An intelligent inspection management method and system for improving security maintenance efficiency

Abstract

The present invention relates to an intelligent inspection management method and system for improving security maintenance efficiency, which relates to the field of inspection management. A binocular camera is used to obtain images of a monitoring area and extract meta-attributes. A pre-trained model is used to determine the weight distribution of multi-dimensional security maintenance indicators. The image is semantically segmented to generate a set of sub-areas. The risk quantification value of each sub-area is then calculated based on the weight distribution, and the value is mapped to an inspection cycle query table to generate an inspection task instruction set. This solves the technical problem of insufficient efficiency and accuracy of security maintenance due to unreasonable inspection task allocation, realizes intelligent management of inspection tasks, and effectively improves the efficiency and accuracy of security maintenance.

Description

Technical Field

[0001] The present invention relates to the field of patrol inspection management, and in particular to an intelligent patrol inspection management method and system for improving security maintenance efficiency. Background Art

[0002] In modern society, with the acceleration of urbanization and the large-scale construction of various public facilities, commercial venues, and industrial areas, the importance of security and maintenance is becoming increasingly prominent. Traditional security and maintenance methods often rely on manual inspections and regular maintenance. This approach is not only inefficient but also susceptible to human factors, resulting in inconsistent inspection quality and making it difficult to effectively prevent and respond to emergencies.

[0003] The rapid development of information technology, especially the widespread application of computer vision, artificial intelligence, and big data, has made it possible to upgrade security and maintenance work to intelligent capabilities. However, existing intelligent security and maintenance systems mostly focus on single functions, such as video surveillance and intrusion detection, and lack comprehensive, dynamic, and intelligent management of security and maintenance tasks. In particular, when it comes to the generation and execution of inspection tasks, they often lack in-depth analysis of the characteristics of the monitored area and dynamic adjustment mechanisms, resulting in irrational allocation of inspection tasks and low inspection efficiency. Summary of the Invention

[0004] The present invention aims to solve the technical problem in the prior art of insufficient efficiency and accuracy of security maintenance caused by unreasonable inspection task allocation, and provides an intelligent inspection management method and system for improving security maintenance efficiency.

[0005] The technical solution of the present invention to solve the above technical problems is as follows:

[0006] In a first aspect, the present invention provides an intelligent inspection and management method for improving security maintenance efficiency, comprising: acquiring digital image data of a target monitoring area through a binocular camera, and extracting meta-attributes of the digital image data, wherein the meta-attributes include regional function type codes, spatial position coordinates, and lighting condition parameters; calling a pre-trained regional feature analysis model, and determining a multi-dimensional security maintenance indicator weight distribution according to the regional function type codes, spatial position coordinates, and lighting condition parameters of the digital image data, wherein the regional feature analysis model is obtained by machine learning training using multiple groups of data, and any group of the multiple groups of data includes Meta-attribute record data, as well as labels that identify security maintenance indicator weight data and function type meta-attributes; based on the multi-dimensional security maintenance indicators, perform semantic segmentation on the digital image data to generate a set of sub-regions with topological connection relationships; based on the multi-dimensional security maintenance indicator weight distribution, perform weighted mean calculation on the multi-dimensional security maintenance indicator characteristic values ​​of each sub-region in the sub-region set to obtain a set of risk quantification values; map the risk quantification values ​​to an inspection cycle query table bound to the meta-attribute, output the corresponding inspection time interval parameters, generate an inspection task instruction set, and send it to the execution terminal device for intelligent inspection management.

[0007] In the second aspect, the present invention provides an intelligent inspection and management system for improving the efficiency of security maintenance, the system comprising: a data acquisition module for acquiring digital image data of a target monitoring area through a binocular camera, and extracting meta-attributes of the digital image data, wherein the meta-attributes include regional function type codes, spatial position coordinates and lighting condition parameters; a weight determination module for calling a pre-trained regional feature analysis model, and determining the weight distribution of multi-dimensional security maintenance indicators according to the regional function type codes, spatial position coordinates and lighting condition parameters of the digital image data, wherein the regional feature analysis model is obtained by machine learning training using multiple groups of data, and any group of the multiple groups of data includes meta-attributes The system comprises a plurality of image processing modules, a plurality of image processing modules and a plurality of image processing modules, and a plurality of image processing modules. The plurality of image processing modules are used to process the image processing data and the plurality of image processing modules respectively. The plurality of image processing modules are used to process the image processing data and the plurality of image processing modules respectively. The plurality of image processing modules are used to process the image processing data and the plurality of image processing modules respectively. The plurality of image processing modules are used to process the image processing data and the plurality of image processing modules respectively.

[0008] The beneficial effects of the present invention are: acquiring images of the monitored area and extracting meta-attributes through a binocular camera, using a pre-trained model to determine the weight distribution of multi-dimensional security and maintenance indicators, performing semantic segmentation on the image to generate a set of sub-areas, and then calculating the risk quantification value of each sub-area based on the weight distribution, and mapping it to the inspection cycle query table to generate an inspection task instruction set, thereby realizing intelligent management of inspection tasks and effectively improving the efficiency and accuracy of security and maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 A flow chart of an intelligent inspection management method for improving security maintenance efficiency provided by the present invention.

[0010] Figure 2 This is a structural diagram of an intelligent inspection and management system provided by the present invention for improving security and maintenance efficiency.

[0011] Explanation of the accompanying drawings: data acquisition module 11, weight determination module 12, semantic segmentation module 13, indicator calculation module 14, intelligent inspection module 15. DETAILED DESCRIPTION

[0012] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts shall fall within the scope of protection of the present invention.

[0013] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, a feature specified as "first" or "second" may explicitly or implicitly include one or more of the specified features. In the description of the present invention, "plurality" means two or more, unless otherwise specifically defined.

[0014] In the description of the present invention, the term "for example" is used to mean "used as an example, illustration or illustration". Any embodiment of the present invention described as "for example" is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is given to enable any person skilled in the art to implement and use the present invention. In the following description, details are listed for the purpose of explanation. It should be understood that a person of ordinary skill in the art can recognize that the present invention can be implemented without using these specific details. In other examples, well-known structures and processes are not elaborated in detail to avoid obscuring the description of the present invention with unnecessary details. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the widest scope consistent with the principles and features disclosed herein.

[0015] Example 1, as Figure 1 As shown, an embodiment of the present invention provides an intelligent inspection management method for improving security maintenance efficiency, including:

[0016] S10: Acquire digital image data of the target monitoring area through a binocular camera, and extract meta-attributes of the digital image data, wherein the meta-attributes include area function type code, spatial position coordinates, and lighting condition parameters.

[0017] For example, in the intelligent inspection management system of this solution, a binocular camera deployed in a target monitoring area captures digital image data of the area.

[0018] The target monitoring area refers to the specific area where the binocular camera is deployed and targets image data. This area is the subject of system monitoring. The binocular camera captures digital image data from this area to implement intelligent inspection and management functions, such as monitoring equipment status, personnel activities, and environmental conditions. For example, in a factory, the target monitoring area might be the production line; in a warehouse, it might be a specific storage area for important supplies.

[0019] The digital image data not only contains intuitive visual information of the monitoring scene, such as the appearance of equipment and personnel activities captured by a binocular camera in an inspection scene, but also forms digital image data after being converted and recorded in digital form. The system can use this data to determine whether the equipment is operating normally and whether personnel operations are in compliance. In addition, digital image data also contains rich meta-attributes. Meta-attributes refer to the more descriptive and characteristic attributes contained in the image data. These attributes can further explain the characteristics of the image data, such as regional functional type codes, spatial location coordinates, and lighting condition parameters. They provide basic information for subsequent intelligent analysis, helping the analysis system better understand and process image data, and providing a foundation for subsequent intelligent analysis.

[0020] Among the meta-attributes, the area function type code represents the functional characteristics of different monitoring areas, such as warehouses, offices, or public passages. These different functional areas may have significantly different security and maintenance focus indicators. For example, a warehouse area may focus more on fire prevention and theft prevention, while an office area may focus more on access control and privacy protection. The spatial location coordinates further refine the geographic location of the monitoring area. Different spatial locations face different security risks due to geographical factors (such as floor height and surrounding environment), which in turn influences the emphasis of security and maintenance indicators. For example, high floors may focus more on wind and earthquake resistance, while low floors or basements may require enhanced moisture and flood control measures. Furthermore, the lighting condition parameter, another important meta-attribute, reflects the changes in lighting conditions in the monitoring area during different time periods (such as daytime and nighttime). This change directly affects the clarity and recognition accuracy of the monitoring image, and thus influences the setting of security and maintenance indicators. For example, at night or in low-light environments, enhanced lighting or infrared monitoring technology may be required to ensure effective monitoring. By extracting these meta-attributes, we can more accurately understand the characteristics of the monitored area, providing strong support for subsequent steps such as determining the weight distribution of multi-dimensional security and maintenance indicators, semantic segmentation, and risk quantification, thereby significantly improving the targetedness and efficiency of security and maintenance.

[0021] S20: Calling a pre-trained regional feature analysis model to determine the weight distribution of multi-dimensional security and maintenance indicators based on the regional function type code, spatial position coordinates and lighting condition parameters of the digital image data, wherein the regional feature analysis model is obtained by machine learning training using multiple groups of data, and any group of the multiple groups of data includes meta-attribute record data, as well as labels identifying the security and maintenance indicator weight data and function type meta-attributes.

[0022] Furthermore, a pre-trained regional feature analysis model is invoked. This model, built using deep learning technology, can deeply understand and analyze the complex characteristics of the monitored area. This regional feature analysis model uses the regional function type code, spatial location coordinates, and lighting condition parameters of the digital image data as core inputs for comprehensive analysis. The regional function type code represents the specific function of the monitored area, such as a data center, parking lot, or public activity area. Different functional areas have different security and maintenance requirements, and the model adjusts the weights of corresponding security and maintenance indicators based on these codes. The spatial location coordinates precisely indicate the specific location of the monitored area in physical space. Different geographical locations may face different natural and man-made risks. For example, coastal areas may require special attention to typhoon prevention measures, while inland cities may focus more on fire prevention and theft prevention. The model further refines the weight distribution based on these factors. The lighting condition parameters cover the changes in light from sunrise to sunset, and even under special weather conditions. These changes directly affect the image quality of the surveillance camera and, in turn, the effectiveness of the security system. The model takes these factors into account, appropriately increasing the weight of features such as infrared monitoring or automatic dimming in conditions such as insufficient light or direct sunlight.

[0023] Specifically, the regional feature analysis model is a deep learning model based on machine learning. It is used to predict the weight distribution and functional type of security and maintenance indicators in a target area based on the metadata recorded in the target area. The regional feature analysis model uses iterative optimization of multiple sets of training data to intelligently analyze the characteristics of different areas, providing a quantitative basis for the dynamic allocation of subsequent inspection tasks. The model utilizes a multi-layer neural network structure, consisting of an input layer, a feature extraction layer, a classification layer, and a regression layer. The input layer receives metadata recorded data. The feature extraction layer extracts spatial and temporal features using a convolutional neural network (CNN) and a long short-term memory network (LSTM). The classification layer outputs the functional type prediction results, and the regression layer outputs the weight values ​​of each security and maintenance indicator.

[0024] Specifically, the input layer receives meta-attribute record data, including multi-dimensional feature vectors encoding region function types, spatial location coordinates, and lighting condition parameters. The input dimension is N×M, where N is the number of samples and M is the dimension of the meta-attribute features. The convolutional neural network (CNN) in the feature extraction layer consists of three convolutional layers, each followed by a batch normalization layer and a Reluctant Unit (ReLU) activation function. The convolution kernel sizes are 3×3, 5×5, and 7×7, respectively, to capture spatial features at different scales. The pooling layer uses max pooling with a stride of 2 to reduce feature dimensionality. The long short-term memory (LSTM) network processes temporal features in the meta-attributes, such as the time series of lighting conditions. The hidden layer has a dimension of 128 and outputs a temporal feature vector. The feature fusion layer concatenates the features extracted by the CNN and LSTM, and reduces the dimensionality to 256 through a fully connected layer. The classification layer consists of two fully connected layers with a Reluctant Unit (ReLU) activation function. The output layer uses a softmax activation function and outputs a K-dimensional vector, corresponding to the probability distribution of K region function types. The regression layer consists of two fully connected layers, with the activation function being ReLU. The output layer uses a linear activation function to output an L-dimensional vector, corresponding to the weight values ​​of the L security and maintenance indicators.

[0025] Preferably, for the model's input meta-attribute record data, the regional functional type is encoded using one-hot encoding. For example, if K = 5 functional types (office area, warehouse, corridor, computer room, public area), the encoding is a 5-dimensional vector. Spatial location coordinates include latitude and longitude, floor height, and surrounding environmental indicators (such as distance to fire protection facilities and pedestrian density), totaling P dimensions. Lighting condition parameters include average illumination, light stability, and diurnal variation curves. Frequency domain features are extracted through Fourier transform, totaling Q dimensions. The input data format is an [N, M] tensor, where M = 5 + P + Q. The output data functional type prediction is a K-dimensional probability vector. For example, [0.1, 0.05, 0.7, 0.05, 0.1] indicates that the area has the highest probability of being a "corridor." The security maintenance index weight is output as an L-dimensional real-valued vector, such as [0.25, 0.15, 0.4, 0.2], which represents the weights of the four indicators of intrusion detection, equipment failure, environmental monitoring, and personnel management.

[0026] During the training process, multiple training data sets each include: meta-attribute records (input features), function type labels (ground truth for the classification task), and security and maintenance indicator weight labels (ground truth for the regression task). Meta-attribute records are collected in real time via binocular cameras and sensor networks. Function type labels are manually annotated or determined through historical records. Security and maintenance indicator weight labels are derived from historical alarm event frequency statistics. For example, if intrusion incidents in a certain area accounted for 30% of the total incidents over the past year, the intrusion detection weight is assigned to 0.3. A multi-task learning loss function is designed, combining weighted classification and regression losses, and optimized using an optimization algorithm. Data preprocessing is performed, and meta-attribute features are normalized to a mean of 0 and a standard deviation of 1. Data augmentation, including rotation, scaling, and noise addition, is performed on the training data to improve model generalization. The batch size for batch training is set to 32, and all training data are traversed in each epoch. The loss function is calculated for each batch, and the model parameters are updated through backpropagation. After each epoch, model performance is evaluated on a validation set. The model parameters corresponding to the best validation set performance are recorded. If the validation set loss does not improve after 20 consecutive epochs, training is terminated early to prevent overfitting. Five models with different initializations are trained and fused using a voting algorithm. The function type prediction uses the category with the highest occurrence across the five models. The security and maintenance indicator weights are the average of the five model predictions.

[0027] By integrating the above content, the regional feature analysis model can dynamically determine the weight distribution of multi-dimensional security and maintenance indicators, ensuring that inspection tasks can be customized to the specific needs of different monitoring areas, thereby significantly improving the efficiency and effectiveness of security and maintenance and achieving optimal allocation and utilization of resources.

[0028] S30: Based on the multi-dimensional security and maintenance indicators, perform semantic segmentation on the digital image data to generate a set of sub-regions with a topological connection relationship.

[0029] Preferably, a semantic segmentation operation is performed on the acquired digital image data based on the determined multi-dimensional security and maintenance indicators. This process involves analyzing each pixel in the image and dividing the image into several sub-areas with clear security significance based on the degree of correlation between the pixel and the preset security and maintenance indicators.

[0030] The multi-dimensional security and maintenance indicators refer to standards for measuring security and maintenance conditions from multiple aspects, such as personnel flow density, equipment distribution density, potential safety hazards, maintenance cycles of monitoring equipment, safety risk levels in different areas, and other security and maintenance requirements. Multi-dimensional security and maintenance indicators are obtained in advance based on the target area. Digital image data refers to image information stored in digital form, such as image data captured by surveillance cameras. The semantic segmentation process aims to classify each pixel in the image into a specific category, such as classifying pixels in a surveillance image into different categories such as people, vehicles, and buildings, so as to more clearly understand and analyze the image content. In the security field, it can be used to identify abnormal situations, monitor equipment status, etc.

[0031] Categorizing pixels based on their close relationship with established multi-dimensional security and maintenance indicators can be understood as follows: for example, if a pixel is closely associated with entrance and exit security indicators, it might be assigned to the category representing the entrance and exit area; if it is closely associated with equipment room maintenance indicators, it might be assigned to the equipment room category. This classification of all pixels in the image allows the entire monitored area to be broken down into distinct sub-areas with clear security significance. These sub-areas not only encompass specific physical spaces like entrances and exits, corridors, and equipment rooms, but can also be further subdivided into high-risk, medium-risk, and low-risk areas based on security requirements.

[0032] During semantic segmentation, it's crucial to preserve the topological connectivity between subregions, ensuring the continuity of adjacent subregions in image space. This characteristic is crucial for subsequent inspection route planning and risk assessment. For example, in a surveillance image of a large warehouse, the system might identify the storage area, loading and unloading area, and office area as distinct subregions, and clearly define the spatial layout relationships between them—for example, the loading and unloading area is adjacent to the storage area, while the office area is located to one side of the warehouse.

[0033] Through this semantic segmentation based on multi-dimensional security and maintenance indicators, a clearly structured and hierarchical set of sub-areas can be generated, providing a basis for subsequent inspection task allocation and risk quantification, significantly improving the accuracy and efficiency of security and maintenance work.

[0034] S40: Based on the weight distribution of the multi-dimensional security and maintenance indicator, a weighted mean calculation is performed on the multi-dimensional security and maintenance indicator characteristic value of each sub-area in the sub-area set to obtain a risk quantification value set.

[0035] In detail, the set of sub-regions obtained through semantic segmentation is analyzed in depth based on the weight distribution of the determined multi-dimensional security and maintenance indicators. For each sub-region, the corresponding multi-dimensional security and maintenance indicator characteristic values ​​are extracted and normalized to ensure comparability between different indicators. These multi-dimensional security and maintenance indicator characteristic values ​​cover data from multiple dimensions, such as intrusion detection sensitivity, fire warning thresholds, and equipment failure rates. Together, they form the basis for evaluating the security status of the sub-region.

[0036] Next, a weighted mean is calculated for the normalized feature values ​​using a predefined weight distribution. For example, if a sub-region is identified as a high-risk area, its intrusion detection sensitivity weight may be increased accordingly, resulting in a larger proportion in the weighted mean calculation.

[0037] This process comprehensively assesses each sub-area's performance across various security and maintenance indicators, quantifies its overall risk level, and ultimately generates a set of risk quantification values. This set of risk quantification values ​​not only provides data support for subsequent adjustments to inspection cycles but also makes the allocation of inspection resources more scientific and rational, ensuring that high-risk areas receive more frequent attention and maintenance, thereby effectively improving the efficiency and effectiveness of security and maintenance work.

[0038] S50: Mapping the risk quantification value to the inspection period query table bound to the meta-attribute, outputting the corresponding inspection time interval parameter, generating an inspection task instruction set, and sending it to the execution terminal device for intelligent inspection management.

[0039] Specifically, the calculated risk quantification value is mapped to an inspection cycle lookup table tied to the monitored area's meta-attributes. These meta-attributes encompass key information such as the area's functional type code, spatial location coordinates, and lighting parameters, which together determine the unique security requirements and risk characteristics of each monitored area.

[0040] As a preset database, the inspection cycle query table predefines the inspection time interval parameters that should be used for different risk levels and regional characteristics based on a comprehensive consideration of risk quantification values ​​and meta-attributes. For example, for high-risk and critical monitoring areas, the inspection interval may be set to a shorter value even during daytime with good lighting conditions to ensure the immediate response of security measures; while for low-risk and relatively unimportant areas, the inspection cycle may be extended accordingly. Through this mapping process, the inspection time interval parameters that match the current risk status of each monitoring area can be automatically output, and then a set of inspection task instructions can be generated. The inspection task instruction set not only clarifies the inspection sequence and time nodes for each area, but may also include specific inspection content and requirements to ensure that each inspection is effective and complete.

[0041] Finally, the inspection task instruction set is quickly sent to the execution terminal device, such as an intelligent inspection robot or a handheld terminal, to realize the automation and intelligent management of the inspection work, significantly improving the response speed and accuracy of security maintenance, effectively reducing human omissions and resource waste, and providing strong technical support for the security of the monitored area.

[0042] In a preferred embodiment, a pre-trained regional feature analysis model is called to determine the weight distribution of multi-dimensional security and maintenance indicators according to the regional function type code, spatial position coordinates and lighting condition parameters of the digital image data, including:

[0043] A training sample library is constructed, wherein the training sample library includes a historical monitoring image data set annotated with multi-dimensional security and maintenance indicator weight true value labels, and the historical monitoring image data set has a meta-attribute record value set.

[0044] Based on the cross entropy loss term and the mean square error loss term, a weight distribution loss function is constructed, wherein the cross entropy loss term is used to statistically predict the cross entropy loss of the regional functional type and the functional type meta-attribute record value, and the mean square error loss term is used to calculate the mean square error loss between the true value of the multi-dimensional security and maintenance indicator weight and the predicted value of the multi-dimensional security and maintenance indicator weight.

[0045] Through the weight distribution loss function, based on the training sample library, with the meta-attribute record value set as input, and the multi-dimensional security and maintenance indicator weight true value label and function type meta-attribute record value as output supervision, the weight distribution matrix is ​​updated through the gradient descent algorithm until the accuracy of the model on the verification set reaches a predetermined threshold.

[0046] Optionally, to ensure that the regional feature analysis model can accurately determine the weight distribution of multi-dimensional security and maintenance indicators, a high-quality training sample library must be constructed. This training sample library compiles a large dataset of historical surveillance images. Each image is annotated with the true value label of the multi-dimensional security and maintenance indicator weights and accompanied by a detailed set of meta-attribute records covering key information such as the regional functional type code, spatial location coordinates, and lighting condition parameters. The function type meta-attribute records serve as a key constraint, ensuring that the model accurately captures the functional characteristics of each region during training and preventing output from deviating from reality.

[0047] Subsequently, a weight distribution loss function was constructed based on the cross-entropy loss term and the mean squared error loss term. The cross-entropy loss term focuses on measuring the difference between the model's predicted regional functional type and the recorded value of the functional type's meta-attribute, effectively preventing deviations or "hallucinations" in the model's functional type predictions. The mean squared error loss term focuses on the error between the true and predicted values ​​of the multi-dimensional security and maintenance indicator weights, ensuring the accuracy of the weight distribution.

[0048] During the training phase, a weight distribution loss function is used, with the meta-attribute record value set as the model input and the true value labels of the multi-dimensional security and maintenance indicator weights and the function type meta-attribute record values ​​as the output supervision. The weight distribution matrix is ​​iteratively optimized using a gradient descent algorithm. By continuously adjusting the model parameters and gradually reducing the loss function value, the model's accuracy on the validation set reaches a predetermined threshold, indicating that the model has the ability to accurately predict the weight distribution of multi-dimensional security and maintenance indicators based on the input meta-attributes.

[0049] The application of regional feature analysis models not only improves the intelligence level of security and maintenance work, but also ensures the rational allocation and efficient execution of inspection tasks, providing technical support for the security of the monitored area.

[0050] In a preferred embodiment, constructing a training sample library includes:

[0051] First historical monitoring image data is collected, wherein the first historical monitoring image data has a first meta-attribute record value set.

[0052] Retrieve security and maintenance alarm event samples that meet the first meta-attribute record value set.

[0053] The event frequency proportions of the multi-dimensional security and maintenance indicators in the security and maintenance alarm event samples are counted respectively, and are set as the weighted true value labels of the multi-dimensional security and maintenance indicators.

[0054] In detail, in the process of constructing a training sample library for regional feature analysis model training, the first historical monitoring image data is collected first. These data sets are accompanied by a first meta-attribute record value set, which covers key information such as regional functional type coding, spatial location coordinates and lighting condition parameters, providing a basic basis for subsequent sample screening and weight setting.

[0055] Based on the collected first meta-attribute record value set, precise retrieval is performed in the security and maintenance event database to screen out security and maintenance alarm event samples that match these meta-attribute record value sets, thereby ensuring a high degree of correlation between the samples and the actual monitoring scenarios, and improving the pertinence and practicality of the training sample library.

[0056] Subsequently, for the selected security and maintenance alarm event samples, the frequency of occurrence of multi-dimensional security and maintenance indicators in each type of event is calculated. These indicators include but are not limited to the number of intrusion detection triggers, fire warning response speed, and equipment failure frequency. By calculating the frequency ratio of each indicator in the event, it is set as the weighted true value label of the multi-dimensional security and maintenance indicator. For example, if a region has frequently experienced fire warning events in the past year, the weighted true value label of the fire warning-related indicators will be set to a higher value to reflect the special fire prevention needs of that area.

[0057] Through this process, the constructed training sample library not only contains rich surveillance image data, but also incorporates multi-dimensional indicator weight information in actual security and maintenance events, providing high-quality, labeled data support for subsequent model training, and effectively improving the application effect and prediction accuracy of the regional feature analysis model in the field of security and maintenance.

[0058] In a preferred embodiment, semantic segmentation is performed on the digital image data based on multi-dimensional security and maintenance indicators to generate a set of sub-regions with topological connections, including:

[0059] Based on the multi-dimensional security and maintenance index, feature normalization extraction is performed on the digital image data to obtain the multi-dimensional security and maintenance index feature value at the first pixel position to the multi-dimensional security and maintenance index feature value at the Q-th pixel position.

[0060] Based on the Euclidean distance threshold, adjacent pixel Euclidean distance clustering is performed on the multi-dimensional security and maintenance index characteristic value at the first pixel position to the multi-dimensional security and maintenance index characteristic value at the Q-th pixel position to obtain a digital image region segmentation result.

[0061] Based on the digital image region segmentation result, the sub-region set having a topological connection relationship is constructed.

[0062] Furthermore, in the image processing phase, semantic segmentation is performed on the digital image data based on multi-dimensional security and maintenance indicators to generate a set of sub-regions with topological connections. The specific process is as follows:

[0063] First, the digital image data is subjected to feature normalization and extraction based on multi-dimensional security and maintenance indicators. In this step, the corresponding multi-dimensional security and maintenance indicator eigenvalues ​​are calculated and extracted for each pixel position in the image, from the first pixel position to the Qth pixel position. These eigenvalues ​​may cover multiple dimensions closely related to security and maintenance, such as light intensity, texture complexity, and target object density. Normalization ensures that different eigenvalues ​​are compared and analyzed on the same scale, providing a reliable data foundation for subsequent clustering and segmentation.

[0064] Next, the Euclidean distance threshold is used to perform Euclidean distance clustering on the extracted multi-dimensional security and maintenance indicator feature values ​​from the first pixel position to the Qth pixel position. Euclidean distance is used to measure the similarity between two pixels in a multidimensional feature space. By setting a reasonable Euclidean distance threshold, the system can cluster adjacent pixels with similar feature values ​​into one category, thereby achieving regional segmentation of the digital image. For example, when monitoring a warehouse scene, if certain adjacent pixels have similar feature values ​​such as light intensity and texture complexity, they will be clustered into the same area, which may represent the warehouse's cargo storage area or aisle area.

[0065] Finally, based on the digital image region segmentation results, a topologically connected set of subregions is constructed. This topological connection reflects the spatial layout and adjacency between subregions, which is crucial for subsequent tasks such as inspection route planning and risk assessment. By constructing such a set of subregions, we can more clearly understand the distribution of security needs within the monitored area, providing strong support for intelligent inspection management and significantly improving inspection efficiency and the accuracy of security maintenance.

[0066] In a preferred embodiment, the steps of constructing the inspection cycle query table include:

[0067] A historical maintenance record data set is collected, wherein the historical maintenance record data set includes a risk quantification value field and collection area meta-attributes.

[0068] Construct the inspection interval optimization equation:

[0069] ,

[0070] in, Characterize the fitness of the inspection plan, Represents the fixed cost of a single inspection, Represents the cumulative risk cost per unit time and is positively correlated with the risk quantification value field. Characterizes the rated minimum inspection cycle, represents the inspection period of the i-th area, and n represents the number of areas.

[0071] Based on the inspection interval optimization equation, combined with the risk quantification value field and the collection area meta-attribute, the inspection interval fitness minimum optimization is performed to obtain the target inspection interval field.

[0072] When the target patrol interval field is inconsistent with the actual patrol interval field, the actual patrol interval field is updated using the target patrol interval field; otherwise, no update is performed.

[0073] The inspection cycle query table is constructed based on the optimized historical maintenance record data set.

[0074] Specifically, constructing the inspection cycle query table is a process based on historical data and optimization algorithms. First, a historical maintenance record dataset is collected. This dataset not only includes risk quantification fields, which measure the security risk level of each area at different points in time, but also includes regional meta-attributes such as area function type and spatial location. These meta-attributes influence the focus and requirements of security maintenance.

[0075] In order to scientifically construct an inspection cycle optimization plan, the formula is set: ;in, Characterizes the fitness of the inspection plan. The smaller the value, the better the comprehensive performance of the inspection plan in terms of cost and efficiency. Represents the fixed cost of a single inspection, covering basic expenses such as manpower and equipment; Represents the cumulative risk cost per unit time and is positively correlated with the risk quantification value field. That is, the higher the risk, the greater the potential loss per unit time. Represents the rated minimum inspection cycle, which is determined by the characteristics of the area. For example, high-risk areas require more frequent inspections. The key variable to be optimized is the inspection cycle for region i, and n represents the number of regions. This formula aims to balance inspection costs with risk prevention and control. By minimizing the f value, we seek the most economical inspection cycle while ensuring safety.

[0076] Then, based on the inspection interval optimization equation, combined with risk quantification and regional meta-attributes, the minimum inspection interval fitness is optimized to find the target inspection interval field that optimizes the overall inspection plan. If the target inspection interval differs from the actual inspection interval, the former is used to update the latter, ensuring that the inspection plan remains close to the optimal state. Finally, based on the optimized historical maintenance record dataset, an inspection cycle query table is constructed, providing accurate and efficient decision-making basis for intelligent inspection management, significantly improving the response speed and resource utilization efficiency of security maintenance.

[0077] In a preferred embodiment, based on the inspection interval optimization equation, combined with the risk quantification value field and the collection area meta-attribute, the inspection interval fitness minimum optimization is performed to obtain the target inspection interval field, including:

[0078] Parameter configuration is performed based on the risk quantification value field and the collection area meta-attribute to obtain a fixed cost of a single inspection of the collection area and a cumulative risk cost per unit time.

[0079] Configure the rated minimum inspection period through the user end.

[0080] Based on the fixed cost of a single inspection of the collection area, the cumulative risk cost per unit time, and the rated minimum inspection period, a minimum fitness search is performed to obtain an initial inspection interval.

[0081] The device abnormality rate that satisfies the risk quantification value field, the collection area meta-attribute and the initial inspection interval is retrieved.

[0082] When the device abnormality rate is greater than the abnormality rate threshold, the initial inspection interval is shortened by 2 times and the cycle is performed.

[0083] When the device abnormality rate is less than the abnormality rate threshold, the initial inspection interval is increased by 2 times and a cycle is performed.

[0084] When the device abnormality rate is equal to the abnormality rate threshold, the initial inspection interval is set to the target inspection interval field.

[0085] For example, to determine the optimal inspection interval, a series of operations are performed based on the inspection interval optimization equation, combining risk quantification fields and collection area meta-attributes. Specifically, parameter configuration is performed based on risk quantification fields, such as the area's risk level and historical failure frequency, as well as collection area meta-attributes such as the area's functional type (production area, storage area, etc.) and equipment density. This results in the fixed cost of a single inspection in the collection area (covering basic expenses such as manpower and testing equipment) and the cumulative risk cost per unit time (which is positively correlated with the risk quantification value; higher risk means greater potential losses per unit time). Next, the user configures a rated minimum inspection cycle based on area characteristics and safety regulations to ensure that the inspection frequency does not fall below the safety limit. Based on these parameters, a minimum fitness search is performed to find the initial inspection interval that optimizes the inspection plan's overall cost and risk prevention and control effectiveness.

[0086] Next, the device anomaly rate is retrieved based on the current risk quantification value field, the collection area meta-attributes, and the initial inspection interval. If the device anomaly rate exceeds the preset anomaly rate threshold, it indicates that the current inspection interval may result in equipment hazards not being discovered in a timely manner. Therefore, the initial inspection interval is reduced by a factor of 2 and the loop is executed to increase the inspection frequency. If the device anomaly rate is less than the anomaly rate threshold, it means that the inspection interval may be too conservative and resources are not fully utilized. In this case, the initial inspection interval is increased by a factor of 2 and the loop is executed to optimize resource allocation. When the device anomaly rate is exactly equal to the anomaly rate threshold, it indicates that the current inspection interval has achieved the optimal balance between cost and risk prevention and control. Therefore, the initial inspection interval is set as the target inspection interval field.

[0087] This series of operations significantly improves the accuracy and efficiency of inspection management by dynamically adjusting inspection intervals, effectively reducing the risk of equipment failure and operation and maintenance costs.

[0088] In a preferred embodiment, parameter configuration is performed based on the risk quantification value field and the collection area meta-attribute to obtain the fixed cost of a single inspection of the collection area and the cumulative risk cost per unit time, including:

[0089] The product of the risk quantification value field and the accumulated time is calculated and set as the accumulated risk cost per unit time.

[0090] Calculate the path distance between the collection area and the inspection starting area, and set it as the fixed cost of a single inspection in the collection area.

[0091] Specifically, to accurately determine the inspection cost parameters for the collection area, calculations must be performed based on the risk quantification value field and the collection area's meta-attributes. First, to calculate the cumulative risk cost per unit time, the risk quantification value field is multiplied by the accumulated time. The risk quantification value field covers indicators such as the regional risk level and potential safety hazards, while the accumulated time reflects the time span within the inspection cycle. For example, if a region has a high risk quantification value and a long inspection cycle, the region's accumulated risk cost per unit time will increase significantly. This calculation method ensures that high-risk areas receive more attention in the inspection plan.

[0092] The fixed cost of a single inspection in a collection area is calculated by measuring the path distance between the collection area and the inspection starting point. Path distance not only affects the inspector's travel time but also directly impacts transportation, labor, and other costs during the inspection process. For example, if the collection area is far from the inspection starting point, inspectors will need to spend more time and resources to reach it, increasing the fixed cost of a single inspection.

[0093] Through the above calculation method, the fixed cost of a single inspection and the cumulative risk cost per unit time in the collection area can be accurately obtained, providing a data basis for inspection interval optimization and resource allocation, and significantly improving the scientificity and economy of inspection management.

[0094] The embodiments of the present invention provide an intelligent inspection and management method for improving security and maintenance efficiency, which has at least the following technical effects:

[0095] 1. By determining the weight distribution of multi-dimensional security and maintenance indicators and calculating the risk quantification value based on them, and comprehensively considering multiple factors such as regional functions, spatial location, and lighting conditions, the risks of each sub-area can be accurately quantified, and then inspection time intervals that are more in line with actual needs can be formulated, thereby improving the scientific nature and accuracy of inspection planning.

[0096] 2. Using the inspection interval optimization equation and a series of optimization steps, the inspection interval is dynamically adjusted according to the equipment abnormality rate, achieving intelligent self-adaptation of the inspection cycle. While ensuring the normal operation of the equipment, it effectively avoids excessive or insufficient inspections, thereby improving inspection efficiency and resource utilization.

[0097] 3. Mapping the risk quantification value to the inspection cycle query table to generate the inspection task instruction set and send it to the execution terminal device, realizing the automatic generation and intelligent management of inspection tasks, reducing manual intervention, and improving the overall efficiency of security maintenance management.

[0098] Example 2, as Figure 2 As shown, based on the same inventive concept as the intelligent inspection management method for improving security maintenance efficiency provided in the first embodiment, the embodiment of the present invention further provides an intelligent inspection management system for improving security maintenance efficiency, the system comprising:

[0099] The data acquisition module 11 is used to acquire digital image data of the target monitoring area through a binocular camera and extract meta-attributes of the digital image data, wherein the meta-attributes include regional function type code, spatial position coordinates and lighting condition parameters.

[0100] The weight determination module 12 is used to call a pre-trained regional feature analysis model to determine the weight distribution of multi-dimensional security and maintenance indicators based on the regional function type code, spatial position coordinates and lighting condition parameters of the digital image data, wherein the regional feature analysis model is obtained by machine learning training through multiple groups of data, and any group of the multiple groups of data includes meta-attribute record data and labels identifying the security and maintenance indicator weight data and function type meta-attributes.

[0101] The semantic segmentation module 13 is used to perform semantic segmentation on the digital image data based on multi-dimensional security and maintenance indicators to generate a set of sub-regions with a topological connection relationship.

[0102] The indicator calculation module 14 is used to calculate the weighted mean of the multi-dimensional security and maintenance indicator characteristic value of each sub-area in the sub-area set based on the multi-dimensional security and maintenance indicator weight distribution to obtain a risk quantification value set.

[0103] The intelligent inspection module 15 is used to map the risk quantification value to the inspection period query table bound to the meta-attribute, output the corresponding inspection time interval parameters, generate an inspection task instruction set, and send it to the execution terminal device for intelligent inspection management.

[0104] Furthermore, the weight determination module 12 is further configured to perform the following steps:

[0105] A training sample library is constructed, wherein the training sample library includes a historical monitoring image data set annotated with a multi-dimensional security and maintenance indicator weight true value label, and the historical monitoring image data set has a meta-attribute record value set; based on a cross-entropy loss term and a mean square error loss term, a weight distribution loss function is constructed, wherein the cross-entropy loss term is used to statistically predict the cross-entropy loss of the regional functional type and the functional type meta-attribute record value, and the mean square error loss term is used to calculate the mean square error loss between the multi-dimensional security and maintenance indicator weight true value and the multi-dimensional security and maintenance indicator weight predicted value; through the weight distribution loss function, based on the training sample library, with the meta-attribute record value set as input, and the multi-dimensional security and maintenance indicator weight true value label and the functional type meta-attribute record value as output supervision, the weight distribution matrix is ​​updated by a gradient descent algorithm until the accuracy of the model on the validation set reaches a predetermined threshold.

[0106] Furthermore, the weight determination module 12 is further configured to perform the following steps:

[0107] Collect first historical monitoring image data, wherein the first historical monitoring image data has a first meta-attribute record value set; retrieve security maintenance alarm event samples that meet the first meta-attribute record value set; and count the event frequency proportions of the multi-dimensional security maintenance indicators in the security maintenance alarm event samples respectively, and set them as the weighted true value labels of the multi-dimensional security maintenance indicators.

[0108] Furthermore, the semantic segmentation module 13 is further configured to perform the following steps:

[0109] Based on the multidimensional security and maintenance index, feature normalization extraction is performed on the digital image data to obtain the multidimensional security and maintenance index characteristic value of the first pixel position to the multidimensional security and maintenance index characteristic value of the Q-th pixel position; based on the Euclidean distance threshold, adjacent pixel Euclidean distance clustering is performed on the multidimensional security and maintenance index characteristic value of the first pixel position to the multidimensional security and maintenance index characteristic value of the Q-th pixel position to obtain a digital image region segmentation result; based on the digital image region segmentation result, the sub-region set with a topological connection relationship is constructed.

[0110] Furthermore, the intelligent inspection module 15 is further configured to perform the following steps:

[0111] Collect a historical maintenance record data set, wherein the historical maintenance record data set includes a risk quantification value field and collection area meta-attributes; construct an inspection interval optimization equation: ,in, Characterize the fitness of the inspection plan, Represents the fixed cost of a single inspection, Represents the cumulative risk cost per unit time and is positively correlated with the risk quantification value field. Characterizes the rated minimum inspection cycle, represents the inspection cycle of the i-th area, and n represents the number of areas; based on the inspection interval optimization equation, combined with the risk quantification value field and the collection area meta-attribute, the inspection interval fitness minimum optimization is performed to obtain the target inspection interval field; when the target inspection interval field is inconsistent with the actual inspection interval field, the target inspection interval field is used to update the actual inspection interval field, otherwise, the update is not performed; based on the historical maintenance record data set after optimization, the inspection cycle query table is constructed.

[0112] Furthermore, the intelligent inspection module 15 is further configured to perform the following steps:

[0113] Parameter configuration is performed based on the risk quantification value field and the collection area meta-attribute to obtain the fixed cost of a single inspection of the collection area and the cumulative risk cost per unit time; the rated minimum inspection period is configured through the user terminal; based on the fixed cost of a single inspection of the collection area, the cumulative risk cost per unit time and the rated minimum inspection period, minimum fitness search is performed to obtain an initial inspection interval; the device abnormality rate that meets the risk quantification value field, the collection area meta-attribute and the initial inspection interval is retrieved; when the device abnormality rate is greater than the abnormality rate threshold, the initial inspection interval is reduced by 2 times and a loop is executed; when the device abnormality rate is less than the abnormality rate threshold, the initial inspection interval is increased by 2 times and a loop is executed; when the device abnormality rate is equal to the abnormality rate threshold, the initial inspection interval is set to the target inspection interval field.

[0114] Furthermore, the intelligent inspection module 15 is further configured to perform the following steps:

[0115] The product of the risk quantification value field and the accumulated time is calculated and set as the accumulated risk cost per unit time; the path distance between the collection area and the inspection starting area is calculated and set as the fixed cost of a single inspection of the collection area.

[0116] Through the above detailed description of an intelligent inspection and management method for improving security and maintenance efficiency in this specification, those skilled in the art can clearly understand an intelligent inspection and management system for improving security and maintenance efficiency in this embodiment. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0117] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An intelligent inspection and management method for improving security maintenance efficiency, characterized in that: include: Acquire digital image data of the target monitoring area through a binocular camera, and extract meta-attributes of the digital image data, wherein the meta-attributes include regional function type code, spatial position coordinates, and lighting condition parameters; Calling a pre-trained regional feature analysis model to determine a multi-dimensional security and maintenance indicator weight distribution based on the regional function type code, spatial position coordinates, and lighting condition parameters of the digital image data, wherein the regional feature analysis model is obtained by machine learning training using multiple sets of data, and any set of the multiple sets of data includes meta-attribute record data, and a label identifying the security and maintenance indicator weight data and the function type meta-attribute, and the function type meta-attribute represents the regional function type code in the meta-attribute record value set; Based on multi-dimensional security and maintenance indicators, performing semantic segmentation on the digital image data to generate a set of sub-regions with topological connection relationships; Based on the weight distribution of the multi-dimensional security and maintenance indicators, a weighted mean calculation is performed on the multi-dimensional security and maintenance indicator characteristic value of each sub-area in the sub-area set to obtain a risk quantification value set; The risk quantification value is mapped to the inspection period query table bound to the meta-attribute, the corresponding inspection time interval parameter is output, an inspection task instruction set is generated, and sent to the execution terminal device for intelligent inspection management.

2. The intelligent inspection and management method for improving security and maintenance efficiency according to claim 1 is characterized in that: The pre-trained regional feature analysis model is called to determine the weight distribution of multi-dimensional security and maintenance indicators according to the regional function type code, spatial position coordinates and lighting condition parameters of the digital image data, including: Constructing a training sample library, wherein the training sample library includes a historical monitoring image dataset annotated with multi-dimensional security and maintenance indicator weight true value labels, and the historical monitoring image dataset has a meta-attribute record value set; Based on the cross entropy loss term and the mean square error loss term, a weight distribution loss function is constructed, wherein the cross entropy loss term is used to statistically predict the cross entropy loss of the regional function type and the function type meta-attribute record value, and the mean square error loss term is used to calculate the mean square error loss between the true value of the multi-dimensional security and maintenance indicator weight and the predicted value of the multi-dimensional security and maintenance indicator weight; Through the weight distribution loss function, based on the training sample library, with the meta-attribute record value set as input, and the multi-dimensional security and maintenance indicator weight true value label and function type meta-attribute record value as output supervision, the weight distribution matrix is ​​updated through the gradient descent algorithm until the accuracy of the model on the verification set reaches a predetermined threshold.

3. The intelligent inspection and management method for improving security and maintenance efficiency according to claim 2 is characterized in that: Build a training sample library, including: Collecting first historical monitoring image data, wherein the first historical monitoring image data has a first meta-attribute record value set; Retrieving security maintenance alarm event samples that meet the first meta-attribute record value set; The event frequency proportions of the multi-dimensional security and maintenance indicators in the security and maintenance alarm event samples are counted respectively, and are set as the weighted true value labels of the multi-dimensional security and maintenance indicators.

4. The intelligent inspection and management method for improving security and maintenance efficiency according to claim 1 is characterized in that: Based on the multi-dimensional security and maintenance indicators, semantic segmentation is performed on the digital image data to generate a set of sub-regions with topological connection relationships, including: Based on the multi-dimensional security and maintenance index, perform feature normalization extraction on the digital image data to obtain the multi-dimensional security and maintenance index characteristic value at the first pixel position to the multi-dimensional security and maintenance index characteristic value at the Q-th pixel position; Based on the Euclidean distance threshold, performing Euclidean distance clustering of adjacent pixels on the multi-dimensional security and maintenance index characteristic values ​​at the first pixel position to the multi-dimensional security and maintenance index characteristic values ​​at the Q-th pixel position to obtain a digital image region segmentation result; Based on the digital image region segmentation result, the sub-region set having a topological connection relationship is constructed.

5. The intelligent inspection and management method for improving security and maintenance efficiency according to claim 1 is characterized in that: The steps to construct the inspection cycle query table include: Collecting a historical maintenance record data set, wherein the historical maintenance record data set includes a risk quantification value field and a collection area meta-attribute; Construct the inspection interval optimization equation: ， in, Characterize the fitness of the inspection plan, Represents the fixed cost of a single inspection, Represents the cumulative risk cost per unit time and is positively correlated with the risk quantification value field. Characterizes the rated minimum inspection cycle, represents the inspection cycle of the i-th area, and n represents the number of areas; Based on the inspection interval optimization equation, combined with the risk quantification value field and the collection area meta-attribute, the inspection interval fitness minimum optimization is performed to obtain the target inspection interval field; When the target inspection interval field is inconsistent with the actual inspection interval field, the actual inspection interval field is updated using the target inspection interval field; otherwise, no update is performed; The inspection cycle query table is constructed based on the optimized historical maintenance record data set.

6. The intelligent inspection and management method for improving security and maintenance efficiency according to claim 5 is characterized in that: Based on the inspection interval optimization equation, combined with the risk quantification value field and the collection area meta-attribute, the inspection interval fitness minimum optimization is performed to obtain the target inspection interval field, including: Parameter configuration is performed based on the risk quantification value field and the collection area meta-attribute to obtain the fixed cost of a single inspection of the collection area and the cumulative risk cost per unit time; Configure the rated minimum inspection cycle through the user terminal; Based on the fixed cost of a single inspection of the collection area, the cumulative risk cost per unit time, and the rated minimum inspection period, performing minimum fitness optimization to obtain an initial inspection interval; Retrieve the device abnormality rate that satisfies the risk quantification value field, the collection area meta-attribute, and the initial inspection interval; When the device abnormality rate is greater than the abnormality rate threshold, the initial inspection interval is shortened by 2 times and the cycle is performed; When the device abnormality rate is less than the abnormality rate threshold, the initial inspection interval is increased by 2 times and the cycle is performed; When the device abnormality rate is equal to the abnormality rate threshold, the initial inspection interval is set to the target inspection interval field.

7. The intelligent inspection and management method for improving security and maintenance efficiency according to claim 6, characterized in that: Parameter configuration is performed based on the risk quantification value field and the collection area meta-attribute to obtain the fixed cost of a single inspection of the collection area and the cumulative risk cost per unit time, including: Calculate the product of the risk quantification value field and the accumulated time, and set it as the accumulated risk cost per unit time; Calculate the path distance between the collection area and the inspection starting area, and set it as the fixed cost of a single inspection in the collection area.

8. An intelligent inspection and management system for improving security and maintenance efficiency, characterized by: The system for implementing the intelligent inspection and management method for improving security and maintenance efficiency according to any one of claims 1 to 7 comprises: A data acquisition module is used to acquire digital image data of the target monitoring area through a binocular camera and extract meta-attributes of the digital image data, wherein the meta-attributes include regional function type code, spatial position coordinates and lighting condition parameters; a weight determination module, configured to call a pre-trained regional feature analysis model to determine a multi-dimensional security and maintenance indicator weight distribution based on the regional function type code, spatial position coordinates, and lighting condition parameters of the digital image data, wherein the regional feature analysis model is obtained by machine learning training using multiple sets of data, each of which includes meta-attribute record data and a label identifying the security and maintenance indicator weight data and the function type meta-attribute, the function type meta-attribute representing the regional function type code in the meta-attribute record value set; A semantic segmentation module, configured to perform semantic segmentation on the digital image data based on multi-dimensional security and maintenance indicators to generate a set of sub-regions having a topological connection relationship; An indicator calculation module, configured to perform weighted mean calculation on the multi-dimensional security and maintenance indicator characteristic values ​​of each sub-area in the sub-area set based on the multi-dimensional security and maintenance indicator weight distribution to obtain a risk quantification value set; The intelligent inspection module is used to map the risk quantification value to the inspection period query table bound to the meta-attribute, output the corresponding inspection time interval parameters, generate the inspection task instruction set, and send it to the execution terminal device for intelligent inspection management.

Images