the subsequent data management and processing efficiency. The choice of lossless compression algorithm is crucial because it can significantly reduce the file size without compromising the image quality, which is especially important for data that needs to be stored for a long time or transmitted frequently. Specifically, in the application scenario of intelligent traffic monitoring, when the system detects and annotates abnormal situations such as pedestrians running red lights and vehicles speeding, these annotated images may contain a large amount of redundant information, such as fixed objects or other irrelevant elements in the background. If these complete images are directly transmitted without processing, a large amount of network bandwidth will be occupied and the storage burden of the remote server will be increased. Therefore, it is particularly necessary to process these annotated abnormal situation images using a preset lossless compression algorithm. This compression method can retain all visually discernible features in the image, including the precise location of the annotation, color coding, and any additional text descriptions, while significantly reducing the size of the image file. For example, in actual operation, the system may use lossless compression formats such as JPEG 2000 or PNG to process the annotated abnormal images. These algorithms use complex mathematical models to analyze the spatial redundancy of image data, such as repeated color patterns or texture structures, and remove unnecessary information through efficient encoding methods. The result is that the generated abnormal image compressed data not only maintains all the important features of the original image, but also greatly improves the speed and efficiency of data transmission. For a busy urban traffic monitoring system, this means that even under poor network conditions, the relevant images of abnormal events can be quickly sent to the remote control center or the mobile devices of the relevant responsible personnel so that they can take immediate action. In addition, due to the use of lossless compression technology, these compressed images can still be fully restored when further review or legal evidence is required, ensuring their validity and reliability as evidence. In this way, whether in daily security monitoring or emergency response to emergencies, this method can provide timely and accurate information support to ensure the effective implementation of public safety measures. In short, by effectively compressing the annotated abnormal situation images, not only the best use of resources is achieved, but also a more robust technical guarantee is provided for complex and changeable practical application scenarios.

[0026] Step S4, sending the abnormal image compressed data to a remote control device, and performing abnormal level classification on the abnormal image compressed data through the remote control device to obtain an abnormal classification result.

[0027] Specifically, after the lossless compression of the abnormal image is completed, the next key step is to "send the abnormal image compressed data to the remote control device, and classify the abnormal image compressed data by the remote control device to obtain the abnormal classification result". This process not only involves the secure transmission of data, but also includes a complex decision-making analysis link, which aims to ensure that the remote monitoring center can timely and accurately understand the abnormal situation occurring on the scene and take appropriate countermeasures accordingly. Specifically, after the edge AI computing and storage integrated chip completes the compression processing of the annotated abnormal situation image, the system will use the preset communication protocol, such as TCP / IP or the more efficient MQTT, to safely and reliably transmit these optimized abnormal image compressed data to the remote control device. This remote control device is usually a server platform that integrates advanced data analysis and management functions, which is responsible for receiving data streams from multiple front-end monitoring nodes. Taking intelligent traffic monitoring as an example, if a camera at a certain intersection detects an abnormal situation of a pedestrian running a red light or a vehicle speeding, the edge device corresponding to the road section will immediately send the relevant compressed image data to the server of the urban traffic management center. Once the abnormal image compressed data successfully reaches the remote control device, the next step is to conduct in-depth analysis on it, especially to implement "abnormal level classification". The core of this link is to establish a scientific and reasonable evaluation system to judge the severity and urgency of each abnormal event. For example, in a traffic monitoring scenario, the system may determine the specific level of abnormal behavior based on factors such as the extent to which the vehicle's speed exceeds the standard and the distance that the pedestrian breaks into the motor vehicle lane. Minor violations (such as speeding at low speed) can be marked as lower levels; while behaviors that may cause serious accidents (such as running a red light at high speed) will be classified as high-level abnormalities. In this way, it can not only help managers quickly identify the issues that need the most attention, but also provide clear guidance for subsequent resource scheduling and emergency response. In order to achieve the above goals, specially designed algorithm models are often deployed inside the remote control device. These models may be rule-based expert systems or well-trained machine learning models, which can automatically classify each received abnormal situation according to pre-defined criteria. Finally, through a comprehensive evaluation of all abnormal events, the system generates a detailed "abnormal classification result", which is not just a simple label, but a report containing rich information, such as the time, location, objects involved and their behavioral characteristics of the abnormality. Such a design enables both the traffic management department and other relevant departments to make more informed decisions based on accurate data support, thereby effectively improving the safety and efficiency of urban traffic. In short, through the carefully designed data transmission and analysis process, this method ensures that the entire process from anomaly detection to processing is both efficient and reliable, providing a solid technical guarantee for building an intelligent urban management system.

[0028] Step S5, using a preset priority sorting model, prioritizes the abnormal situations based on the abnormal classification results, obtains an abnormal event priority sequence table, and automatically generates an alarm signal based on the abnormal event priority sequence table, and distributes the alarm signal to the mobile devices or workstations of relevant responsible personnel through a preset communication protocol for early warning prompts.

[0029] Specifically, after the abnormal level classification is completed, the next important step is to "use a preset priority sorting model to prioritize the abnormal situations based on the abnormal classification results, obtain an abnormal event priority sequence list, and automatically generate an alarm signal based on the abnormal event priority sequence list, and distribute the alarm signal to the mobile device or workstation of the relevant responsible personnel through a preset communication protocol for early warning prompts". This process not only ensures that abnormal events can be handled in a timely manner according to their urgency, but also provides clear action guidelines for relevant responsible personnel, thereby effectively improving the efficiency and accuracy of responding to emergencies. Specifically, when the remote control device generates a detailed abnormal classification result based on the abnormal image compression data, the system will start a pre-configured priority sorting model. The task of this model is to comprehensively evaluate all identified abnormal situations and sort them according to their potential impact and urgency. Taking intelligent traffic monitoring as an example, assume that different types of abnormal situations occur at multiple intersections at the same time during a certain period of time, such as pedestrians running red lights, vehicles speeding, and traffic accidents. At this time, the priority sorting model will decide which events should be handled first based on pre-set rules, such as factors such as the severity of the accident, the number of people involved, and the possible harm. For those situations that may cause major safety hazards or require urgent intervention (such as serious traffic accidents), the system will place them at the highest level; while relatively minor behaviors (such as short-term low-speed overspeed) will be assigned to lower priority positions. Once the priority order of each abnormal event is determined, the system will generate an "abnormal event priority sequence list". This list is not just a simple sorting result. It contains key information of each abnormal event, such as the time, location, type and potential impact of the occurrence. Based on this list, the system can further automatically generate corresponding alarm signals. These alarm signals are not only an instant notification of abnormal situations, but also designed to ensure that relevant personnel can take quick action. For example, in the urban traffic management center, the on-duty personnel can quickly deploy resources and formulate effective response strategies based on the different levels of alarms received. More importantly, the alarm signal will be accurately sent to the mobile devices or workstations of the relevant responsible personnel responsible for handling abnormal events in a specific area or type through preset communication protocols, such as SMS, email, instant messaging tools or special application programming interfaces (APIs). Such a mechanism ensures efficient information transmission and coordinated command even in a complex and changing urban environment. For example, when a serious traffic accident occurs at an intersection, traffic police and emergency teams can receive high-priority alerts at the first time and immediately go to the scene to deal with it; for other more routine violations, lower-level alerts can be used to remind relevant departments to follow up and handle them.In summary, by introducing a priority ranking model and an automated alarm generation and distribution mechanism, this method not only improves the scientificity and standardization of abnormal event management, but also provides strong technical support for building a more intelligent and reliable urban safety management system. This not only helps to improve the level of public safety, but also significantly improves the quality of life and travel experience of citizens.

[0030] In a specific embodiment, the step of inputting the image data to be processed into a preset edge AI computing and storage integrated chip for data classification to obtain sub-image data to be processed includes: Inputting the image data to be processed into a preset edge AI computing and storage integrated chip to obtain input data, and performing data preprocessing on the input data to obtain preprocessed data; Performing spectral analysis on the preprocessed data to obtain spectral feature data, performing boundary detection on the spectral feature data using an adaptive threshold segmentation algorithm to obtain boundary features, and performing topological analysis on the boundary features to obtain topological feature data; Performing dimensionality reduction processing on topological feature data through kernel principal component analysis to obtain dimensionality reduction features, and performing manifold learning on the dimensionality reduction features to obtain manifold feature data; A deep belief network is used to classify the manifold feature data to obtain sub-image data to be processed.

[0031] Specifically, in the process of realizing "inputting the image data to be processed into the preset edge AI computing and storage integrated chip for data classification to obtain the sub-image data to be processed", this step not only includes simple image input and preliminary classification, but also involves a series of complex data processing technologies to ensure that the sub-image data finally extracted is both accurate and representative. Specifically, when the original image data is transmitted to the edge AI computing and storage integrated chip, the system first receives these data and converts them into "input data". Next, in order to optimize the effect of subsequent processing, the system will perform necessary "data preprocessing" on the input data. This step may include operations such as removing noise and adjusting brightness and contrast to generate clearer and easier to analyze "preprocessed data". On this basis, the system further performs "spectral analysis" on the preprocessed data, which is a method of revealing the internal structural characteristics of the image through frequency domain transformation. By analyzing different frequency components, "spectral feature data" reflecting the texture and shape characteristics of the image can be obtained. Subsequently, the adaptive threshold segmentation algorithm is used to perform boundary detection on these spectral feature data to identify the contours of each object or area in the image to form "boundary features". This boundary-based analysis helps to distinguish between background and foreground, laying the foundation for subsequent feature extraction. Next, the system will perform "topological analysis" on the boundary features, that is, studying the connection relationship and spatial distribution pattern between each part, so as to obtain more abstract but more descriptive "topological feature data". In order to further simplify the feature representation and improve the computational efficiency, the system will use "Kernel Principal Component Analysis" (KPCA) to reduce the dimension of the topological feature data to obtain "reduced dimension features". KPCA is a nonlinear dimensionality reduction method that can reduce the feature dimension while maintaining the original data structure, making subsequent processing more efficient. After that, the system will also implement "manifold learning" on the reduced dimension features, which is a process aimed at discovering the inherent low-dimensional manifold structure of high-dimensional data. The resulting "manifold feature data" not only retains the key information of the original image, but also enhances its expressive power and generalization performance. Finally, in order to accurately separate the target objects or abnormal situations that need to be paid attention to from the numerous manifold feature data, the system uses "Deep Belief Network" (DBN) for data classification. As a deep neural network model, DBN can automatically learn complex patterns in image data, and finely classify each frame of image according to the pre-trained knowledge base, and finally output "sub-image data to be processed". For example, in the application scenario of intelligent traffic monitoring, after the above series of processing, the system can accurately identify vehicles, pedestrians and various traffic signs on the road, and select those behaviors that may pose safety hazards or violate traffic rules as sub-image data to be processed.In this way, not only can the accuracy of abnormal event detection be effectively improved, but also the workload of the subsequent processing stage can be significantly reduced, ensuring the efficient operation of the entire system. In short, through such a well-designed data processing process, this method not only achieves efficient classification of image data, but also provides a solid technical guarantee for subsequent abnormal event detection and response.

[0032] In a specific embodiment, the sub-image data to be processed is identified by a preset identification algorithm to determine whether there is an abnormality in the sub-image data to be processed, and if so, the sub-image data to be processed is marked based on the abnormality to obtain an image marked with abnormality, including: Using a preset convolutional neural network to extract features from the sub-image data to be processed, to obtain a sub-image feature vector set; wherein each feature vector in the sub-image feature vector set corresponds to a key area in the sub-image data to be processed; Inputting the sub-image feature vector set into a preset recognition algorithm for recognition to determine whether there is an abnormality in the sub-image data to be processed; If so, classifying and labeling the sub-image feature vector set based on the abnormal situation to distinguish between a normal sub-image feature vector set and an abnormal sub-image feature vector set; Using a preset target detection algorithm, the sub-image data to be processed is region-located based on the abnormal sub-image feature vector set to obtain an abnormal region frame; Based on the abnormal region frame, the sub-image data to be processed is classified at the pixel level to obtain a binary segmentation map of the abnormal region and the normal region; wherein the abnormal region is represented by a specific color or grayscale value in the sub-image data to be processed; Performing morphological processing on the binary segmentation image to obtain an optimized abnormal region boundary; Through a preset text generation network, descriptive annotation text is automatically generated for the optimized abnormal region boundary, and a text description is given to the abnormal region to obtain an annotated abnormal situation image.

[0033] Specifically, in the process of realizing "identifying the sub-image data to be processed by a preset recognition algorithm to determine whether there is an abnormality in the sub-image data to be processed", the system first needs to use the preset convolutional neural network (CNN) to perform in-depth feature extraction on these sub-image data. As a powerful deep learning model, CNN is good at automatically capturing complex patterns and structural features from images. When the sub-image data to be processed is input into the edge AI computing and storage integrated chip, CNN will perform multi-layer convolution operations, pooling operations, and nonlinear activation function processing on it, and finally generate one or more "sub-image feature vector sets". Each feature vector set contains information corresponding to a specific key area, which can highly summarize the important attributes of the object in the area, such as shape, texture, color, etc. For example, in the application scenario of intelligent traffic monitoring, assuming that a camera at a certain intersection captures a speeding vehicle, then the CNN in the system will focus on extracting the key features of the vehicle and its surrounding environment, such as license plate number, body color, driving direction, etc. In this way, the system can ensure that the object of subsequent analysis is a truly valuable data fragment, rather than redundant background information. The obtained sub-image feature vector set not only reflects the spatial distribution of the original image, but also implies the dynamic changes in the time dimension, providing a solid foundation for the next step of recognition. Next, the system will input these sub-image feature vector sets into the preset recognition algorithm for further analysis to determine whether there are indeed abnormal conditions in the sub-image data to be processed. The recognition algorithm here may include but is not limited to support vector machine (SVM), random forest (Random Forest) or other machine learning methods, which have been trained with a large number of samples and have the ability to distinguish between normal and abnormal conditions. For intelligent traffic monitoring, this means that the system can evaluate whether the currently captured behavior meets the expected standards based on historical data and rule bases. If the detection results show that there is an abnormality - such as the aforementioned vehicle speeding - then enter the next step: classify and label the sub-image feature vector set based on the abnormal situation, so as to distinguish between the normal sub-image feature vector set and the abnormal sub-image feature vector set. In order to more accurately locate the area where the abnormality occurs, the system will use the preset target detection algorithm to locate the area of ​​the sub-image data to be processed based on the abnormal sub-image feature vector set to obtain the "abnormal area frame". Object detection algorithms usually combine the functions of bounding box regression and category prediction to mark all possible locations of abnormal behavior in the image and give their specific coordinate ranges. Taking intelligent traffic monitoring as an example, if a car running a red light is found, the system will accurately circle the location of the car in the image and record its license plate number and other relevant information. This not only helps to preserve evidence later, but also provides an intuitive operation interface for managers.Once the specific location of the abnormal area is determined, the system will further obtain a binary segmentation map of the abnormal area and the normal area through a pixel-level classification method. In this step, each pixel will be judged as belonging to the abnormal area or the normal area, and represented by a specific color or grayscale value. For example, the abnormal area may be highlighted in red, while the normal part remains unchanged in its original color. Such a binary segmentation map is not only intuitive and clear, but also convenient for subsequent morphological processing. Through morphological operations such as expansion and corrosion, the system can optimize the boundaries of the abnormal area and remove unnecessary noise interference, making the final result smoother and easier to understand. Finally, in order to provide more detailed information support, the system will also automatically generate descriptive annotation text for the optimized abnormal area boundaries through a preset text generation network, and provide a text description of the abnormal area to obtain a complete "annotated abnormal situation image". The text generation network can generate concise and clear text descriptions based on natural language processing technology and pre-defined templates or rules, such as "the vehicle ran a red light at XX intersection at XX hours and XX minutes". These descriptive texts not only enhance the readability and practicality of image information, but also provide a reliable written basis for legal proceedings or other formal purposes. For example, in traffic accident investigations or law enforcement actions, image data with detailed annotations can help relevant personnel quickly understand the full picture of the incident and improve work efficiency. In summary, through a series of complex and orderly operations such as feature extraction, classification recognition, region positioning, pixel-level segmentation, morphological processing, and text generation of sub-image data, this method not only achieves accurate detection and annotation of abnormal situations, but also provides relevant responsible personnel with rich decision-making support information. Whether in daily security monitoring or emergency response to emergencies, this method can effectively improve the intelligence level of the system and ensure the stable maintenance of public safety and social order.

[0034] In a specific embodiment, the image compression processing is performed on the annotated abnormal situation image by using a preset lossless compression algorithm to obtain abnormal image compressed data, including: Using an image edge detection algorithm to extract edges of the annotated abnormal situation image to obtain edge image data, and performing pixel value statistics on the edge image data to obtain a pixel value frequency distribution table; Performing Huffman tree construction on the pixel value frequency distribution table through a Huffman tree construction algorithm to obtain a Huffman coding tree, and performing coding table generation on the Huffman coding tree to obtain a Huffman coding table; Performing Huffman coding on the annotated abnormal situation image based on the Huffman coding table to obtain a Huffman coding data stream; Performing run-length encoding on the Huffman-encoded data stream to obtain RLE-encoded data; Segmenting the RLE encoded data into data blocks using a preset lossless compression algorithm to obtain segmented RLE encoded data; The segmented RLE encoded data is compressed using a preset lossless compression algorithm to obtain abnormal image compressed data.

[0035] Specifically, in the process of realizing "compressing the annotated abnormal situation image by a preset lossless compression algorithm to obtain abnormal image compressed data", the system will first use an image edge detection algorithm to extract the edge of the annotated abnormal situation image. This process aims to highlight the areas with significant changes in the image, such as object contours or important feature points, so as to provide a clearer and more concise basis for subsequent data compression. Edge detection algorithms are usually based on methods such as gradient calculation or Laplace operator, which can effectively identify the boundaries with large brightness changes in the image. For example, in the application scenario of intelligent traffic monitoring, assuming that a camera at a certain intersection captures the behavior of a pedestrian running a red light, and the behavior has been accurately marked, then the system will use the edge detection algorithm to extract the outline of the pedestrian, road markings and other key objects from the annotated image to generate "edge image data". Next, the system will perform pixel value statistics on these edge image data to establish a detailed "pixel value frequency distribution table". This table records the number of times each pixel value (i.e., grayscale or color value) appears, reflecting the distribution of different brightness or color components in the image. For the above-mentioned example of intelligent traffic monitoring, most of the background area in the edge image may appear to be a relatively uniform color, while the outline of the pedestrian is composed of a series of high-contrast edges. Therefore, by statistically analyzing the frequency distribution of these pixel values, we can better understand which parts of the image contain more detailed information and which parts are relatively smooth. With the pixel value frequency distribution table, the system will further apply the Huffman tree construction algorithm to create a "Huffman coding tree". Huffman coding is a classic lossless data compression technology that assigns different coding lengths according to the probability of the occurrence of symbols, so that frequently occurring symbols use shorter codes, while less common symbols use longer codes. Specifically, in this step, the system will construct a Huffman tree according to the data in the pixel value frequency distribution table, in which the pixel values ​​with higher probabilities are located near the root node, while the pixel values ​​with lower probabilities are located at the ends of the farther branches. In this way, the system generates a "Huffman coding table", which defines the optimal binary coding sequence corresponding to each pixel value, ensuring that the compressed data not only retains all the details of the original image, but also minimizes redundant information. Next, the system performs Huffman encoding on the anomaly-labeled image based on the generated Huffman encoding table, thereby obtaining a "Huffman-encoded data stream". This means that the entire image is converted into a series of optimized binary strings that represent the encoding results of each pixel in the original image. The advantage of Huffman encoding is that it can significantly reduce the file size without losing any information, and is particularly suitable for data that needs to be stored for a long time or transmitted frequently. However, in order to further improve the compression efficiency, the system will also perform run-length encoding (RLE) on the Huffman-encoded data stream.RLE is a simple lossless compression method that is suitable for data sequences containing a large number of consecutive identical values. For example, in a Huffman-encoded image, if there are large areas of the same background color, RLE can simplify these data by recording the number of repetitions to obtain "RLE encoded data". In order to ensure that the compressed data can be stored and transmitted efficiently, the system also needs to segment the RLE encoded data into data blocks to obtain "segmented RLE encoded data". This step is mainly to adapt to different application scenarios and technical requirements, because in some cases, large blocks of data may not be suitable for direct processing or storage. Through a reasonable block strategy, the entire encoded data can be divided into multiple smaller units for subsequent operations. Finally, the system will call the preset lossless compression algorithm again to perform the final data compression on these segmented RLE encoded data to obtain "abnormal image compressed data". This compression not only retains all the visual features and annotation information in the image, but also greatly improves the speed and storage efficiency of data transmission. For example, in intelligent traffic monitoring, when a speeding vehicle is detected and marked, the system will execute the above compression process in sequence. First, through edge detection and pixel value statistics, the system determines the most important features in the image; then, Huffman coding and RLE coding are used to further reduce the amount of data; then, by reasonably dividing the encoded data into blocks, it ensures that it is easy to manage and transmit; finally, with the help of an efficient lossless compression algorithm, the system generates a compact and complete abnormal image compressed data. This compressed data can not only be quickly uploaded to a remote server for management personnel to review, but also completely restore the original image when necessary, as a supporting material for legal evidence or other formal purposes. In short, through this series of carefully designed compression steps, the system not only achieves effective processing of abnormal images, but also provides reliable technical support for complex and changeable practical application scenarios.

[0036] In a specific embodiment, the abnormal level classification of the abnormal image compression data by the remote control device to obtain the abnormal classification result includes: Extracting features from the abnormal image compression data using an image feature extraction algorithm in the remote control device to obtain an abnormal feature data set, and sorting the abnormal feature data set to obtain an ordered feature sequence; Recalling a historical ordered feature sequence from a database, performing quartile calculation on the historical ordered feature sequence through a quartile calculator to obtain a historical quartile distribution, and performing outlier detection on the historical quartile distribution to obtain a historical outlier range; Binning the abnormal feature data set based on the historical abnormal value range to obtain a feature binning result, and performing boundary optimization on the feature binning result to obtain an optimized binning boundary; A hierarchical clustering algorithm is used to perform cluster analysis on the optimized bin boundaries to obtain abnormal level clustering results, and density estimation is performed on the abnormal level clustering results to obtain level density distribution; The grade density distribution is scored and calculated by a multi-index comprehensive scorer to obtain an abnormal grade score, and the abnormal grade score is graded and mapped to obtain an abnormal grade label; The abnormal image compression data is graded based on the abnormal grade label to obtain the graded abnormal data, and the graded abnormal data is encapsulated to obtain the abnormal classification result.

[0037] Specifically, in the process of realizing "using the remote control device to classify the abnormal image compressed data into abnormal levels and obtain the abnormal classification results", the system first needs to use the image feature extraction algorithm in the remote control device to perform detailed feature extraction on the abnormal image compressed data. This link aims to recover and identify those key feature information that can reflect abnormal situations from the compressed image data. For example, in the application scenario of intelligent traffic monitoring, when the system receives the abnormal image compressed data of a speeding vehicle captured by a camera at a certain intersection, it will start the image feature extraction algorithm to parse the data. The algorithm may combine a variety of technical means, such as edge detection, color histogram analysis, and texture descriptors, to ensure that every subtle change in the image can be captured comprehensively and accurately. In this way, the system generates an "abnormal feature data set" containing all relevant features, and performs preliminary data sorting on it to form an "ordered feature sequence". In order to further optimize the accuracy of the abnormal level classification, the system will then call the historical ordered feature sequence from the database and use the quartile calculator to perform quartile calculations on these historical data. Quartile is a statistical method used to divide a set of data into four equal parts, so as to better understand the distribution of the data. Specifically, by calculating the quartiles of the historical ordered feature sequence, the system can obtain a "historical quartile distribution". This distribution not only reflects the feature distribution law of similar abnormal events in the past, but also provides an important reference benchmark for the current abnormal feature data set. On this basis, the system will perform outlier detection on the historical quartile distribution and determine the "historical outlier range". For example, in intelligent traffic monitoring, if vehicle speeding occurs frequently in a certain period of time, the feature values ​​corresponding to these behaviors may be regarded as outliers and thus included in the historical outlier range. With the historical outlier range as a reference, the system then performs binning processing on the abnormal feature data set based on this to obtain the "feature binning result". Binning is a method of data discretization, which divides continuous data into several intervals (or boxes) according to certain rules for subsequent analysis and processing. In this step, the system will reasonably set the binning boundaries according to the historical outlier range, and optimize these boundaries to ensure that the data in each box is similar and can effectively distinguish different abnormal situations. After boundary optimization, the system obtains the "optimized binning boundaries", which lays a solid foundation for further clustering analysis. Next, the system uses a hierarchical clustering algorithm to perform cluster analysis on the optimized bin boundaries to reveal the internal connections and hierarchical structures between different abnormal situations. The hierarchical clustering algorithm gradually merges or splits data points to construct a tree structure that represents the similarity relationship between data, and finally obtains the "abnormal level clustering result."This clustering result can not only intuitively show the category of each abnormal event, but also reflect the difference in their relative importance. In order to more accurately measure the importance of each cluster, the system will also perform density estimation on the abnormal level clustering results to obtain the "level density distribution". Density estimation can help identify which areas are clustered with a large number of abnormal events, thereby providing a scientific basis for the determination of abnormal levels. Finally, the system uses a multi-indicator comprehensive scorer to score and calculate the level density distribution to obtain the "abnormal level score". This scoring process takes into account multiple factors, such as the frequency, severity and potential hazards of abnormal occurrences, to ensure that each abnormal event can obtain a reasonable score. Subsequently, the system will perform level mapping on the abnormal level score to generate a clear "abnormal level label". These labels not only clearly indicate the urgency of each abnormal event, but also provide specific response suggestions for managers. For example, in intelligent traffic monitoring, when the system detects a serious traffic accident, it will be assigned a higher abnormal level label to remind relevant personnel to take immediate action; while for some minor violations, lower level labels can be assigned to arrange subsequent processing. Based on the above abnormal level labels, the system graded the abnormal image compression data to form "labeled abnormal data". For ease of management and use, the system will also encapsulate the results of these annotated abnormal data, and finally obtain a complete "abnormal classification result". In summary, through a series of complex and orderly operations such as feature extraction, historical data analysis, binning, cluster analysis, density estimation, and multi-indicator comprehensive scoring of abnormal image compression data, this method not only achieves accurate classification of abnormal events, but also provides detailed decision support information for relevant responsible personnel. Whether in daily security monitoring or emergency response to sudden incidents, this method can significantly improve the intelligence level of the system and ensure the stable maintenance of public safety and social order.

[0038] In a specific embodiment, the preset priority sorting model is used to prioritize the abnormal situations based on the abnormal classification results to obtain an abnormal event priority sequence table, including: An abnormality level analyzer is used to numerically quantize the abnormality classification result to obtain an abnormality level quantization vector; Performing timeliness calculation on the abnormality level quantization vector through a time decay function to obtain a timeliness abnormality feature, and performing an influence range analysis on the timeliness abnormality feature to obtain an influence range vector; Assigning weights to the influence range vector based on the hierarchical analysis method to obtain a weighted influence range vector; Using a preset priority sorting model, the abnormal situations are prioritized based on the weight impact range vector to obtain an initial priority sequence, and conflict detection and elimination are performed on the initial priority sequence to obtain an initial priority sequence after elimination; The initial priority sequence after elimination is subjected to constraint optimization through preset priority constraint conditions to obtain an optimized priority sequence; The optimized priority sequence is formatted to obtain an abnormal event priority sequence table.

[0039] Specifically, in the process of realizing "using a preset priority sorting model, prioritizing the abnormal situations based on the abnormal classification results, and obtaining an abnormal event priority sequence table", the system first needs to use an abnormal level parser to numerically quantify the abnormal classification results. The core of this link is to convert qualitative abnormal level labels into specific numerical representations, thereby providing basic data support for subsequent calculations and analysis. For example, in the application scenario of intelligent traffic monitoring, assuming that the system has generated a series of events with clear abnormal level labels (such as speeding, running red lights, etc.) based on historical data analysis and multi-indicator comprehensive scoring, then the abnormal level parser will map these labels to corresponding numerical scores to form an "abnormal level quantization vector". This quantization process not only simplifies the data processing process, but also enables different types of abnormal situations to be compared and sorted on the same scale. Next, in order to consider the time factor of abnormal events, the system will calculate the timeliness of the abnormal level quantization vector through the time decay function. The time decay function is a mathematical model that can reflect the changing trend of the importance of events over time. Specifically, for abnormal situations that occurred in a relatively recent period of time, their weights will be assigned higher values; while those that occurred at an earlier time point will gradually lose some of their attention as time passes. By applying the time decay function to the abnormal level quantification vector, the system can obtain the "time-sensitive abnormal characteristics", which ensures that urgent and recent abnormal situations can occupy a more important position in the sorting. In addition, the system will also analyze the impact range of these time-sensitive abnormal characteristics to determine the area or population size that each abnormal event may affect, and finally generate an "impact range vector". Based on the Analytic Hierarchy Process (AHP), the system will assign weights to the above impact range vectors to obtain a "weighted impact range vector". The Analytic Hierarchy Process is a widely used decision analysis tool that evaluates the relative importance of various factors by constructing a hierarchical structure model and combining expert judgment or historical data. In this step, the system will assign appropriate weights to the impact range of each abnormal event based on pre-set criteria, such as event type, geographical location, number of people affected, etc. This not only improves the scientificity and rationality of the sorting results, but also better reflects the complex needs in actual application scenarios. For example, in intelligent traffic monitoring, if a serious traffic accident occurs at an intersection, its impact range may involve road traffic in multiple directions, so the event will be given a higher weight to ensure that it is at the top of the priority sorting. With the weighted impact range vector as input, the system then uses a preset priority sorting model to comprehensively sort all abnormal situations. This priority sorting model integrates multiple algorithms and technologies to comprehensively consider various factors and give the best sorting solution.The system first preliminarily sorts the abnormal situations based on the weighted impact range vector to obtain the "initial priority sequence". However, since some abnormal events may conflict or overlap (such as traffic accidents occurring at multiple intersections at the same time), the system needs to further perform conflict detection and elimination to ensure that each event can obtain a reasonable processing order. After conflict detection and elimination, the system obtains the "initial priority sequence after elimination". This step effectively avoids resource waste and duplication of work and improves the overall response efficiency. In order to further optimize the sorting results, the system will constrain and optimize the initial priority sequence after elimination through preset priority constraints to obtain the "optimized priority sequence". These constraints may be stipulated by policies and regulations, emergency plans or specific business rules, in order to ensure that the sorting results meet the actual situation and management requirements. For example, in intelligent traffic monitoring, if a certain type of accident requires special equipment or professionals to be on site for processing, the system will give priority to the availability and deployment of these resources when sorting. Finally, for ease of use and display, the system will convert the optimized priority sequence format and finally generate a clear and easy-to-understand "abnormal event priority sequence table". In summary, through a series of carefully designed operations such as numerical quantification, timeliness calculation, impact range analysis, weight assignment, priority sorting, and constraint optimization of the abnormal division results, this method not only achieves accurate priority sorting of abnormal events, but also provides detailed and intuitive information support for relevant responsible personnel. Whether in daily security monitoring or emergency response to sudden incidents, this method can significantly improve the intelligence level of the system and ensure the effective maintenance of public safety and social order. For example, in the urban traffic management center, the on-duty personnel can quickly allocate resources and formulate effective response strategies based on this abnormal event priority sequence list, so as to deal with various abnormal situations in a timely manner and protect the lives and property of citizens.

[0040] In a specific embodiment, the automatic generation of an alarm signal based on the abnormal event priority sequence table and the distribution of the alarm signal to the mobile device or workstation of the relevant responsible personnel for early warning through a preset communication protocol include: Performing semantic analysis on the abnormal event priority sequence table through a natural language generator to obtain an event semantic vector, and performing context association analysis on the event semantic vector to obtain associated semantic data; wherein the associated semantic data includes event description, impact degree and disposal suggestion; Performing information integration on the associated semantic data to obtain rich media alarm content, and automatically generating an alarm signal based on the rich media alarm content; Performing distributed transaction processing on the alarm distribution task sequence to obtain an alarm distribution task sequence, and performing priority scheduling analysis on the alarm distribution task sequence based on the abnormal event priority sequence table to obtain a task scheduling plan; Dynamically allocating resources to the alarm distribution task sequence based on the task scheduling scheme to obtain a resource allocation strategy table, and optimizing the network topology of the alarm distribution task sequence based on the resource allocation strategy table to obtain an optimal transmission path; Through a preset communication protocol, the alarm distribution task sequence is distributed to the mobile devices or workstations of relevant responsible personnel based on the optimal transmission path for early warning prompts.

[0041] Specifically, in the process of realizing "automatically generating an alarm signal based on the abnormal event priority sequence table, and distributing the alarm signal to the mobile device or workstation of the relevant responsible personnel for early warning through a preset communication protocol", the system first needs to use a natural language generator to perform semantic analysis on the abnormal event priority sequence table. This link aims to convert the abstract priority sorting results into easy-to-understand text descriptions to facilitate subsequent information transmission and processing. Specifically, when the system receives a precisely sorted abnormal event priority sequence table, the natural language generator will perform in-depth semantic analysis on it, extract the key information of each event, such as time, location, type, etc., and convert it into a structured "event semantic vector". For example, in the application scenario of intelligent traffic monitoring, if a serious traffic accident occurs at a certain intersection, the natural language generator will parse the specific location of the accident (such as the intersection of XX Road and YY Road), the time of occurrence (such as 14:30) and the objects involved (such as a truck and two cars), thereby forming a semantic vector containing these detailed information. Next, the system will perform context association analysis on these event semantic vectors to obtain richer and more concrete "association semantic data". This kind of association analysis not only considers the characteristics of a single event itself, but also combines the influence of the surrounding environment and other relevant factors to ensure that each alarm can provide comprehensive and accurate information support. For example, for the aforementioned traffic accident, the system may further analyze its impact range (such as whether it causes congestion of surrounding roads), potential hazards (such as whether there are casualties) and corresponding disposal suggestions (such as immediately dispatching traffic police and emergency teams). In this way, the system can generate complete association semantic data including "event description, impact level and disposal suggestions", laying a solid foundation for the subsequent alarm content production. With detailed association semantic data, the system will integrate the information to obtain "rich media alarm content". The rich media mentioned here is not limited to text descriptions, but can also include pictures, video clips and even audio files. It aims to improve the effectiveness and intuitiveness of alarms through multi-dimensional information presentation. For example, in intelligent traffic monitoring, in addition to the above-mentioned text descriptions, the system can also attach photos or real-time video streams of the accident site to help managers understand the specific situation more clearly. In addition, to ensure the coherence and logic of the alarm content, the system will also organize and arrange all the information reasonably, and finally automatically generate a complete alarm signal, which contains both necessary text explanations and rich multimedia materials, so that the receiver can grasp the key information and take action in the shortest time. In order to ensure that the alarm signal can be efficiently and orderly transmitted to the relevant personnel, the system will perform distributed transaction processing on the alarm distribution process to obtain the "alarm distribution task sequence".Distributed transaction processing is an important mechanism to ensure the consistency of operations between multiple nodes. It can help the system coordinate the order of task execution of different devices in a complex network environment. On this basis, the system will perform priority scheduling analysis on the alarm distribution task sequence according to the abnormal event priority sequence table and formulate a reasonable "task scheduling plan". For example, in the urban traffic management center, if multiple abnormal situations of different levels occur at the same time, the system will give priority to the alarm sending tasks corresponding to high-priority events to ensure that the most important information can be conveyed to the relevant responsible personnel as soon as possible. In this way, the system not only improves the response speed, but also avoids delays or confusion caused by resource contention. After determining the task scheduling plan, the system will dynamically allocate resources for the alarm distribution task sequence based on this to obtain a "resource allocation strategy table". This strategy table clarifies the computing resources, bandwidth requirements, and expected completion time required for each task, providing specific guidance for subsequent operations. In order to further optimize the transmission efficiency, the system will also optimize the network topology of the alarm distribution task sequence based on the resource allocation strategy table to find the "optimal transmission path". Network topology optimization involves a series of technical means such as selecting the best route and adjusting node configuration, with the aim of ensuring that the alarm signal can be delivered to the target terminal at the fastest speed and with the highest reliability. For example, in intelligent traffic monitoring, if an alarm at a certain intersection needs to be notified to the nearby traffic police brigade, the system will select the shortest path from the monitoring center to the traffic police brigade and ensure that the network nodes along the way have sufficient bandwidth and stability to ensure the timely transmission of the alarm information. Finally, based on the determined optimal transmission path, the system will distribute the alarm distribution task sequence to the mobile devices or workstations of the relevant responsible personnel for early warning prompts. This means that whether through mobile phone text messages, application push or dedicated workstation interface, the personnel responsible for handling abnormal situations can quickly receive alarm information and take appropriate countermeasures accordingly. For example, in intelligent traffic monitoring, when the on-duty traffic police receive an alarm of a serious traffic accident, they can immediately go to the scene to direct and guide, ensuring that the traffic order is restored to normal as soon as possible. In short, through the preset communication protocol, through a series of carefully designed operations such as automatic generation of alarm signals, semantic analysis, information integration, task scheduling, resource allocation and network optimization, this method not only achieves efficient early warning and response to abnormal events, but also provides detailed and intuitive information support for relevant responsible personnel, significantly improving the intelligence level and emergency handling capabilities of the system.

[0042] The above describes the data processing method based on the edge AI computing and storage integrated chip in the embodiment of the present invention. The following describes the data processing device based on the edge AI computing and storage integrated chip in the embodiment of the present invention. Figure 2 In one embodiment of the present invention, a data processing device based on an edge AI computing and storage integrated chip includes: The classification module 21 is used to input the image data to be processed into a preset edge AI computing and storage integrated chip for data classification to obtain sub-image data to be processed; The recognition module 22 is used to recognize the sub-image data to be processed by a preset recognition algorithm to determine whether there is an abnormality in the sub-image data to be processed. If there is, the sub-image data to be processed is marked based on the abnormality to obtain a marked abnormality image; wherein the abnormality has at least one and is different from each other; A compression module 23 is used to perform image compression processing on the annotated abnormal situation image by using a preset lossless compression algorithm to obtain abnormal image compressed data; A classification module 24, used for sending the abnormal image compressed data to a remote control device, and performing abnormal level classification on the abnormal image compressed data through the remote control device to obtain an abnormal classification result; The sorting module 25 is used to adopt a preset priority sorting model to prioritize the abnormal situations based on the abnormal classification results, obtain an abnormal event priority sequence table, and automatically generate an alarm signal based on the abnormal event priority sequence table, and distribute the alarm signal to the mobile devices or workstations of relevant responsible personnel through a preset communication protocol for early warning prompts.

[0043] In this embodiment, for the specific implementation of each unit in the above system embodiment, please refer to the above method embodiment, which will not be repeated here.

[0044] Reference Figure 3 The present invention also provides a computer device in an embodiment, wherein the internal structure of the computer device can be as follows: Figure 3 As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0045] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0046] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0047] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0048] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0049] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A data processing method based on an edge AI computing and storage integrated chip, characterized in that: The following steps are involved: The image data to be processed is input into the preset edge AI computing and storage integrated chip for data classification to obtain the sub-image data to be processed; The sub-image data to be processed is identified by a preset identification algorithm to determine whether there is an abnormality in the sub-image data to be processed; if so, the sub-image data to be processed is marked based on the abnormality to obtain an image marked with abnormality; wherein the abnormality has at least one and is different from each other; Performing image compression processing on the annotated abnormal situation image by using a preset lossless compression algorithm to obtain abnormal image compressed data; The abnormal image compressed data is sent to a remote control device, and the abnormal image compressed data is classified into abnormal levels by the remote control device to obtain an abnormal classification result; A preset priority sorting model is adopted to prioritize the abnormal situations based on the abnormal classification results to obtain an abnormal event priority sequence list, and an alarm signal is automatically generated based on the abnormal event priority sequence list. The alarm signal is distributed to the mobile devices or workstations of relevant responsible personnel through a preset communication protocol for early warning prompts.

2. The data processing method based on the edge AI computing and storage integrated chip according to claim 1 is characterized in that: The step of inputting the image data to be processed into a preset edge AI computing and storage integrated chip for data classification to obtain sub-image data to be processed includes: Inputting the image data to be processed into a preset edge AI computing and storage integrated chip to obtain input data, and performing data preprocessing on the input data to obtain preprocessed data; Performing spectral analysis on the preprocessed data to obtain spectral feature data, performing boundary detection on the spectral feature data using an adaptive threshold segmentation algorithm to obtain boundary features, and performing topological analysis on the boundary features to obtain topological feature data; Performing dimensionality reduction processing on topological feature data through kernel principal component analysis to obtain dimensionality reduction features, and performing manifold learning on the dimensionality reduction features to obtain manifold feature data; A deep belief network is used to classify the manifold feature data to obtain sub-image data to be processed.

3. The data processing method based on the edge AI computing and storage integrated chip according to claim 1 is characterized in that: The method of identifying the sub-image data to be processed by a preset recognition algorithm to determine whether there is an abnormality in the sub-image data to be processed, and if so, marking the sub-image data to be processed based on the abnormality to obtain a marked abnormality image, includes: Using a preset convolutional neural network to extract features from the sub-image data to be processed, to obtain a sub-image feature vector set; wherein each feature vector in the sub-image feature vector set corresponds to a key area in the sub-image data to be processed; Inputting the sub-image feature vector set into a preset recognition algorithm for recognition to determine whether there is an abnormality in the sub-image data to be processed; If so, classifying and labeling the sub-image feature vector set based on the abnormal situation to distinguish between a normal sub-image feature vector set and an abnormal sub-image feature vector set; Using a preset target detection algorithm, the sub-image data to be processed is region-located based on the abnormal sub-image feature vector set to obtain an abnormal region frame; Based on the abnormal region frame, the sub-image data to be processed is classified at the pixel level to obtain a binary segmentation map of the abnormal region and the normal region; wherein the abnormal region is represented by a specific color or grayscale value in the sub-image data to be processed; Performing morphological processing on the binary segmentation image to obtain an optimized abnormal region boundary; Through a preset text generation network, descriptive annotation text is automatically generated for the optimized abnormal region boundary, and a text description is given to the abnormal region to obtain an annotated abnormal situation image.

4. The data processing method based on the edge AI computing and storage integrated chip according to claim 1 is characterized in that: The step of performing image compression processing on the annotated abnormal situation image by using a preset lossless compression algorithm to obtain abnormal image compressed data includes: Using an image edge detection algorithm to extract edges of the annotated abnormal situation image to obtain edge image data, and performing pixel value statistics on the edge image data to obtain a pixel value frequency distribution table; Performing Huffman tree construction on the pixel value frequency distribution table through a Huffman tree construction algorithm to obtain a Huffman coding tree, and performing coding table generation on the Huffman coding tree to obtain a Huffman coding table; Performing Huffman coding on the annotated abnormal situation image based on the Huffman coding table to obtain a Huffman coding data stream; Performing run-length encoding on the Huffman-encoded data stream to obtain RLE-encoded data; Segmenting the RLE encoded data into data blocks using a preset lossless compression algorithm to obtain segmented RLE encoded data; The segmented RLE encoded data is compressed using a preset lossless compression algorithm to obtain abnormal image compressed data.

5. The data processing method based on the edge AI computing and storage integrated chip according to claim 1 is characterized in that: The step of performing abnormal level classification on the abnormal image compression data by the remote control device to obtain an abnormal classification result includes: Extracting features from the abnormal image compression data using an image feature extraction algorithm in the remote control device to obtain an abnormal feature data set, and sorting the abnormal feature data set to obtain an ordered feature sequence; Recalling a historical ordered feature sequence from a database, performing quartile calculation on the historical ordered feature sequence through a quartile calculator to obtain a historical quartile distribution, and performing outlier detection on the historical quartile distribution to obtain a historical outlier range; Binning the abnormal feature data set based on the historical abnormal value range to obtain a feature binning result, and performing boundary optimization on the feature binning result to obtain an optimized binning boundary; A hierarchical clustering algorithm is used to perform cluster analysis on the optimized bin boundaries to obtain abnormal level clustering results, and density estimation is performed on the abnormal level clustering results to obtain level density distribution; The grade density distribution is scored and calculated by a multi-index comprehensive scorer to obtain an abnormal grade score, and the abnormal grade score is graded and mapped to obtain an abnormal grade label; The abnormal image compression data is graded based on the abnormal grade label to obtain the graded abnormal data, and the graded abnormal data is encapsulated to obtain the abnormal classification result.

6. The data processing method based on the edge AI computing and storage integrated chip according to claim 1 is characterized in that: The preset priority sorting model is used to prioritize the abnormal situations based on the abnormal classification results to obtain an abnormal event priority sequence table, including: An abnormality level analyzer is used to numerically quantize the abnormality classification result to obtain an abnormality level quantization vector; Performing timeliness calculation on the abnormality level quantization vector through a time decay function to obtain a timeliness abnormality feature, and performing an influence range analysis on the timeliness abnormality feature to obtain an influence range vector; Assigning weights to the influence range vector based on the hierarchical analysis method to obtain a weighted influence range vector; Using a preset priority sorting model, the abnormal situations are prioritized based on the weight impact range vector to obtain an initial priority sequence, and conflict detection and elimination are performed on the initial priority sequence to obtain an initial priority sequence after elimination; The initial priority sequence after elimination is subjected to constraint optimization through preset priority constraint conditions to obtain an optimized priority sequence; The optimized priority sequence is formatted to obtain an abnormal event priority sequence table.

7. The data processing method based on the edge AI computing and storage integrated chip according to claim 1 is characterized in that: The automatic generation of an alarm signal based on the abnormal event priority sequence table, and the distribution of the alarm signal to the mobile device or workstation of the relevant responsible personnel for early warning through a preset communication protocol, includes: Performing semantic analysis on the abnormal event priority sequence table through a natural language generator to obtain an event semantic vector, and performing context association analysis on the event semantic vector to obtain associated semantic data; wherein the associated semantic data includes event description, impact degree and disposal suggestion; Performing information integration on the associated semantic data to obtain rich media alarm content, and automatically generating an alarm signal based on the rich media alarm content; Performing distributed transaction processing on the alarm distribution task sequence to obtain an alarm distribution task sequence, and performing priority scheduling analysis on the alarm distribution task sequence based on the abnormal event priority sequence table to obtain a task scheduling plan; Dynamically allocating resources to the alarm distribution task sequence based on the task scheduling scheme to obtain a resource allocation strategy table, and optimizing the network topology of the alarm distribution task sequence based on the resource allocation strategy table to obtain an optimal transmission path; Through a preset communication protocol, the alarm distribution task sequence is distributed to the mobile devices or workstations of relevant responsible personnel based on the optimal transmission path for early warning prompts.

8. A data processing device based on an edge AI computing and storage integrated chip, characterized in that: include: A classification module is used to input the image data to be processed into a preset edge AI computing and storage integrated chip for data classification to obtain sub-image data to be processed; an identification module, used to identify the sub-image data to be processed by a preset identification algorithm to determine whether there is an abnormality in the sub-image data to be processed, and if so, to mark the sub-image data to be processed based on the abnormality to obtain an image marked with abnormality; wherein the abnormality has at least one and is different from each other; A compression module, used for performing image compression processing on the annotated abnormal situation image by using a preset lossless compression algorithm to obtain abnormal image compressed data; A classification module, used for sending the abnormal image compressed data to a remote control device, and performing abnormal level classification on the abnormal image compressed data through the remote control device to obtain an abnormal classification result; A sorting module is used to adopt a preset priority sorting model to prioritize the abnormal situations based on the abnormal classification results, obtain an abnormal event priority sequence table, and automatically generate an alarm signal based on the abnormal event priority sequence table, and distribute the alarm signal to the mobile devices or workstations of relevant responsible personnel through a preset communication protocol for early warning prompts.

9. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images