A big data-based underground pipe gallery fault data traceability processing method and system

Abstract

This invention relates to the field of image processing technology, and discloses a method and system for tracing and processing fault data in underground utility tunnels based on big data. The method includes: acquiring multi-view monitoring image data of underground utility tunnels; downsampling the multi-view monitoring image data to obtain differentiated resolution images, and performing edge feature enhancement on the differentiated resolution images according to a pyramid structure to obtain an enhanced feature map; identifying abnormal regions in the enhanced feature map; verifying the authenticity of the abnormal regions in the spatial structure of the underground utility tunnel to obtain multi-dimensional feature descriptors; constructing a spatiotemporal relationship feature map based on the spatiotemporal feature information in the multi-dimensional feature descriptors and the feature dimension correlation of the abnormal regions; and performing multi-angle trajectory tracing on the multi-dimensional feature descriptors based on the spatiotemporal relationship feature map to obtain anomaly tracing information. This invention can improve the efficiency of tracing and processing fault data in underground utility tunnels based on big data.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method and system for tracing and processing fault data in underground utility tunnels based on big data. Background Technology

[0002] As a critical carrier of urban infrastructure, the accuracy and efficiency of fault data tracing in underground utility tunnels directly impact the timeliness of operation and maintenance. Existing technologies, when processing multi-view monitoring image data, fail to effectively address the challenges of varying data resolutions and complex feature information. Feature enhancement methods lack specificity and struggle to fully extract key fault-related features, resulting in insufficient accuracy in identifying abnormal areas. This leads to false alarms or missed genuine anomalies, creating potential problems for subsequent tracing efforts.

[0003] In the fault tracing and analysis phase, existing technologies often overlook the inherent correlation between spatiotemporal characteristics and the characteristic dimensions of abnormal areas, lacking a systematic mechanism for constructing spatiotemporal relationships. The tracing process is mostly limited to single-dimensional trajectory tracking, failing to achieve multi-angle and comprehensive feature correlation analysis. This makes it difficult for the tracing link to fully present the evolution process and propagation path of the fault, resulting in vague fault root cause location, low tracing efficiency, and an inability to meet the actual operation and maintenance needs of rapid troubleshooting and precise handling of underground utility tunnel faults. Summary of the Invention

[0004] This invention provides a method and system for tracing and processing fault data in underground utility tunnels based on big data, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, this invention provides a method for tracing and processing fault data in underground utility tunnels based on big data, comprising:

[0006] S1. Acquire multi-view monitoring image data of underground utility tunnels;

[0007] S2. Downsample the multi-view monitoring image data to obtain a differentiated resolution image of the underground utility tunnel, and perform edge feature enhancement on the differentiated resolution image according to the pyramid structure to obtain an enhanced feature map of the underground utility tunnel.

[0008] S3. Identify the abnormal regions in the enhanced feature map based on the fault-related abnormal dynamic features in the enhanced feature map;

[0009] S4. Verify the authenticity of the abnormal area in the spatial structure of the underground utility tunnel to obtain the multi-dimensional feature descriptor of the enhanced feature map;

[0010] S5. Based on the spatiotemporal feature information in the multi-dimensional feature descriptor and the correlation between the feature dimensions of the abnormal area, construct the spatiotemporal relationship feature map of the underground utility tunnel;

[0011] S6. Based on the spatiotemporal relationship feature map, perform multi-angle trajectory tracing on the multi-dimensional feature descriptor to obtain the anomaly tracing information of the underground utility tunnel.

[0012] In a preferred embodiment, acquiring multi-view monitoring image data of the underground utility tunnel includes:

[0013] Collect continuous image sequence data from different perspectives in the underground utility tunnel to obtain the original image data of the underground utility tunnel;

[0014] Video frames are extracted from the original image data to obtain a multi-view image frame set of the underground utility tunnel;

[0015] Unify the image format of the multi-view image frame set to generate multi-view monitoring image data of the underground utility tunnel;

[0016] The multi-view monitoring image data of the underground utility tunnel is obtained by verifying the view coverage integrity and temporal consistency of the multi-view monitoring image data.

[0017] In a preferred embodiment, the step of downsampling the multi-view monitoring image data to obtain a differentiated resolution image of the underground utility tunnel, and then performing edge feature enhancement on the differentiated resolution image according to a pyramid structure to obtain an enhanced feature map of the underground utility tunnel, includes:

[0018] The multi-view monitoring image data is downsampled to obtain a differentiated resolution image of the underground utility tunnel.

[0019] Based on the pyramid structure, image edges at different resolution levels in the differential resolution image are detected to obtain a multi-scale edge map of the underground utility tunnel;

[0020] Based on the grayscale difference changes in the multi-scale edge map, the edge features of the multi-scale edge map are enhanced to obtain the edge-enhanced image of the underground utility tunnel;

[0021] The edge-enhanced images are fused to obtain an enhanced feature map of the underground utility tunnel.

[0022] In a preferred embodiment, identifying the abnormal region of the enhanced feature map based on the fault-related abnormal dynamic features in the enhanced feature map includes:

[0023] Extract the dynamic feature sequence that is temporally related to the fault from the enhanced feature map;

[0024] Based on the abnormal mutation trend of image points in the dynamic feature sequence, abnormal feature points that deviate from normal changes in the enhanced feature map are identified.

[0025] The spatial distribution density of the abnormal feature points is statistically analyzed to obtain candidate abnormal regions of the enhanced feature map;

[0026] Spatial consistency verification is performed on the candidate abnormal regions to filter out the abnormal regions in the enhanced feature map.

[0027] In a preferred embodiment, identifying anomalous feature points that deviate from normal changes in the enhanced feature map based on the anomalous abrupt change trend of image points in the dynamic feature sequence includes:

[0028] Based on the dynamic feature sequence, feature points in the enhanced feature map are connected in chronological order to obtain the feature change trajectory of the dynamic feature sequence;

[0029] The deviation of the feature change trajectory is compared with a preset normal feature change pattern to obtain the degree of deviation of the feature change trajectory. The formula for calculating the degree of deviation is as follows:

[0030] ;

[0031] In the formula, The degree of deviation, The time window length of the feature change trajectory is, At the current time point, The first-order difference of the characteristic change trajectory at adjacent time points. This represents the first-order difference of the normal characteristic change pattern at adjacent time points. The second-order difference of the characteristic change trajectory at adjacent time points, The second-order difference of the normal feature change pattern at adjacent time points. The standard deviation of the first difference in the normal characteristic change pattern is given. The standard deviation of the second difference in the normal characteristic change pattern is given. These are the weighting coefficients of the first-order difference term. These are the weighting coefficients of the second-order difference term. The feature change trajectory at time point eigenvalues, The normal feature change pattern at time points The baseline eigenvalues;

[0032] The deviation degree is compared with a threshold to obtain the abnormal feature points of the dynamic feature sequence.

[0033] In a preferred embodiment, verifying the authenticity of the abnormal region within the spatial structure of the underground utility tunnel to obtain the multi-dimensional feature descriptor of the enhanced feature map includes:

[0034] The internal components and spatial layout information of the underground utility tunnel are used as spatial structure data.

[0035] Spatial registration is performed between the abnormal region and the spatial structure data to obtain the spatial positional relationship between the abnormal region and the corresponding pipe gallery component;

[0036] Based on the spatial location relationship, the consistency between the abnormal area and the structural topology relationship of the pipe gallery components is verified, and abnormal structural areas of components that conform to the spatial structural data are selected.

[0037] The texture features, shape features, and dynamic change trend features of the abnormal region of the component structure are integrated to obtain the multi-dimensional feature descriptor of the enhanced feature map.

[0038] In a preferred embodiment, the step of constructing a spatiotemporal relationship feature map of the underground utility tunnel based on the correlation between the spatiotemporal feature information in the multi-dimensional feature descriptor and the feature dimensions of the abnormal region includes:

[0039] The points representing different time points and different spatial locations in the multi-dimensional feature descriptor are used as graph nodes;

[0040] The node paths connecting continuous time series in the multi-dimensional feature descriptor are used as time-series edges;

[0041] The spatial edges are defined as the propagation paths of adjacent spatial location nodes in the multi-dimensional feature descriptor along the spatial dimension.

[0042] The similarity of the feature vectors in the multi-dimensional feature descriptor is measured, and the feature edges of the multi-dimensional feature descriptor are established based on the measurement results and the degree of node clustering in the feature space.

[0043] Based on the graph nodes, the temporal edges, the spatial edges, and the feature edges, a spatiotemporal relationship feature graph of the underground utility tunnel is constructed.

[0044] In a preferred embodiment, the formula for calculating the feature edge weights in the multi-dimensional feature descriptor is as follows:

[0045] ;

[0046] In the formula, For the nodes in the multi-dimensional feature descriptor With nodes The feature edge weights between them For the nodes in the multi-dimensional feature descriptor With nodes The cosine distance between the eigenvectors. The mean of the cosine distances between all node pairs in the multi-dimensional feature descriptor is given. The standard deviation of the cosine distance between all node pairs in the multi-dimensional feature descriptor is given. For the nodes in the multi-dimensional feature descriptor With nodes The space between them is Euclidean distance. For the nodes in the multi-dimensional feature descriptor With nodes The absolute time difference between them The preset spatial scale factor, The preset time scale factor, It is a natural constant. It is a natural constant.

[0047] In a preferred embodiment, the step of performing multi-angle trajectory tracing on the multi-dimensional feature descriptors based on the spatiotemporal relationship feature map to obtain the anomaly tracing information of the underground utility tunnel includes:

[0048] The spatiotemporal association paths corresponding to the multi-dimensional feature descriptors in the spatiotemporal relationship feature map are used as the candidate tracing path set for the underground utility tunnel;

[0049] From the perspectives of temporal evolution, spatial propagation, and feature association in the spatiotemporal relationship feature map, multi-angle trajectory analysis is performed on the candidate source tracing path set to obtain the preliminary source tracing trajectory of the candidate source tracing path set.

[0050] The preliminary tracing trajectory is resolving trajectory conflicts to obtain the complete tracing link of the underground utility tunnel;

[0051] Based on the complete tracing link, the abnormal feature information in the multi-dimensional feature descriptor is integrated to obtain the abnormal tracing information of the underground utility tunnel.

[0052] To address the aforementioned problems, this invention also provides a big data-based system for tracing and processing fault data in underground utility tunnels, the system comprising:

[0053] The multi-source data acquisition module is used to acquire multi-view monitoring image data of underground utility tunnels;

[0054] A multi-scale feature enhancement module is used to downsample the multi-view monitoring image data to obtain a differentiated resolution image of the underground utility tunnel, and to enhance the edge features of the differentiated resolution image according to a pyramid structure to obtain an enhanced feature map of the underground utility tunnel.

[0055] An abnormal region identification module is used to identify abnormal regions in the enhanced feature map based on fault-related abnormal dynamic features in the enhanced feature map.

[0056] The feature verification and description module is used to verify the authenticity of the abnormal area in the spatial structure of the underground utility tunnel and obtain the multi-dimensional feature descriptor of the enhanced feature map.

[0057] The spatiotemporal relationship construction module is used to construct the spatiotemporal relationship feature map of the underground utility tunnel based on the spatiotemporal feature information in the multi-dimensional feature descriptor and the correlation between the feature dimensions of the abnormal area.

[0058] The trajectory tracing analysis module is used to perform multi-angle trajectory tracing on the multi-dimensional feature descriptor based on the spatiotemporal relationship feature map to obtain the abnormal tracing information of the underground utility tunnel.

[0059] Compared with the prior art, the present invention has the following beneficial effects:

[0060] 1. This invention downsamples multi-view monitoring images of underground utility tunnels to obtain images with varying resolutions, and combines this with edge feature enhancement using a pyramid structure. This fully leverages image edge information at different resolution levels, effectively strengthening fault-related key features and generating more accurate enhanced feature maps. Simultaneously, based on fault-related abnormal dynamic feature extraction and spatial consistency verification within the enhanced feature maps, it accurately identifies real abnormal areas, reduces feature misjudgments, provides reliable basic data support for subsequent fault tracing, and improves the effectiveness of early-stage data processing in fault tracing.

[0061] 2. This invention constructs a spatiotemporal relationship feature map, deeply associating spatiotemporal feature information in multi-dimensional feature descriptors with the feature dimensions of abnormal regions, achieving collaborative representation of spatiotemporal and feature dimensions, and making the correlation of fault-related features clearer. Based on this, trajectory tracing of multi-dimensional feature descriptors is performed from multiple angles, and a complete tracing link is obtained through trajectory conflict resolution. Finally, abnormal feature information is integrated to generate accurate abnormal tracing information, significantly improving the completeness and efficiency of fault tracing, and providing comprehensive and reliable tracing basis for the precise location of underground utility tunnel faults. Attached Figure Description

[0062] Figure 1 This is a flowchart illustrating a method for tracing and processing fault data in underground utility tunnels based on big data, provided in an embodiment of the present invention.

[0063] Figure 2 A functional module diagram of an underground utility tunnel fault data tracing and processing system based on big data, provided in an embodiment of the present invention;

[0064] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0065] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0066] This application provides a method for tracing and processing fault data in underground utility tunnels based on big data. The executing entity of this method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, this method can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cluster of cloud servers. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0067] Reference Figure 1 The diagram shown is a flowchart illustrating a method for tracing and processing fault data in underground utility tunnels based on big data, according to an embodiment of the present invention. In this embodiment, the method includes:

[0068] S1. Acquire multi-view monitoring image data of underground utility tunnels;

[0069] In this embodiment of the invention, acquiring multi-view monitoring image data of the underground utility tunnel includes:

[0070] Collect continuous image sequence data from different perspectives in the underground utility tunnel to obtain the original image data of the underground utility tunnel;

[0071] Video frames are extracted from the original image data to obtain a multi-view image frame set of the underground utility tunnel;

[0072] Unify the image format of the multi-view image frame set to generate multi-view monitoring image data of the underground utility tunnel;

[0073] The multi-view monitoring image data of the underground utility tunnel is obtained by verifying the view coverage integrity and temporal consistency of the multi-view monitoring image data.

[0074] High-definition network cameras are installed at key locations within the underground utility tunnel, including the tunnel's top, side walls, and the areas surrounding various critical equipment. Top cameras are angled to cover the entire tunnel, side wall cameras focus on the pipe connections, and cameras around critical equipment are positioned close to the equipment itself, ensuring each camera captures a unique, pre-defined perspective. All cameras are connected to a local storage server within the tunnel via network cables. The cameras are set to a fixed high frame rate and activated from the tunnel's operational launch, continuously recording. The resulting sequence of images from each perspective is transmitted in real-time to the storage server via network cables and stored in folders on the server's hard drive, each folder corresponding to a camera's perspective. This process ultimately yields raw image data of the underground utility tunnel covering all pre-defined monitoring perspectives.

[0075] The raw image data of the underground utility tunnel, categorized by viewpoint and stored on the local server hard drive, is imported one by one into the video frame extraction tool. For each continuous image sequence from a viewpoint, the "Extraction Interval" setting is located in the tool's interface. The extraction interval is set to extract one frame at a fixed time interval, ensuring data continuity while avoiding redundancy. After setting this, the "Start Extraction" button is clicked. The tool automatically extracts corresponding single frames from the currently imported continuous image sequence at the set fixed time intervals, automatically adding a filename containing the viewpoint number and extraction time for each frame extracted. Once the continuous image sequence for one viewpoint is extracted, all extracted single frames from that viewpoint are organized into a new folder, with the folder name labeled with the corresponding viewpoint number. This extraction and organization process is then repeated for all other continuous image sequences from different viewpoints. After completion, all image frame folders corresponding to all viewpoints are merged into a single main folder, resulting in a multi-view image frame set of the underground utility tunnel.

[0076] Launch the image format conversion tool. In the tool's "Output Format" option, select JPEG as the unified image format. Then, in the "Image Resolution" settings module, adjust the output resolution to a unified resolution that meets the monitoring clarity requirements. This resolution ensures image clarity while controlling file size. From the multi-view image frame set of the underground utility tunnel, open each image frame folder sequentially according to the viewpoint order and import the image frames in each folder into the format conversion tool's processing queue. For each imported image frame, the tool automatically identifies its original format and resolution and processes it according to the preset JPEG format and unified resolution. During processing, the tool converts the image's pixel arrangement and encoding method to ensure that the converted image conforms to the JPEG format standard and has a unified resolution. After processing, click the "Save" button to save the converted image frames to a new storage path labeled "Unified Format," maintaining the original viewpoint classification and file naming rules. Repeat this import, conversion, and save operation until all image frames in the multi-view image frame set have completed format conversion and resolution unification, ultimately generating multi-view monitoring image data of the underground utility tunnel.

[0077] First, conduct a view coverage integrity verification. Retrieve the detailed design and construction drawings of the underground utility tunnel and mark all key areas requiring monitoring on the drawings with a red marker. These key areas include all pipe interfaces within the tunnel, operating surfaces of various valve devices, air inlets and outlets of ventilation equipment, lampshade areas of lighting fixtures, and every corner of the tunnel passageway, ensuring no key areas are missed. Next, access the multi-view monitoring image data of the underground utility tunnel and review all images from each viewpoint in sequence. Compare the images with the key areas marked on the drawings to confirm that each red-marked key area is clearly visible in at least one viewpoint. If a key area is not found in any viewpoint, return to check for any deviations in the initial camera installation angles, adjust them, re-collect data, and repeat the subsequent steps. If all key areas are covered by images from their corresponding viewpoints, the view coverage integrity verification is successful. Next, a temporal consistency verification was performed. Images from different perspectives targeting the same key area of ​​the utility tunnel were selected from the multi-view monitoring image data. These images were then sorted by the extraction time in their filenames. The timestamp information in each image filename was then checked to ensure that the sorted images had consecutive timestamps in chronological order, without any jumps, repetitions, or misalignments. If the timestamps were completely consecutive and consistent, the temporal consistency verification passed. After both verifications passed, the final multi-view monitoring image data of the underground utility tunnel was obtained.

[0078] The beneficial effects include the continuous acquisition and categorized storage of raw image data through the deployment of multi-view high-definition cameras at key locations in underground utility tunnels. Combined with a fixed-time-interval image frame extraction strategy, data continuity is ensured while redundancy is avoided. After standardized format conversion and resolution adjustment, both image clarity and file size are considered appropriately. Verification of view coverage completeness ensures no key areas are missed, and temporal consistency verification guarantees the logical continuity of data over time. The resulting multi-view monitoring image data is characterized by comprehensive coverage, standardized timing, unified format, and reliable quality, providing a high-quality data foundation for subsequent image analysis and related applications in underground utility tunnels.

[0079] S2. Downsample the multi-view monitoring image data to obtain a differentiated resolution image of the underground utility tunnel, and perform edge feature enhancement on the differentiated resolution image according to the pyramid structure to obtain an enhanced feature map of the underground utility tunnel.

[0080] In this embodiment of the invention, the step of downsampling the multi-view monitoring image data to obtain a differentiated resolution image of the underground utility tunnel, and then performing edge feature enhancement on the differentiated resolution image according to a pyramid structure to obtain an enhanced feature map of the underground utility tunnel, includes:

[0081] The multi-view monitoring image data is downsampled to obtain a differentiated resolution image of the underground utility tunnel.

[0082] Based on the pyramid structure, image edges at different resolution levels in the differential resolution image are detected to obtain a multi-scale edge map of the underground utility tunnel;

[0083] Based on the grayscale difference changes in the multi-scale edge map, the edge features of the multi-scale edge map are enhanced to obtain the edge-enhanced image of the underground utility tunnel;

[0084] The edge-enhanced images are fused to obtain an enhanced feature map of the underground utility tunnel.

[0085] Open the image sampling tool and import the multi-view monitoring image data of the underground utility tunnel. In the "Sampling Method" option of the tool's operation interface, select mean sampling. This method ensures that the sampled image retains the overall grayscale characteristics of the original image. For each imported monitoring image, the tool automatically divides the image into square pixel blocks of the same size. The grayscale values ​​of all pixels within each pixel block are summed and then divided by the total number of pixels in that pixel block to obtain the average grayscale value of that pixel block. This average grayscale value is then used to replace all pixels in the entire pixel block, completing one downsampling operation. Perform the above mean sampling operation on all images in the multi-view monitoring image data, and set different pixel block sizes for different images, so that some images are sampled with larger pixel blocks and some with smaller pixel blocks. Finally, a set of images with different resolutions is obtained, namely, the differentiated resolution images of the underground utility tunnel.

[0086] Differential resolution images of the underground utility tunnel are arranged in descending order of resolution to form a pyramid structure. The highest resolution image forms the bottom layer of the pyramid, the second highest resolution image forms the second layer, and so on, with the lowest resolution image forming the top layer, creating a clear hierarchy of resolutions. Starting from the bottom layer of the pyramid, the differential resolution images of that layer are imported into an edge detection tool. The tool scans each pixel horizontally and vertically, starting from the top left corner of the image. During the scan, it compares the grayscale values ​​of the current pixel with those of its adjacent pixels above, below, to the left, and to the right. If the difference in grayscale values ​​between any pair of adjacent pixels exceeds a preset significant difference range, the current pixel is marked as an edge point. Using the same scanning and comparison method, edge detection is performed on the differential resolution images of all layers in the pyramid structure. The edge points detected at each layer are connected to form complete edge lines. Finally, the edge detection results from all layers are integrated to obtain a multi-scale edge map of the underground utility tunnel.

[0087] Open the image grayscale analysis tool, import the multi-scale edge map of the underground utility tunnel, and the tool will automatically count the grayscale values ​​of all pixels in the image, filtering out pixels belonging to edge areas and pixels belonging to non-edge areas. It will then calculate the average grayscale value of pixels in these two categories to determine the original grayscale difference between edge and non-edge areas. Next, launch the grayscale adjustment tool, import the multi-scale edge map into it, and based on the previously counted grayscale differences, adjust the grayscale value of each pixel in the edge areas by increasing it by a certain margin. Simultaneously, adjust the grayscale value of each pixel in the non-edge areas by decreasing it by a certain margin. This adjustment method further amplifies the grayscale difference between edge and non-edge areas, making the edge lines more prominent in the image. After completing the grayscale adjustment for all pixels in the multi-scale edge map, save the adjusted image to obtain the enhanced edge image of the underground utility tunnel.

[0088] The fusion method is determined to be weighted fusion. Weights are assigned to each edge enhancement image based on its resolution; higher-resolution images receive greater weights. Specifically, the weights are set according to the ratio of each image's resolution to the highest resolution among all edge enhancement images. For example, images with the highest resolution receive higher weights, while images with half the resolution receive lower weights. All edge enhancement images are imported into an image fusion tool. The tool first determines an alignment reference for all images, using the highest-resolution edge enhancement image as the reference. Other resolution edge enhancement images are then aligned pixel-wise with the reference image. After alignment, the tool scans each pixel in the image, multiplying the grayscale value of the corresponding pixel in each edge enhancement image by its corresponding weight value. All products are summed, and then divided by the sum of all weight values ​​to obtain the fused grayscale value for that pixel. This fused grayscale value is used as the pixel value for that position in the final image. After all pixel positions have undergone fusion calculations, the final generated image is saved, resulting in the enhanced feature map of the underground utility tunnel.

[0089] The beneficial effects are that by constructing an image pyramid structure through differential resolution sampling, and combining multi-scale edge detection, the edge information of underground utility tunnel images is accurately captured. After grayscale adjustment, the grayscale difference between edge and non-edge areas is further expanded. Then, the effective features of each level are integrated through a resolution-based weighted fusion strategy. The final enhanced feature map can fully retain the overall features and details of the image, significantly improve the prominence of edge areas and detection accuracy, and has a clear hierarchical structure and distinct feature expression, providing high-quality data support for image analysis and subsequent applications related to underground utility tunnels.

[0090] S3. Identify the abnormal regions in the enhanced feature map based on the fault-related abnormal dynamic features in the enhanced feature map;

[0091] In this embodiment of the invention, identifying the abnormal region of the enhanced feature map based on the fault-related abnormal dynamic features in the enhanced feature map includes:

[0092] Extract the dynamic feature sequence that is temporally related to the fault from the enhanced feature map;

[0093] Based on the abnormal mutation trend of image points in the dynamic feature sequence, abnormal feature points that deviate from normal changes in the enhanced feature map are identified.

[0094] The spatial distribution density of the abnormal feature points is statistically analyzed to obtain candidate abnormal regions of the enhanced feature map;

[0095] Spatial consistency verification is performed on the candidate abnormal regions to filter out the abnormal regions in the enhanced feature map.

[0096] Based on the dynamic feature sequence, feature points in the enhanced feature map are connected in chronological order to obtain the feature change trajectory of the dynamic feature sequence;

[0097] The deviation of the feature change trajectory is compared with a preset normal feature change pattern to obtain the degree of deviation of the feature change trajectory. The formula for calculating the degree of deviation is as follows:

[0098] ;

[0099] In the formula, The degree of deviation, The time window length of the feature change trajectory is, At the current time point, The first-order difference of the characteristic change trajectory at adjacent time points. This represents the first-order difference of the normal characteristic change pattern at adjacent time points. The second-order difference of the characteristic change trajectory at adjacent time points, The second-order difference of the normal feature change pattern at adjacent time points. The standard deviation of the first difference in the normal characteristic change pattern is given. The standard deviation of the second difference in the normal characteristic change pattern is given. These are the weighting coefficients of the first-order difference term. These are the weighting coefficients of the second-order difference term. The feature change trajectory at time point eigenvalues, The normal feature change pattern at time points The baseline eigenvalues;

[0100] The deviation degree is compared with a threshold to obtain the abnormal feature points of the dynamic feature sequence.

[0101] The enhanced feature images of underground utility tunnels are arranged chronologically to form a temporal enhanced feature image set. From this set, feature types related to common tunnel faults are selected, specifically including the curvature of edge lines, grayscale value changes in specific areas, and pixel aggregation. For each selected fault-related feature, numerical information is extracted from the temporal enhanced feature image set. The numerical changes of the feature are recorded at time points, with time as the horizontal axis and feature value as the vertical axis. For example, the curvature angle of the edge lines at pipe joints and the grayscale value of areas where leakage may occur are recorded at fixed time intervals. All temporal numerical records of fault-related features are integrated into an ordered dataset, with each dataset corresponding to the temporal changes of one fault feature. This results in a dynamic feature sequence in the enhanced feature image set that is temporally related to the fault.

[0102] Historical dynamic feature sequences under normal operating conditions of the utility tunnel are collected. The changes in feature values ​​for each image point in the sequence are statistically analyzed to calculate the normal fluctuation range of the feature values ​​for each image point. This range includes all possible feature values ​​for that image point during normal operation. The dynamic feature sequence to be analyzed is compared point-by-point with the normal fluctuation range of the corresponding image points to observe the trend of feature value changes for each image point. If the feature value of an image point suddenly jumps outside the normal fluctuation range within several consecutive time points, and the magnitude of the change far exceeds the maximum magnitude of the normal fluctuation, an abnormal abrupt change trend has occurred. All image points exhibiting such abnormal abrupt change trends are marked and compiled into a list, ultimately identifying anomalous feature points that deviate from normal changes in the enhanced feature map.

[0103] A uniformly divided square grid is overlaid on the enhanced feature map, with each grid serving as an independent spatial statistical unit. The grid size is determined to clearly reflect the aggregation of anomalous feature points, ensuring that each grid can accommodate a certain number of pixels. The number of anomalous feature points in each grid is counted, and the count is recorded. Based on the feature point density of anomalous areas in the historical fault data of the utility tunnel, a density threshold is set. This threshold serves as the standard for determining whether a grid belongs to a dense region. When the number of anomalous feature points in a grid exceeds this threshold, the grid is marked as a dense grid. Adjacent dense grids are merged to form continuous regions. These continuous regions are the candidate anomalous regions for the enhanced feature map. Finally, the location and extent information of all candidate anomalous regions are recorded in the image annotation file.

[0104] Examine the feature attributes of all pixels within each candidate anomaly region, including the direction of edge lines, the distribution pattern of grayscale values, and the clustering pattern of pixels. Determine if the feature attributes of all pixels within the region are consistent. If there are clusters of pixels with significantly different feature attributes within the region, and these clusters differ greatly from the overall features of the region, then the spatial consistency of the candidate region is not up to standard. Compare the features of the candidate anomaly region with those of the surrounding normal region, observing whether there is a clear and continuous boundary between the candidate region and the normal region, and whether the feature changes at the boundary conform to the feature difference pattern between the faulty region and the normal region. If the boundary between the candidate region and the normal region is blurred, or the feature changes at the boundary do not conform to the fault characteristics, then the spatial consistency of the candidate region is not up to standard. Select candidate anomaly regions that meet the spatial consistency standard, and remove those that do not, finally obtaining the anomaly regions of the enhanced feature map. Mark the location and extent of these anomaly regions on the image with a striking color.

[0105] Key information corresponding to each time point is extracted from the dynamic feature sequence. This information includes the specific location identifiers of feature points in the enhanced feature map. The location identifiers are determined by the gray value distribution and edge contour attributes of the feature points in the image, and the horizontal and vertical coordinates of each feature point are clearly defined in the image coordinate system. Following the chronological order from earliest to latest, the coordinates of feature points corresponding to adjacent time points are associated. Straight line segments are used to connect the feature points of the preceding and following time points sequentially, ensuring that the line segments accurately pass through the coordinate positions of the two feature points, and each line segment is labeled with the corresponding time interval information. After all feature points at all time points have been connected adjacently, a continuous line with chronological markers is formed. This line represents the feature change trajectory of the dynamic feature sequence.

[0106] A preset normal characteristic change pattern is obtained. This pattern is generated by collecting multiple sets of dynamic characteristic sequences of underground utility tunnels under long-term normal operation. Following the method described above for generating characteristic change trajectories, normal characteristic change trajectories are generated for each set of sequences. The average position coordinates of all normal trajectories at the same time node are then taken to form a standard normal characteristic change pattern curve. Each time node in the curve corresponds to a fixed standard position coordinate. The obtained characteristic change trajectory is compared with this standard curve node by node. For each time node, the difference between the horizontal coordinate of the feature point on the characteristic change trajectory and the horizontal coordinate of the corresponding node on the standard curve, as well as the difference between the vertical coordinate and the vertical coordinate of the feature point on the characteristic change trajectory, are calculated. The two differences are added together to obtain the single-point deviation value for that time node. The single-point deviation values ​​for all time nodes are arranged in chronological order to form a complete deviation degree data set, which represents the deviation degree of the characteristic change trajectory.

[0107] A threshold for the degree of deviation is determined by analyzing deviation data from historical failure cases in underground utility tunnels. The minimum deviation value at the first occurrence of an abnormal signal before a failure is selected and set as the threshold. This threshold is ensured to be lower than all failure-related deviation values ​​and higher than the maximum deviation value that may occur under normal operating conditions. Each individual deviation value in the deviation data set of the characteristic change trajectory is compared with the set threshold. If the deviation value of a single point at a certain time node exceeds the threshold, the feature point on the characteristic change trajectory corresponding to that time node is marked as an abnormal feature point. The marking includes an "abnormal" label and the corresponding time node next to the feature point's coordinates. After all individual deviation values ​​have been compared against the threshold, all feature points marked with "abnormal" are compiled into a list in chronological order. The list includes the time node of each abnormal feature point, its coordinate position in the enhanced feature map, and the corresponding individual deviation value, ultimately yielding the abnormal feature points of the dynamic feature sequence.

[0108] The time window length is the total number of time nodes extracted from a pre-defined continuous time interval in the feature change trajectory. The process involves defining the continuous time range to be analyzed within the feature change trajectory and counting the specific number of independent time nodes within that range; this number is the time window length. The current time point is a specific time marker in the time series of the feature change trajectory where deviation calculations are currently being performed. The selection process involves determining the specific time position currently involved in the calculation from all time nodes in the feature change trajectory; the time corresponding to this position is the current time point.

[0109] The first-order difference of the characteristic change trajectory at adjacent time points is obtained by calculating the difference between the characteristic value at the current time point and the characteristic value at the previous adjacent time point in the characteristic change trajectory. The calculation process involves subtracting the characteristic value corresponding to the current time point from the characteristic value corresponding to the previous adjacent time point. The result is the first-order difference of that adjacent time point. The first-order difference of the normal characteristic change pattern at adjacent time points is obtained by calculating the difference between the baseline characteristic value at the current time point and the baseline characteristic value at the previous adjacent time point in the normal characteristic change pattern. The calculation process involves subtracting the baseline characteristic value corresponding to the current time point from the baseline characteristic value corresponding to the previous adjacent time point. The result is the first-order difference of that adjacent time point.

[0110] The second-order difference of the characteristic change trajectory at adjacent time points is obtained by calculating the difference between two adjacent first-order differences in the characteristic change trajectory. The calculation process involves subtracting the first-order difference of the next adjacent time point from the first-order difference of the previous adjacent time point. The result is the second-order difference at that adjacent time point. Similarly, the second-order difference of the normal characteristic change pattern at adjacent time points is obtained by calculating the difference between two adjacent first-order differences in the normal characteristic change pattern. The calculation process involves subtracting the first-order difference of the next adjacent time point from the first-order difference of the previous adjacent time point. The result is the second-order difference at that adjacent time point.

[0111] The standard deviation of the first difference in the normal characteristic change pattern is obtained by statistically calculating the first differences of all adjacent time points in the normal characteristic change pattern. The calculation process is as follows: First, the first differences of all adjacent time points in the normal characteristic change pattern are summed. The summation result is divided by the total number of first differences to obtain the average value. Then, the average value is subtracted from each first difference to obtain the corresponding deviation value. The square of each deviation value is calculated to obtain the squared deviation value. The sum of all squared deviation values ​​is divided by the total number of squared deviation values ​​to obtain the mean squared deviation value. Finally, the square root of the mean squared deviation value is taken. The result is the standard deviation of the first difference in the normal characteristic change pattern.

[0112] The standard deviation of the second difference in the normal characteristic change pattern is obtained by statistically calculating the second differences of all adjacent time points in the normal characteristic change pattern. The calculation process is as follows: First, the second differences of all adjacent time points in the normal characteristic change pattern are summed. The summation result is divided by the total number of second differences to obtain the average value. Then, the average value is subtracted from each second difference to obtain the corresponding deviation value. The square of each deviation value is calculated to obtain the squared deviation value. The sum of all squared deviation values ​​is divided by the total number of squared deviation values ​​to obtain the mean squared deviation value. Finally, the square root of the mean squared deviation value is taken. The result is the standard deviation of the second difference in the normal characteristic change pattern.

[0113] The weighting coefficient of the first-order difference term is a pre-set value based on the importance of the first-order difference in the deviation assessment. The setting process involves analyzing the influence of the first-order difference on the deviation of the characteristic change trajectory from the normal characteristic change pattern, and determining a fixed value based on this influence. This value reflects the contribution proportion of the first-order difference term in the overall deviation calculation. Similarly, the weighting coefficient of the second-order difference term is a pre-set value based on the importance of the second-order difference in the deviation assessment. The setting process involves analyzing the influence of the second-order difference on the deviation of the characteristic change trajectory from the normal characteristic change pattern, and determining a fixed value based on this influence. This value reflects the contribution proportion of the second-order difference term in the overall deviation calculation.

[0114] The feature value of the feature change trajectory at a given time point is extracted from the data source corresponding to the feature change trajectory. The extraction process involves locating the record entry corresponding to the time point in the feature change trajectory and obtaining the corresponding quantized feature value from that record entry; this value is the feature value at that time point. The baseline feature value of the normal feature change pattern at a given time point is extracted from the baseline data source corresponding to the normal feature change pattern. The extraction process involves locating the baseline record entry corresponding to the time point in the normal feature change pattern and obtaining the corresponding baseline quantized feature value from that baseline record entry; this value is the baseline feature value at that time point.

[0115] To calculate the deviation of the feature change trajectory from the normal feature change pattern, firstly, the difference between the first-order difference of the feature change trajectory and the first-order difference of the normal feature change pattern is calculated. This difference is then divided by the standard deviation of the first-order difference in the normal feature change pattern, and the result is squared to obtain the quantified deviation value for the first-order difference dimension. Similarly, the difference between the second-order difference of the feature change trajectory and the second-order difference of the normal feature change pattern is calculated, divided by the standard deviation of the second-order difference in the normal feature change pattern, and the result is squared to obtain the quantified deviation value for the second-order difference dimension.

[0116] Multiply the deviation quantization value of the first-order difference dimension by the weight coefficient of the first-order difference term to obtain the weighted deviation quantization value of the first-order difference dimension. Multiply the deviation quantization value of the second-order difference dimension by the weight coefficient of the second-order difference term to obtain the weighted deviation quantization value of the second-order difference dimension. Accumulate the weighted deviation quantization values ​​of the first-order difference dimension and the weighted deviation quantization values ​​of the second-order difference dimension for all time points within the time window to obtain the total weighted deviation quantization value within the time window.

[0117] Dividing the total weighted deviation quantization value by the time window length yields the degree of deviation of the feature change trajectory from the normal feature change pattern. This result can accurately quantify the difference level between the feature change trajectory and the normal feature change pattern. The higher the difference level, the greater the deviation, and the lower the difference level, the smaller the deviation.

[0118] The beneficial effects are as follows: by extracting the temporal dynamic features related to the fault in the enhanced feature map, and combining them with the feature fluctuation range under normal conditions, the abnormal change trend of image points can be accurately captured, and abnormal feature points that deviate from normal changes can be accurately identified; by statistically analyzing the spatial density of abnormal feature points through grids, candidate abnormal regions are determined, and then regions with chaotic feature attributes and blurred boundaries with normal regions are eliminated through spatial consistency verification, ensuring that the selected abnormal regions are real and reliable. The overall process can effectively locate abnormal regions in the enhanced feature map of underground utility tunnels, providing accurate and reliable basis for the identification of utility tunnel faults and helping to discover potential problems in utility tunnels in a timely manner.

[0119] By generating continuous feature change trajectories by associating feature points in the enhanced feature map in chronological order, the temporal evolution of features can be clearly presented. By comparing this trajectory with a standard pattern built based on long-term normal data, the degree of deviation at each time point can be accurately obtained, providing a reliable basis for anomaly judgment. Scientific thresholds are determined using historical fault data, and abnormal feature points can be accurately screened through threshold comparison. The entire process achieves accurate identification of abnormal feature points, providing strong support for underground utility tunnel fault monitoring and helping to discover potential risks in a timely manner.

[0120] S4. Verify the authenticity of the abnormal area in the spatial structure of the underground utility tunnel to obtain the multi-dimensional feature descriptor of the enhanced feature map;

[0121] In this embodiment of the invention, verifying the authenticity of the abnormal region within the spatial structure of the underground utility tunnel to obtain the multi-dimensional feature descriptor of the enhanced feature map includes:

[0122] The internal components and spatial layout information of the underground utility tunnel are used as spatial structure data.

[0123] Spatial registration is performed between the abnormal region and the spatial structure data to obtain the spatial positional relationship between the abnormal region and the corresponding pipe gallery component;

[0124] Based on the spatial location relationship, the consistency between the abnormal area and the structural topology relationship of the pipe gallery components is verified, and abnormal structural areas of components that conform to the spatial structural data are selected.

[0125] The texture features, shape features, and dynamic change trend features of the abnormal region of the component structure are integrated to obtain the multi-dimensional feature descriptor of the enhanced feature map.

[0126] The process involves collecting design and construction drawings and as-built data for the underground utility tunnel, extracting information on its internal components, including pipe material type, diameter, and laying direction; bracket installation location and support method; valve model and connection port; and installation locations of ventilation equipment and lighting devices, ensuring no critical components are overlooked. Simultaneously, on-site measurements are conducted to record the relative distances, height differences, and connection methods between components, determining the overall spatial layout of the tunnel, such as the vertical distance between pipes and brackets, the parallel spacing between different pipes, and the connection positions of valves and pipe interfaces. The extracted component information is integrated with the measured spatial layout information, and the data is segmented along the length of the tunnel into structured data. Each segment corresponds to a set of records containing component details and spatial locations, ultimately forming the spatial structure data of the underground utility tunnel.

[0127] Fixed markers within the utility tunnel are selected as registration reference points from the spatial structure data. These markers include embedded parts on the tunnel sidewalls, flanges at pipe interfaces, and fixing bolts on the top of supports. These markers have clear and unique coordinate positions in the spatial structure data. In the enhanced feature map containing the anomalous area, the corresponding fixed markers in the spatial structure data are found by observing image details. The image coordinates of these markers are then marked on the enhanced feature map using an image annotation tool. The actual coordinates of the reference points in the spatial structure data and the image coordinates of the reference points in the enhanced feature map are input into a coordinate transformation tool. The tool establishes a conversion rule between actual spatial coordinates and image coordinates based on the correspondence between the two sets of coordinates. This rule converts the image coordinates of the anomalous area in the enhanced feature map into the actual spatial coordinates of the utility tunnel. By comparing the actual spatial coordinates of the anomalous area with the coordinate ranges of each utility tunnel component in the spatial structure data, it is determined which component's coordinate range the anomalous area falls within, the distance between the anomalous area and the edge of that component, and the specific part of the component it covers. Finally, the spatial positional relationship between the anomalous area and the corresponding utility tunnel component is obtained.

[0128] Based on the spatial structure data, the structural topological relationships between each utility tunnel component are determined. For example, the topological relationship between a pipe and a support is that the support provides vertical support below the pipe, and the center point of the support is aligned vertically with the central axis of the pipe. The topological relationship between a valve and a pipe is that the two ends of the valve are fully connected to the pipe ports, and the central axis of the valve is on the same straight line as the central axis of the pipe. Combining the obtained anomaly areas with the spatial positional relationship of the corresponding utility tunnel components, it is determined whether the connection method between the component in the anomaly area and the surrounding components conforms to the preset topological relationship. If the anomaly area is located at the connection between the pipe and the support, it is checked whether the support is still below the pipe and whether its center point is aligned with the central axis of the pipe. If they are aligned, the topological relationship is consistent; if the support is offset to the side of the pipe, the topological relationship is inconsistent. Anomaly areas with consistent topological relationships are filtered out. These areas are anomalies inherent in the utility tunnel components themselves, rather than false anomalies caused by coordinate registration errors or misidentification of markers. Finally, the structural anomaly areas of the components that conform to the spatial structure data are obtained.

[0129] Image texture extraction methods are used to process abnormal areas in the component structure. By scanning the image of the abnormal area line by line, the grayscale value of each pixel is recorded. The frequency and amplitude of grayscale value changes between adjacent pixels are statistically analyzed to determine the texture type of the abnormal area. For example, water stains caused by pipe leakage appear as gradually changing and continuous stripes of grayscale values, while cracks caused by structural deformation appear as abruptly changing and discontinuous lines of grayscale values. These texture features are then compiled into textual descriptions. Image contour extraction tools are used to delineate the edges of the abnormal areas in the component structure. Based on the curvature, turning angle, and closure state of the edges, the shape characteristics of the abnormal areas are determined. For example, the edges of circular rusted areas are smooth with no obvious turning points, while the edges of irregular deformed areas have multiple turning angles. The specific manifestations of the shape characteristics are recorded. Enhanced feature images of the abnormal areas in the component structure at different time points are retrieved. The area size and grayscale distribution range of the abnormal areas at each time point are compared to analyze whether there is a trend of area expansion and increased grayscale difference in the abnormal areas, and the dynamic change trend characteristics are summarized. The organized texture features, shape features, and dynamic change trend features are integrated to form a structured document containing detailed descriptions of the three types of features. Each feature corresponds to a specific description, ultimately resulting in a multi-dimensional feature descriptor for the enhanced feature map.

[0130] The beneficial effects include: forming accurate spatial structure data by integrating information on utility tunnel components and spatial layout, providing a solid basis for anomaly verification; achieving accurate registration between anomaly areas and spatial structure data with the help of fixed markers, clearly defining the spatial positional relationship between anomaly areas and utility tunnel components; effectively eliminating false anomalies and screening out real component structural anomaly areas by verifying the consistency of structural topology; and integrating the texture, shape, and dynamic change characteristics of anomaly areas to form multi-dimensional feature descriptors, comprehensively presenting anomaly attributes, providing reliable and rich evidence for accurate judgment and subsequent handling of underground utility tunnel faults, and improving the accuracy and reliability of fault identification.

[0131] S5. Based on the spatiotemporal feature information in the multi-dimensional feature descriptor and the correlation between the feature dimensions of the abnormal area, construct the spatiotemporal relationship feature map of the underground utility tunnel;

[0132] In this embodiment of the invention, the step of constructing a spatiotemporal relationship feature map of the underground utility tunnel based on the spatiotemporal feature information in the multi-dimensional feature descriptor and the feature dimension correlation of the abnormal region includes:

[0133] The points representing different time points and different spatial locations in the multi-dimensional feature descriptor are used as graph nodes;

[0134] The node paths connecting continuous time series in the multi-dimensional feature descriptor are used as time-series edges;

[0135] The spatial edges are defined as the propagation paths of adjacent spatial location nodes in the multi-dimensional feature descriptor along the spatial dimension.

[0136] The similarity of the feature vectors in the multi-dimensional feature descriptor is measured, and the feature edges of the multi-dimensional feature descriptor are established based on the measurement results and the degree of node clustering in the feature space.

[0137] Based on the graph nodes, the temporal edges, the spatial edges, and the feature edges, a spatiotemporal relationship feature graph of the underground utility tunnel is constructed.

[0138] The formula for calculating the feature edge weights in the multi-dimensional feature descriptor is as follows:

[0139] ;

[0140] In the formula, For the nodes in the multi-dimensional feature descriptor With nodes The feature edge weights between them For the nodes in the multi-dimensional feature descriptor With nodes The cosine distance between the eigenvectors. The mean of the cosine distances between all node pairs in the multi-dimensional feature descriptor is given. The standard deviation of the cosine distance between all node pairs in the multi-dimensional feature descriptor is given. For the nodes in the multi-dimensional feature descriptor With nodes The space between them is Euclidean distance. For the nodes in the multi-dimensional feature descriptor With nodes The absolute time difference between them The preset spatial scale factor, The preset time scale factor, It is a natural constant. It is a natural constant.

[0141] Specific content representing temporal and spatial information is extracted from multi-dimensional feature descriptors. Temporal information corresponds to different data collection times within the anomaly area, and each time point must accurately correlate with the dynamic trend characteristics of the anomaly area at that moment. Spatial information corresponds to the specific spatial location of the anomaly area within the utility tunnel, and each spatial location must precisely match the actual installation location of internal components within the tunnel, such as pipe interfaces or support connection points. A unique identifier is assigned to each extracted time point and spatial location point, containing both temporal and spatial attributes to ensure each point is distinguishable. The uniquely identified time points and spatial locations are treated as independent node units and uniformly incorporated into a node set of a graph structure. Each node unit must clearly record the corresponding temporal or spatial information, ultimately forming the graph nodes of the spatiotemporal relationship feature map of the underground utility tunnel.

[0142] Nodes containing only time attributes are selected from the graph nodes. These time nodes are then sorted chronologically based on the collection time of abnormal areas recorded in the multi-dimensional feature descriptor, ensuring that adjacent nodes correspond to consecutive collection periods. For example, if one node corresponds to a collection time at a certain moment, the next node corresponds to the immediately following collection time. The node paths for the continuous time series are determined, covering all nodes sorted by time, and the path direction must be consistent with the direction of time flow. Line segments with time labels are used to connect the sorted consecutive time nodes. The line segments must be labeled with temporal attributes such as the time interval and time flow direction of the connected nodes. These attributes must be directly associated with the temporal correlation information of the dynamic trend features in the multi-dimensional feature descriptor, ultimately forming the temporal edges of the underground utility tunnel spatiotemporal relationship feature graph.

[0143] Nodes containing only spatial attributes are selected from the graph nodes. Based on the spatial structure data of the underground utility tunnel, the spatial adjacency relationship between nodes at each spatial location is determined. Spatial adjacency requires that the corresponding utility tunnel components are directly adjacent in the actual layout, such as two adjacent pipe sections or two supports installed side by side. The propagation path of adjacent spatial location nodes in the spatial dimension is analyzed. The propagation path must conform to the spatial distribution pattern of the components inside the utility tunnel. For example, from the middle node of one pipe section to the middle node of an adjacent pipe section, the path must extend along the length of the utility tunnel or the connection direction of the components. Spatially adjacent nodes are connected using line segments with spatial identifiers. The line segments must be labeled with spatial attributes such as spatial distance and spatial direction of the connecting nodes. These attributes must directly match the spatial layout-related feature information in the multi-dimensional feature descriptor, ultimately forming the spatial edges of the spatiotemporal relationship feature graph of the underground utility tunnel.

[0144] Feature vectors corresponding to each graph node are extracted from the multi-dimensional feature descriptor. These feature vectors must contain numerical descriptions of specific features such as texture and shape characteristics of abnormally associated regions. These numerical descriptions must be quantified by recording the visual representation of the features; for example, texture direction is explicitly described as "horizontal" or "vertical" and converted into corresponding identifiers. Feature vectors of any two graph nodes are compared feature-by-feature. During the comparison, it is necessary to determine whether the descriptions of each feature dimension are consistent, such as whether the texture density is within the same range or whether the shape contour has a similar form. The similarity between two feature vectors is determined by the number of consistent feature dimensions; the more consistent dimensions, the higher the similarity. Simultaneously, the distribution of all graph nodes in the feature space is observed, and it is determined whether nodes with highly similar feature vectors cluster together. The degree of clustering is measured by the proportion of nodes within the cluster to the total number of nodes. Nodes that reach a set clustering level are connected using line segments with similarity labels. The line segments must be labeled with similarity level, common feature dimensions, and other information, ultimately forming the feature edges of the spatiotemporal relationship feature map of the underground utility tunnel.

[0145] The generated graph nodes are arranged on the graph structure canvas according to the actual spatial layout and time sequence of the underground utility tunnel. Spatial nodes must correspond to the actual relative positions of the tunnel components, and time nodes must be arranged linearly along one side of the canvas in the order of data collection. Temporal edges are connected to the graph nodes ordered by time, ensuring that the temporal flow of the edges is consistent with the node sorting direction. Spatial edges are connected to spatially adjacent graph nodes, ensuring that the extension direction of the spatial edges conforms to the spatial distribution of the tunnel components. Feature edges are connected to graph nodes with high similarity clusters, ensuring that the identification information of the feature edges matches the similarity results of the node feature vectors. Attribute annotations are added to all nodes and edges in the graph structure, including the temporal / spatial attributes of the nodes and the temporal / spatial / feature attributes of the edges. All nodes and edges are checked to ensure they are correctly connected and complete. After confirmation, the complete graph structure is saved, ultimately constructing the spatiotemporal relationship feature graph of the underground utility tunnel.

[0146] Nodes in multi-dimensional feature descriptors With nodes The cosine distance between the eigenvectors of two nodes is obtained by calculating the eigenvectors of the two nodes. The calculation first extracts the eigenvectors of the nodes... and nodes The multi-dimensional features are transformed into numerical vectors of uniform dimension. The dot product of two vectors is then calculated, and the magnitudes of both vectors are calculated. The dot product is divided by the product of the magnitudes of the two vectors, and the result is the node. With nodes The cosine distance between the eigenvectors.

[0147] The mean cosine distance between all node pairs in the multi-dimensional feature descriptor is obtained by statistically analyzing the cosine distances of all node pairs. During the statistical analysis, all nodes in the multi-dimensional feature descriptor are traversed first to generate all possible combinations of node pairs. The cosine distance of each node pair is calculated, and the values ​​of all cosine distances are summed. The summation result is divided by the total number of node pairs, and the average value is the mean cosine distance between all node pairs.

[0148] In the multi-dimensional feature descriptor, the standard deviation of the cosine distance between all node pairs is a quantitative result that measures the dispersion of the cosine distance of all node pairs. The calculation first obtains the difference between the cosine distance of each node pair and the mean of the cosine distance between all node pairs. The difference is squared for each difference. All the squared results are summed. The sum is divided by the total number of node pairs to obtain the mean squared difference. The square root of the mean squared difference is then taken to obtain the standard deviation of the cosine distance between all node pairs.

[0149] Nodes in multi-dimensional feature descriptors With nodes The spatial Euclidean distance between nodes is calculated using the spatial coordinates of the nodes. The calculation begins by determining the nodes. and nodes In the preset spatial coordinate system, calculate the difference between the corresponding coordinate values ​​of two nodes, square each coordinate difference, sum all the squared results, and then take the square root of the sum. The result is the node's coordinate. With nodes The space between them is Euclidean distance.

[0150] Nodes in multi-dimensional feature descriptors With nodes The absolute time difference between the two nodes is calculated using their timestamps. The calculation first retrieves the timestamps of the nodes. and nodes For each corresponding timestamp, subtract the smaller timestamp value from the larger timestamp value; the difference is the node value. With nodes The absolute time difference between them.

[0151] The preset spatial scale factor is a fixed value that is set in advance. When setting it, the range of spatial Euclidean distance distribution of all node pairs in the multi-dimensional feature descriptor and the density of node spatial distribution are taken into account. Combined with the expected value range of feature edge weights, the adjustment effect of different values ​​on spatial distance is simulated, and the value that makes the influence of spatial distance on weights conform to the actual correlation law is selected. This value is the preset spatial scale factor.

[0152] The preset time scale factor is a fixed value set in advance. When setting it, the distribution characteristics of the absolute time difference of node pairs in the multi-dimensional feature descriptor and the time correlation requirements of node features are analyzed. The influence of time factors on the strength of node correlation in historical data is referenced. By simulating the adjustment effect of different values ​​on the time difference, the value that can reasonably quantify the role of time factors is selected. This value is the preset time scale factor.

[0153] Feature edge weights are used to measure the weight of nodes in a multi-dimensional feature descriptor. With nodes A quantitative indicator of the degree of correlation between nodes, whose numerical value directly reflects the comprehensive correlation strength between two nodes in three dimensions: features, space, and time. The product of the square root of the natural constant and the standard deviation of the cosine distance between all node pairs, multiplied by the square root of pi, and the reciprocal of this product, is used to determine the node... and The squared difference between the cosine distance between the feature vectors and the mean of the cosine distance between all node pairs is divided by the square of twice the standard deviation of the cosine distance between all node pairs. The negative of this result is then used to calculate the natural exponent. The exponent result is multiplied by the inverse of the above. The resulting Gaussian distribution term is used to measure the degree of similarity concentration between the feature vectors of the two nodes. The closer the cosine distance is to the mean, the larger this part is, and the higher the contribution of feature similarity to the association strength.

[0154] Divide the preset spatial scale factor by the number of nodes. and The sum of the spatial Euclidean distance and the preset spatial scale factor yields the spatial distance adjustment term. The smaller the spatial Euclidean distance, the larger the value of this adjustment term, and the more significant the effect of spatial proximity on the association strength.

[0155] Divide by the node using the preset time scale factor. and The sum of the absolute time difference and the preset time scale factor yields the time difference adjustment term. The smaller the absolute time difference, the larger the value of this adjustment term, and the more prominent the effect of time synchronization on the correlation strength.

[0156] Multiplying the Gaussian distribution term, spatial distance adjustment term, and time difference adjustment term together yields the node. With nodes The feature edge weights between nodes indicate that the larger the value, the better the overall performance of the two nodes in terms of feature similarity, spatial proximity, and temporal synchronization, and the closer the relationship. The smaller the value, the weaker the overall relationship. This provides the core weight basis for subsequent analysis of multi-dimensional feature descriptors.

[0157] The beneficial effects include using time points and spatial location points in multi-dimensional feature descriptors as graph nodes, ensuring that nodes have both clear attributes and distinguishability, laying the foundation for graph structure construction; constructing temporal edges based on continuous time series paths, clearly presenting the temporal evolution relationship of abnormal features; constructing spatial edges based on adjacent spatial propagation paths, accurately matching the actual spatial distribution pattern of utility tunnel components; establishing feature edges based on feature vector similarity and node clustering degree, effectively associating nodes with similar feature attributes; and finally integrating graph nodes and various edges to construct a spatiotemporal relationship feature graph, comprehensively integrating the spatiotemporal information and feature relationships of abnormal areas, providing structured feature basis for tracing the root causes of underground utility tunnel failures and predicting development trends, and improving the systematicness and accuracy of failure analysis.

[0158] S6. Based on the spatiotemporal relationship feature map, perform multi-angle trajectory tracing on the multi-dimensional feature descriptor to obtain the anomaly tracing information of the underground utility tunnel.

[0159] In this embodiment of the invention, the step of performing multi-angle trajectory tracing on the multi-dimensional feature descriptor based on the spatiotemporal relationship feature map to obtain the anomaly tracing information of the underground utility tunnel includes:

[0160] The spatiotemporal association paths corresponding to the multi-dimensional feature descriptors in the spatiotemporal relationship feature map are used as the candidate tracing path set for the underground utility tunnel;

[0161] From the perspectives of temporal evolution, spatial propagation, and feature association in the spatiotemporal relationship feature map, multi-angle trajectory analysis is performed on the candidate source tracing path set to obtain the preliminary source tracing trajectory of the candidate source tracing path set.

[0162] The preliminary tracing trajectory is resolving trajectory conflicts to obtain the complete tracing link of the underground utility tunnel;

[0163] Based on the complete tracing link, the abnormal feature information in the multi-dimensional feature descriptor is integrated to obtain the abnormal tracing information of the underground utility tunnel.

[0164] Key feature information is extracted from the multi-dimensional feature descriptor. This information includes the texture and shape features of the abnormal area, as well as the corresponding time acquisition information and spatial location information, ensuring that the extracted feature information can be completely mapped to the nodes and edges of the spatiotemporal relationship feature graph. In the spatiotemporal relationship feature graph, associated paths containing these features are filtered based on the extracted feature information. During the filtering process, each graph node and edge in the graph must be matched one by one. Only when the node attributes and edge feature identifiers in the path completely match the key features of the multi-dimensional feature descriptor is the path included in the candidate range. All the filtered matching paths are organized into a set, and each path needs to be labeled with corresponding feature matching items to ensure that the association between the path and the multi-dimensional feature descriptor is traceable, ultimately obtaining the candidate tracing path set for the underground utility tunnel.

[0165] From a temporal evolution perspective, for each path in the candidate source tracing path set, temporal edges and corresponding time nodes are extracted from the path. The connection relationships of the temporal edges are sorted out in chronological order, and the changes of abnormal features with time nodes are observed. For example, the density changes of a certain texture feature at different time nodes, and the size evolution of shape features over continuous time, are recorded to record the rules of feature evolution over time. From a spatial propagation perspective, spatial edges and corresponding spatial nodes are extracted from the path. Combined with the spatial structure data of underground utility tunnels, the adjacency relationships of spatial nodes and the propagation direction of spatial edges are analyzed to determine the spatial diffusion path of abnormal features among the tunnel components. For example, the process of propagating from a spatial node of a certain pipe segment to a spatial node of an adjacent pipe segment through a spatial edge is recorded to record the spatial propagation trajectory of the feature. From a feature association perspective, feature edges and corresponding feature vectors are extracted from the path. The similarity level of the feature edge annotation is compared with the consistency dimension of the feature vector to determine the feature association logic between different nodes. For example, the association basis for nodes with similar shape features connected by feature edges is recorded to record the feature association rules. The analysis results from the three perspectives are integrated into each candidate path, supplementing the path's temporal evolution pattern, spatial propagation trajectory, and feature association logic, ultimately yielding the preliminary source tracing trajectory of the candidate source tracing path set.

[0166] Each preliminary tracing trajectory was examined individually for its temporal evolution logic, spatial propagation logic, and feature association logic to identify conflicting sections. Temporal evolution conflicts manifest as a discrepancy between the temporal flow direction of a temporal edge and the actual data collection sequence; for example, the time interval labeled on a temporal edge may differ from the data collection time interval recorded in the multi-dimensional feature descriptor. Spatial propagation conflicts manifest as a mismatch between the propagation direction of a spatial edge and the spatial structure data of the underground utility tunnel; for example, the two spatial nodes connected by a spatial edge are not adjacent components in the actual tunnel layout. Feature association conflicts manifest as a mismatch between the similarity level of a feature edge and the actual consistent dimension of the feature vector; for example, a feature edge may be labeled with high similarity, but the consistent dimension of the feature vector is minimal. For each type of conflict, the path connection relationships of the conflicting sections were corrected by comparing the original feature information of the multi-dimensional feature descriptor with the spatial structure data of the underground utility tunnel. This included adjusting the temporal flow direction of the temporal edge, changing the connecting nodes of the spatial edge, and relabeling the similarity level of the feature edge to ensure that the corrected trajectory logic was completely consistent with the actual feature information and spatial structure. All corrected trajectories were integrated, and conflicting paths that could not be corrected were eliminated to obtain the complete tracing link of the underground utility tunnel.

[0167] Anomaly feature information corresponding to all nodes and edges in the complete traceability chain is extracted. Time nodes are associated with the temporal evolution data of anomaly features, spatial nodes with the spatial distribution data of anomaly features, temporal edges with the temporal variation patterns of features, spatial edges with the spatial propagation trajectories of features, and feature edges with the similarity information of features. This anomaly feature information is integrated according to the "time-space-feature" dimension. The time dimension organizes the evolution patterns and key time nodes of features, the spatial dimension organizes the propagation paths of features and associated utility tunnel components, and the feature dimension organizes the similarity associations and attribute changes of features, forming a structured information set. Each information item in the set needs to be labeled with the corresponding complete traceability chain node or edge to ensure the correlation between information and the chain, ultimately obtaining the anomaly traceability information of the underground utility tunnel.

[0168] The beneficial effects include: screening candidate tracing paths through key feature matching of multi-dimensional feature descriptors to ensure a strong correlation between the path and the abnormal features; analyzing the path from multiple perspectives such as temporal evolution, spatial propagation, and feature association to obtain a preliminary tracing trajectory, comprehensively covering the spatiotemporal and attribute association logic of abnormal features; correcting trajectory contradictions and eliminating invalid paths through conflict resolution to ensure the logical accuracy and reliability of the complete tracing link; and integrating multi-dimensional abnormal feature information based on the complete link to form structured abnormal tracing information, providing a comprehensive and accurate basis for the root cause location and development process tracing of underground utility tunnel failures, and assisting in the formulation of targeted failure handling solutions.

[0169] like Figure 2 The diagram shown is a functional module diagram of an underground utility tunnel fault data tracing and processing system based on big data, provided by an embodiment of the present invention.

[0170] The underground utility tunnel fault data tracing and processing system 100 based on big data described in this invention can be installed in an electronic device. Depending on the functions implemented, the system 100 may include a multi-source data acquisition module 101, a multi-scale feature enhancement module 102, an anomaly area identification module 103, a feature verification and description module 104, a spatiotemporal relationship construction module 105, and a trajectory tracing and analysis module 106. The modules described in this invention can also be referred to as units, which are a series of computer program segments that can be executed by the processor of an electronic device and perform a fixed function, stored in the memory of the electronic device.

[0171] In this embodiment, the functions of each module / unit are as follows:

[0172] The multi-source data acquisition module 101 is used to acquire multi-view monitoring image data of the underground utility tunnel;

[0173] The multi-scale feature enhancement module 102 is used to downsample the multi-view monitoring image data to obtain a differentiated resolution image of the underground utility tunnel, and to enhance the edge features of the differentiated resolution image according to the pyramid structure to obtain an enhanced feature map of the underground utility tunnel.

[0174] The abnormal region identification module 103 is used to identify abnormal regions in the enhanced feature map based on fault-related abnormal dynamic features in the enhanced feature map.

[0175] The feature verification and description module 104 is used to verify the authenticity of the abnormal area in the spatial structure of the underground utility tunnel and obtain the multi-dimensional feature descriptor of the enhanced feature map.

[0176] The spatiotemporal relationship construction module 105 is used to construct the spatiotemporal relationship feature map of the underground utility tunnel based on the spatiotemporal feature information in the multi-dimensional feature descriptor and the feature dimension correlation of the abnormal area.

[0177] The trajectory tracing analysis module 106 is used to perform multi-angle trajectory tracing on the multi-dimensional feature descriptor based on the spatiotemporal relationship feature map to obtain the abnormal tracing information of the underground utility tunnel.

[0178] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0179] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0180] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0181] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0182] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0183] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for tracing and processing fault data in underground utility tunnels based on big data, characterized in that, The method includes: S1. Acquire multi-view monitoring image data of underground utility tunnels; S2. Downsample the multi-view monitoring image data to obtain a differentiated resolution image of the underground utility tunnel, and perform edge feature enhancement on the differentiated resolution image according to the pyramid structure to obtain an enhanced feature map of the underground utility tunnel. S3. Based on the fault-related abnormal dynamic features in the enhanced feature map, identify the abnormal regions in the enhanced feature map, including: Extract the dynamic feature sequence that is temporally related to the fault from the enhanced feature map; Based on the abnormal mutation trend of image points in the dynamic feature sequence, abnormal feature points that deviate from normal changes in the enhanced feature map are identified. The spatial distribution density of the abnormal feature points is statistically analyzed to obtain candidate abnormal regions of the enhanced feature map; Spatial consistency verification is performed on the candidate abnormal regions to filter out the abnormal regions in the enhanced feature map; The step of identifying anomalous feature points that deviate from normal changes in the enhanced feature map based on the anomalous abrupt change trend of image points in the dynamic feature sequence includes: Based on the dynamic feature sequence, feature points in the enhanced feature map are connected in chronological order to obtain the feature change trajectory of the dynamic feature sequence; The deviation of the feature change trajectory is compared with a preset normal feature change pattern to obtain the degree of deviation of the feature change trajectory. The formula for calculating the degree of deviation is as follows: ； In the formula, The degree of deviation, The time window length of the feature change trajectory is, At the current time point, The first-order difference of the characteristic change trajectory at adjacent time points. This represents the first-order difference of the normal characteristic change pattern at adjacent time points. The second-order difference of the characteristic change trajectory at adjacent time points, The second-order difference of the normal feature change pattern at adjacent time points. The standard deviation of the first difference in the normal characteristic change pattern is given. The standard deviation of the second difference in the normal characteristic change pattern is given. These are the weighting coefficients of the first-order difference term. These are the weighting coefficients of the second-order difference term. The feature change trajectory at time point eigenvalues, The normal feature change pattern at time points The baseline eigenvalues; The deviation degree is compared with a threshold to obtain the abnormal feature points of the dynamic feature sequence; S4. Verify the authenticity of the abnormal area in the spatial structure of the underground utility tunnel to obtain the multi-dimensional feature descriptor of the enhanced feature map; S5. Based on the spatiotemporal feature information in the multi-dimensional feature descriptor and the correlation between the feature dimensions of the abnormal area, construct the spatiotemporal relationship feature map of the underground utility tunnel; S6. Based on the spatiotemporal relationship feature map, perform multi-angle trajectory tracing on the multi-dimensional feature descriptor to obtain the anomaly tracing information of the underground utility tunnel.

2. The method for tracing and processing underground utility tunnel fault data based on big data as described in claim 1, characterized in that, The acquisition of multi-view monitoring image data of underground utility tunnels includes: Collect continuous image sequence data from different perspectives in the underground utility tunnel to obtain the original image data of the underground utility tunnel; Video frames are extracted from the original image data to obtain a multi-view image frame set of the underground utility tunnel; Unify the image format of the multi-view image frame set to generate multi-view monitoring image data of the underground utility tunnel; The multi-view monitoring image data of the underground utility tunnel is obtained by verifying the view coverage integrity and temporal consistency of the multi-view monitoring image data.

3. The method for tracing and processing underground utility tunnel fault data based on big data as described in claim 1, characterized in that, The process of downsampling the multi-view monitoring image data to obtain a differentiated resolution image of the underground utility tunnel, and then performing edge feature enhancement on the differentiated resolution image according to a pyramid structure to obtain an enhanced feature map of the underground utility tunnel, includes: The multi-view monitoring image data is downsampled to obtain a differentiated resolution image of the underground utility tunnel. Based on the pyramid structure, image edges at different resolution levels in the differential resolution image are detected to obtain a multi-scale edge map of the underground utility tunnel; Based on the grayscale difference changes in the multi-scale edge map, the edge features of the multi-scale edge map are enhanced to obtain the edge-enhanced image of the underground utility tunnel; The edge-enhanced images are fused to obtain an enhanced feature map of the underground utility tunnel.

4. The method for tracing and processing underground utility tunnel fault data based on big data as described in claim 1, characterized in that, The verification of the authenticity of the abnormal region within the spatial structure of the underground utility tunnel yields a multi-dimensional feature descriptor for the enhanced feature map, including: The internal components and spatial layout information of the underground utility tunnel are used as spatial structure data. Spatial registration is performed between the abnormal region and the spatial structure data to obtain the spatial positional relationship between the abnormal region and the corresponding pipe gallery component; Based on the spatial location relationship, the consistency between the abnormal area and the structural topology relationship of the pipe gallery components is verified, and abnormal structural areas of components that conform to the spatial structural data are selected. The texture features, shape features, and dynamic change trend features of the abnormal region of the component structure are integrated to obtain the multi-dimensional feature descriptor of the enhanced feature map.

5. The method for tracing and processing underground utility tunnel fault data based on big data as described in claim 1, characterized in that, The step of constructing a spatiotemporal relationship feature map of the underground utility tunnel based on the correlation between the spatiotemporal feature information in the multi-dimensional feature descriptor and the feature dimensions of the abnormal region includes: The points representing different time points and different spatial locations in the multi-dimensional feature descriptor are used as graph nodes; The node paths connecting continuous time series in the multi-dimensional feature descriptor are used as time-series edges; The spatial edges are defined as the propagation paths of adjacent spatial location nodes in the multi-dimensional feature descriptor along the spatial dimension. The similarity of the feature vectors in the multi-dimensional feature descriptor is measured, and the feature edges of the multi-dimensional feature descriptor are established based on the measurement results and the degree of node clustering in the feature space. Based on the graph nodes, the temporal edges, the spatial edges, and the feature edges, a spatiotemporal relationship feature graph of the underground utility tunnel is constructed.

6. The method for tracing and processing underground utility tunnel fault data based on big data as described in claim 5, characterized in that, The formula for calculating the feature edge weights in the multi-dimensional feature descriptor is as follows: ； In the formula, For the nodes in the multi-dimensional feature descriptor With nodes The feature edge weights between them For the nodes in the multi-dimensional feature descriptor With nodes The cosine distance between the eigenvectors. The mean of the cosine distances between all node pairs in the multi-dimensional feature descriptor is given. The standard deviation of the cosine distance between all node pairs in the multi-dimensional feature descriptor is given. For the nodes in the multi-dimensional feature descriptor With nodes The space between them is Euclidean distance. For the nodes in the multi-dimensional feature descriptor With nodes The absolute time difference between them The preset spatial scale factor, The preset time scale factor, It is a natural constant. It is a natural constant.

7. The method for tracing and processing fault data in underground utility tunnels based on big data as described in claim 1, characterized in that, Based on the spatiotemporal relationship feature map, multi-angle trajectory tracing is performed on the multi-dimensional feature descriptors to obtain anomaly tracing information of the underground utility tunnel, including: The spatiotemporal association paths corresponding to the multi-dimensional feature descriptors in the spatiotemporal relationship feature map are used as the candidate tracing path set for the underground utility tunnel; From the perspectives of temporal evolution, spatial propagation, and feature association in the spatiotemporal relationship feature map, multi-angle trajectory analysis is performed on the candidate source tracing path set to obtain the preliminary source tracing trajectory of the candidate source tracing path set. The preliminary tracing trajectory is resolving trajectory conflicts to obtain the complete tracing link of the underground utility tunnel; Based on the complete tracing link, the abnormal feature information in the multi-dimensional feature descriptor is integrated to obtain the abnormal tracing information of the underground utility tunnel.

8. A big data-based system for tracing and processing fault data in underground utility tunnels, characterized in that, The system is used to implement the big data-based underground utility tunnel fault data tracing and processing method as described in claim 1, the system comprising: The multi-source data acquisition module is used to acquire multi-view monitoring image data of underground utility tunnels; A multi-scale feature enhancement module is used to downsample the multi-view monitoring image data to obtain a differentiated resolution image of the underground utility tunnel, and to enhance the edge features of the differentiated resolution image according to a pyramid structure to obtain an enhanced feature map of the underground utility tunnel. An abnormal region identification module is used to identify abnormal regions in the enhanced feature map based on fault-related abnormal dynamic features in the enhanced feature map. The feature verification and description module is used to verify the authenticity of the abnormal area in the spatial structure of the underground utility tunnel and obtain the multi-dimensional feature descriptor of the enhanced feature map. The spatiotemporal relationship construction module is used to construct the spatiotemporal relationship feature map of the underground utility tunnel based on the spatiotemporal feature information in the multi-dimensional feature descriptor and the correlation between the feature dimensions of the abnormal area. The trajectory tracing analysis module is used to perform multi-angle trajectory tracing on the multi-dimensional feature descriptor based on the spatiotemporal relationship feature map to obtain the abnormal tracing information of the underground utility tunnel.

Images