A Deep Learning-Based Method and System for Identifying Defects in Water Supply and Drainage Pipelines

Abstract

This invention provides a method and system for identifying defects in water supply and drainage pipe networks based on deep learning, relating to the field of deep learning technology. First, it acquires a set of visual detection data within the pipe network area to be identified, including consecutive frames of visual image units within the pipe and corresponding pipe mileage coordinates. Next, it performs visual texture primitive decomposition on the set of visual detection data to obtain a texture primitive distribution map and a spatial dependency relationship map. Then, it calls a pre-constructed pipe wall texture primitive evolution prediction network to perform temporal evolution prediction, generating a predicted texture primitive evolution map. Based on this predicted texture primitive evolution map, it performs defect pattern recognition to obtain the category identifier and expansion trend description information of potential defects. Finally, it generates a pipe network defect treatment instruction based on the category identifier and expansion trend description information of potential defects and sends it to the pipe network maintenance and scheduling terminal, achieving efficient, accurate, and forward-looking pipe network defect identification and treatment.

Description

Technical Field

[0001] This invention relates to the field of deep learning technology, and more specifically, to a method and system for identifying defects in water supply and drainage pipe networks based on deep learning. Background Technology

[0002] In the management and maintenance of water supply and drainage pipe networks, traditional methods for identifying defects mainly rely on manual inspections and periodic physical testing. Manual inspections require staff to enter the pipe network or observe from the ground to discover defects. These methods are not only inefficient, but also pose significant safety risks for pipe network areas that are deeply buried underground, in complex locations, or in harsh environments. Furthermore, they are difficult to comprehensively and promptly identify potential defects.

[0003] While regular physical inspections can improve the accuracy of disease detection to some extent, the inspection cycle is long, making it impossible to monitor the health of the pipeline network in real time. Furthermore, physical inspections typically require specialized equipment, which is costly, and comprehensive inspections are difficult for large-scale water supply and drainage pipeline systems. In addition, most existing image recognition-based pipeline disease detection methods only perform simple feature extraction and classification on currently acquired images, without considering the development process and trends of pipeline diseases, making it difficult to accurately predict the spread of diseases. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for identifying defects in water supply and drainage pipe networks based on deep learning, the method comprising: Obtain the pipe-in-pipe visual detection data set corresponding to the water supply and drainage network area to be identified. The pipe-in-pipe visual detection data set includes continuous frames of pipe-in-pipe visual image units and the pipe mileage positioning coordinates corresponding to each pipe-in-pipe visual image unit. The visual texture primitive decomposition process is performed on the visual detection data set inside the pipe to obtain the texture primitive distribution map of the pipe wall surface in the visual image unit inside the pipe and the spatial dependency relationship map between the texture primitives on the pipe wall surface. The pre-built pipe wall texture primitive evolution prediction network is invoked to perform temporal evolution prediction processing on the texture primitive distribution map and the spatial dependency relationship map, generating a predicted texture primitive evolution map of the pipe wall surface at future time points; Based on the predicted texture primitive evolution map, disease pattern recognition processing is performed to obtain the category identifier of potential diseases in the water supply and drainage network area to be identified and the description information of the expansion trend of potential diseases. Based on the category identifier and the extended trend description information, a pipeline network defect treatment instruction is generated, and the pipeline network defect treatment instruction is sent to the pipeline network maintenance and dispatch terminal.

[0005] Furthermore, embodiments of the present invention also provide a deep learning-based system for identifying defects in water supply and drainage pipe networks, comprising: A processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to execute the aforementioned deep learning-based method for identifying defects in water supply and drainage networks by executing the machine-executable instructions.

[0006] Based on the above, by acquiring a set of in-pipe visual detection data containing continuous frames of in-pipe visual image units and their corresponding pipeline mileage positioning coordinates, visual texture primitive decomposition processing is performed on the in-pipe visual detection data set to obtain a texture primitive distribution map of the pipe wall surface and a spatial dependency relationship map between texture primitives. This deeply mines the microscopic feature information of the pipe wall surface. A pre-constructed pipe wall texture primitive evolution prediction network is called to perform temporal evolution prediction processing on the texture primitive distribution map and spatial dependency relationship map, generating a predicted texture primitive evolution map for future time points. This achieves forward-looking prediction of the development trend of pipeline network defects, breaking through the limitation of traditional methods that can only detect current defects. Based on the predicted texture primitive evolution map, defect pattern recognition processing can accurately obtain the category identification of potential defects and the description information of expansion trends. Finally, based on this information, pipeline network defect treatment instructions are generated and sent to the pipeline network maintenance and scheduling terminal, realizing rapid response and accurate treatment of pipeline network defects, effectively improving the management efficiency and maintenance quality of water supply and drainage pipeline networks. Attached Figure Description

[0007] Figure 1 This is a schematic diagram of the execution flow of the deep learning-based method for identifying defects in water supply and drainage pipe networks provided in an embodiment of the present invention.

[0008] Figure 2 This is a schematic diagram of exemplary hardware and software components of the deep learning-based water supply and drainage network defect identification system provided in an embodiment of the present invention. Detailed Implementation

[0009] Figure 1 This is a flowchart illustrating a method for identifying defects in water supply and drainage pipe networks based on deep learning, provided in one embodiment of the present invention. A detailed description follows.

[0010] Step S110: Obtain the pipe visual detection data set corresponding to the water supply and drainage network area to be identified. The pipe visual detection data set includes pipe visual image units of consecutive frames and pipe mileage positioning coordinates corresponding to each pipe visual image unit.

[0011] In a scenario applied to the detection of defects in urban drainage pipelines, the area to be identified in the water supply and drainage network is a section of drainage pipeline in a certain city. Step S110 obtains the set of visual inspection data inside the pipe corresponding to the area to be identified. This set of visual inspection data includes consecutive frames of visual image units inside the pipe and the pipeline mileage positioning coordinates corresponding to each visual image unit. The visual image units inside the pipe are continuously captured by an industrial camera carried by a pipeline robot as it travels inside the pipe, and each image unit records the texture and structural information of the pipe wall surface. The pipeline mileage positioning coordinates are synchronously recorded by the robot's odometer sensor and correspond one-to-one with each image unit. The collection of all data has been authorized by the pipeline management department and complies with data privacy and security regulations.

[0012] Step S120: Perform visual texture primitive decomposition processing on the visual detection data set inside the pipe to obtain the texture primitive distribution map of the pipe wall surface in the visual image unit inside the pipe and the spatial dependency relationship map between the texture primitives on the pipe wall surface.

[0013] Step S121: Extract the pipeline mileage positioning coordinates corresponding to each pipeline visual image unit in the pipeline visual detection data set. Arrange all pipeline visual image units with the same pipeline mileage positioning coordinates in the order of acquisition time to construct a single-point visual image time series corresponding to each pipeline mileage positioning coordinate. For the first pipeline visual image unit in each single-point visual image time series, use a preset texture primitive perception filter group to perform multi-directional texture response calculation processing on the first pipeline visual image unit to generate a multi-directional texture response vector for each pixel position in the first pipeline visual image unit.

[0014] For each visual image unit within the pipe-in-pipe detection dataset, its pipeline mileage location coordinates are read from its metadata. A hash mapping table is established using the pipeline mileage location coordinates as the key. All visual image units within the pipe are traversed, and image units with the same pipeline mileage location coordinates are added to the same list. Then, the image units in each list are sorted in ascending order of their acquisition timestamps to form a single-point visual image time series corresponding to that pipeline mileage location coordinate. For the first frame of the visual image unit within each single-point visual image time series, it is input into a pre-designed texture primitive sensing filter bank. This texture primitive sensing filter bank consists of a set of two-dimensional Gaussian second-order derivative filters with different directions and scales. Specifically, a set of direction parameters is set, containing multiple equally spaced angle values. A set of scale parameters is set, containing multiple increasing standard deviation values. For each combination of direction and scale parameters, a two-dimensional Gaussian second-order derivative filter kernel is generated. This filter kernel is convolved with the image to obtain the texture response map at that direction and scale. For each pixel location, collect texture response values ​​at all directions and all scales to form a multidimensional vector. The dimension of this multidimensional vector is equal to the number of directions multiplied by the number of scales, which serves as the multi-directional texture response vector.

[0015] Step S122: Perform directional main peak detection processing on the multi-directional texture response vector of each pixel position, extract the texture main response direction and texture main response intensity of each pixel position, and perform region aggregation processing on adjacent pixels with similar texture main response directions based on the texture main response direction to generate the initial texture primitive region set in the first intra-tube visual image unit. Perform region shape regularization processing on each initial texture primitive region in the initial texture primitive region set, extract the region skeleton line of each initial texture primitive region, and perform equal curvature segment fitting processing on the region skeleton line along the extension direction of the region skeleton line to generate the tube wall surface texture primitive fragment corresponding to each initial texture primitive region.

[0016] For each pixel location output in step S121, the multi-directional texture response vector is traversed along the direction dimension to find the direction index with the largest component value in the multi-directional texture response vector. The angle value corresponding to this direction index is the dominant texture response direction at that pixel location, and the largest component value is the dominant texture response intensity. A direction angle difference threshold is set. All pixels are traversed, and for each pixel, the absolute value of the difference between the dominant texture response direction angle of its neighboring pixels and the dominant texture response direction angle of this pixel is checked. If the absolute value of the difference is less than the threshold, the neighboring pixels are included in the same candidate region as the current pixel. All adjacent pixels that meet the direction consistency condition are connected using the disjoint-set data structure algorithm to form multiple initial texture primitive regions. For each initial texture primitive region, hole filling processing is first performed to include the empty pixels inside the region. Then, the distance transformation method is used to calculate the distance from each pixel in the region to the region boundary, and the local maxima of distance are connected to form a line, forming the region skeleton line. This region skeleton line is a curve with a single pixel width. The skeleton line is segmented according to the curvature change: the curvature value of each point on the skeleton line is calculated, and the curvature calculation formula is the reciprocal of the radius of the circle determined by three adjacent points. Along the skeleton line direction, when the curvature value change exceeds a preset curvature change threshold, a segmentation point is set at that point. A segment of the skeleton line between adjacent segmentation points is treated as a curve segment. This curve segment is fitted into a quadratic curve using the least squares method, and the parameters of the fitted curve are recorded. Each fitted curve segment corresponds to a pipe wall surface texture primitive fragment.

[0017] Step S123: Input the remaining intra-pipe visual image units in the single-point visual image time series into the texture primitive perception filter group in sequence, and combine the position information of the pipe wall surface texture primitive fragments extracted from the previous intra-pipe visual image units to perform texture primitive tracking and extraction processing guided by the position of texture primitive fragments on the current intra-pipe visual image unit, generating a spatiotemporal position sequence of the pipe wall surface texture primitive fragments in each intra-pipe visual image unit. Based on the spatiotemporal position sequence of each pipe wall surface texture primitive fragment, extract the spatial offset vector of the pipe wall surface texture primitive fragment at adjacent time points, and extract the direction angle offset and offset distance change of the spatial offset vector to generate a motion offset descriptor for each pipe wall surface texture primitive fragment.

[0018] For each pipe-side visual image unit after the first frame in a single-point visual image time series, the spatial coordinates of all pipe wall surface texture primitive fragments extracted from the previous frame are first obtained. These spatial coordinates are projected onto the image coordinate system of the current frame as the predicted position of the corresponding texture primitive fragment in the current frame. A search window is defined around each predicted position, with the window size preset according to the pipe robot's movement speed. Within the search window of the current frame, steps S121 and S122 are executed to extract the pipe wall surface texture primitive fragments in the current frame. For each extracted texture primitive fragment, the offset between its centroid coordinates and the centroid coordinates of the corresponding fragment in the previous frame is calculated to obtain a spatial offset vector. The spatial offset vector is decomposed into a direction angle offset (the angle between the vector direction and the horizontal axis of the image) and an offset distance change (the magnitude of the vector). The direction angle offset and offset distance change are recorded in the motion record of the fragment in chronological order. The motion records of all time points are combined to form the motion offset descriptor of the pipe wall surface texture primitive fragment.

[0019] Step S124: Arrange all pipe wall surface texture primitive fragments at the same time point in the single-point visual image time series according to their spatial location attributes to generate a texture primitive distribution map of the pipe wall surface corresponding to each time point. The texture primitive distribution map records the geometric morphological attribute set and spatial location coordinates of each pipe wall surface texture primitive fragment.

[0020] For each time point in the single-point visual image time series, all extracted pipe wall surface texture primitive fragments at that time point are collected. A blank two-dimensional matrix with the same size as the original image is created. For each pipe wall surface texture primitive fragment, the pixel region covered by its fitted curve segment is marked in the two-dimensional matrix, and the set of geometric morphological attributes of the fragment is stored at the corresponding position in the matrix. The set of geometric morphological attributes includes: the length of the equivalent ellipse major axis, the length of the equivalent ellipse minor axis, the area of ​​the fragment, the perimeter of the fragment boundary, the mean and variance of the curvature of each point within the fragment, and the coefficients of the quadratic equation of the fitted curve. For each fragment, its centroid spatial coordinates are also recorded. The labels and attributes of all fragments constitute the texture primitive distribution map at that time point.

[0021] Step S125: For any two spatially adjacent pipe wall surface texture primitive fragments in the texture primitive distribution map, calculate the angle difference of the main response directions of the two pipe wall surface texture primitive fragments as the direction angle, calculate the relative distance and direction between the boundaries of the two pipe wall surface texture primitive fragment regions as the boundary relative position offset, and construct the spatial dependency relationship vector between the two pipe wall surface texture primitive fragments based on the direction angle and the boundary relative position offset.

[0022] In the texture primitive distribution map, it is determined whether the boundaries of two pipe wall surface texture primitive fragments share an edge or the shortest distance between their boundaries is less than a preset distance threshold. If the conditions are met, the two fragments are marked as spatially adjacent. For each pair of adjacent fragments, their respective texture principal response direction angles are obtained from step S122, and the absolute value of the difference between the two angles is calculated to obtain the direction angle. The two closest points on the boundaries of the two fragments are found, and the Euclidean distance between these two points is calculated as the relative distance. The unit direction vector from the nearest point of the first fragment to the nearest point of the second fragment is calculated as the relative direction. The relative distance and relative direction are combined into a two-dimensional vector. The direction angle and this two-dimensional vector are concatenated into a three-dimensional vector, which serves as the spatial dependency vector between the two pipe wall surface texture primitive fragments.

[0023] Step S126: Summarize and organize the spatial dependency vectors of all spatially adjacent pipe wall surface texture primitive fragment pairs in the texture primitive distribution map according to the spatial adjacency order to generate a spatial dependency graph between the pipe wall surface texture primitives. The spatial dependency graph includes the topological structure information of nodes and edges and the spatial dependency vector corresponding to each edge.

[0024] Create a graph data structure containing a set of nodes and a set of edges. Traverse all pipe wall surface texture primitive fragments in the texture primitive distribution graph, creating a node for each fragment. The node identifier is the fragment's unique number, and the node attributes include the set of geometric attributes recorded in step S124. Traverse all adjacent fragment pairs, creating an undirected edge between the corresponding two nodes. The edge's attributes are the spatial dependency vector calculated in step S125. Number the nodes according to their spatial position in the image, serialize and store the graph structure, and obtain the spatial dependency graph.

[0025] Step S127: Merge the texture primitive distribution map and spatial dependency graph corresponding to each pipeline mileage positioning coordinate to form the output data of the visual texture primitive decomposition processing. The texture primitive distribution map in the output data contains the complete geometric morphological attributes of the texture primitive fragments on the pipe wall surface at the pipeline mileage positioning coordinate, and the spatial dependency graph in the output data contains the complete spatial constraint relationship between the texture primitive fragments on the pipe wall surface at the pipeline mileage positioning coordinate.

[0026] For each pipeline mileage location coordinate, the texture primitive distribution map generated in step S124 and the spatial dependency relationship map generated in step S126 are associated and stored. The texture primitive distribution map records the geometric morphology information of the texture primitives on the pipe wall surface at that mileage location, and the spatial dependency relationship map records the spatial layout and mutual constraints between these texture primitives. Together, they constitute the visual texture primitive decomposition output data for that mileage location.

[0027] Step S130: Call the pre-built pipe wall texture primitive evolution prediction network to perform temporal evolution prediction processing on the texture primitive distribution map and the spatial dependency relationship map, and generate the predicted texture primitive evolution map of the pipe wall surface at future time points.

[0028] Step S131: Extract the set of geometric morphological attributes for each texture primitive fragment on the pipe wall surface from the texture primitive distribution map, and encode the set of geometric morphological attributes into a fixed-dimensional texture primitive geometric embedding vector; simultaneously, extract the spatial position coordinates for each texture primitive fragment on the pipe wall surface from the texture primitive distribution map, and encode the spatial position coordinates into a fixed-dimensional texture primitive position embedding vector; perform vector concatenation processing on the texture primitive geometric embedding vector and the texture primitive position embedding vector for each texture primitive fragment on the pipe wall surface to generate a texture primitive comprehensive embedding vector corresponding to each texture primitive fragment on the pipe wall surface, wherein the texture primitive comprehensive embedding vector contains a joint representation of the geometric morphological information and spatial position information of the texture primitive fragment on the pipe wall surface.

[0029] For each texture primitive fragment on the pipe wall surface in the texture primitive distribution map, all values ​​in its geometric attribute set are arranged in a fixed order to form an original feature vector. This original feature vector is input into the first fully connected layer, which maps it to a fixed-dimensional (e.g., 128-dimensional) space and outputs a texture primitive geometric embedding vector. Simultaneously, the centroid space coordinates (x-coordinate and y-coordinate values) of this fragment are input into a second fully connected layer, which maps it to the same fixed-dimensional space and outputs a texture primitive position embedding vector. These two embedding vectors are concatenated along the vector dimension direction to generate a 256-dimensional texture primitive composite embedding vector.

[0030] Step S132: Input the spatial dependency vector between each pair of pipe wall surface texture primitive fragments in the spatial dependency relationship diagram into a preset spatial relationship encoding module, and perform nonlinear mapping processing on the spatial dependency vector through the spatial relationship encoding module to generate a fixed-dimensional spatial relationship embedding vector.

[0031] The spatial relation encoding module is a multilayer perceptron with the following structure: The input layer receives a 3D spatial dependency vector. The first hidden layer contains 64 neurons, employing a modified linear unit activation function (MRU) for linear transformation and nonlinear activation: the output of the first hidden layer equals the weight matrix of the first hidden layer multiplied by the input vector plus the bias vector, then the MRU is applied. The second hidden layer contains 32 neurons, also employing the MRU. The output layer outputs a 16-dimensional spatial relation embedding vector. For each edge in the spatial dependency graph, its spatial dependency vector is input into the multilayer perceptron, computed through forward propagation, and output as a 16-dimensional vector, serving as the spatial relation embedding vector.

[0032] Step S133: Using the texture primitive composite embedding vector as the graph node feature and the spatial relationship embedding vector as the graph edge feature, construct a texture primitive spatial dependency graph structure with the texture primitive fragments on the pipe wall surface as nodes. The initial feature of each node in the texture primitive spatial dependency graph structure is the texture primitive composite embedding vector of the corresponding texture primitive fragment.

[0033] Create a graph data structure where the number of nodes equals the number of texture primitive fragments on the pipe wall surface in the texture primitive distribution graph. For the i-th node, use its texture primitive synthesis embedding vector (256 dimensions) as the initial feature vector of that node. For each edge in the spatial dependency graph, create an edge between the corresponding two nodes in the newly created graph data structure, and use the 16-dimensional spatial dependency embedding vector calculated in step S132 as the feature vector of that edge. The final constructed graph serves as the texture primitive spatial dependency graph structure.

[0034] Step S134: Using the graph attention feature propagation module in the pipe wall texture primitive evolution prediction network, perform multiple graph attention message passing processes on the texture primitive spatial dependency graph structure. In each graph attention message passing process, calculate the message passing attention weight based on the spatial relationship embedding vector between the target node and its neighboring nodes, and aggregate the texture primitive comprehensive embedding vectors of the neighboring nodes into the texture primitive comprehensive embedding vector of the target node based on the message passing attention weight, thereby generating the context-aware texture primitive feature vector of each node.

[0035] The graph attention feature propagation module comprises multiple graph attention layers. In the first graph attention layer, for each target node i, the set of all its neighboring nodes j is obtained. The attention coefficient between target node i and neighboring nodes j is calculated, utilizing the feature vectors of node i, node j, and the feature vectors of the edges connecting i and j. Specifically, the feature vectors of node i, node j, and the edge feature vectors are concatenated and input into a single-layer feedforward neural network. After processing with the LeakyReLU activation function, the original attention score is obtained. Then, the original attention scores of all neighboring nodes j are Softmax normalized to obtain attention weights. The feature vectors of neighboring nodes j are weighted and summed using the attention weights to obtain the aggregated feature vector. The aggregated feature vector is residually concatenated with the original feature vector of node i, and then subjected to layer normalization to obtain the updated feature vector of node i. After propagation through multiple graph attention layers, the feature vector of each node incorporates information from its surrounding neighboring nodes, serving as a context-aware texture primitive feature vector.

[0036] Step S135: Input the context-aware texture primitive feature vector of each pipe wall surface texture primitive fragment and the motion offset descriptor of the pipe wall surface texture primitive fragment into the temporal state update module in the pipe wall texture primitive evolution prediction network. The historical motion state is encoded by the gated recurrent processing unit in the temporal state update module to generate an initial motion state vector.

[0037] The temporal state update module employs a gated recurrent unit (ROU) network. The sequence of motion offset descriptors for each pipe wall surface texture primitive fragment, arranged chronologically, is input into the ROU. At each time step, the ROU receives the current motion offset descriptor (including orientation angle offset and offset distance change) and the hidden state from the previous time step, and outputs the hidden state for the current time step. The ROU internally contains update and reset gates. The update gate controls how much of the hidden state from the previous time step is retained in the current time step, while the reset gate controls the degree of integration between the current input and the hidden state from the previous time step. After processing the sequence across all time steps, the hidden state from the last time step is taken as the initial motion state vector.

[0038] Step S136: Perform feature-gated fusion processing on the initial motion state vector and the context-aware texture primitive feature vector. Selectively fuse the initial motion state vector and the context-aware texture primitive feature vector through the gating mechanism of the gated loop processing unit to generate a fused time-dependent texture primitive dynamic evolution feature vector. Input the fused time-dependent texture primitive dynamic evolution feature vector into the spatial coordinate regression branch of the pipe wall texture primitive evolution prediction network. Perform spatial position offset regression calculation on the texture primitive dynamic evolution feature vector through the spatial coordinate regression branch to generate the predicted spatial position offset of each pipe wall surface texture primitive fragment at future time points.

[0039] The initial motion state vector obtained in step S135 is fused with the context-aware texture primitive feature vector obtained in step S134. The fusion method employs a gating mechanism: first, a gating coefficient is calculated, which is obtained by linearly transforming the initial motion state vector and the context-aware texture primitive feature vector and then passing it through the Sigmoid function. Then, the gating coefficient is multiplied by the context-aware texture primitive feature vector, and then (1 minus the gating coefficient) is multiplied by the initial motion state vector to obtain the fused feature vector, which serves as the dynamic evolution feature vector of the texture primitive. This vector is input into the spatial coordinate regression branch. The spatial coordinate regression branch is a multilayer perceptron containing three fully connected layers. The first fully connected layer maps the input vector to 128 dimensions, the second fully connected layer to 64 dimensions, and the third fully connected layer outputs a 2-dimensional vector, representing the predicted spatial position offset of the texture primitive fragment in the horizontal and vertical directions of the image at future time points, respectively.

[0040] Step S137: Simultaneously input the dynamic evolution feature vector of the texture primitive into the geometric morphology regression branch of the pipe wall texture primitive evolution prediction network, and perform geometric morphology parameter change regression calculation on the dynamic evolution feature vector of the texture primitive through the geometric morphology regression branch to generate the predicted geometric morphology attribute change of each pipe wall surface texture primitive fragment at future time points.

[0041] The texture primitive dynamic evolution feature vector generated in step S136 is simultaneously input into the geometric morphology regression branch. The geometric morphology regression branch is also a multilayer perceptron, with a structure similar to the spatial coordinate regression branch but a different output dimension. The first fully connected layer maps the input vector to 128 dimensions, the second fully connected layer to 64 dimensions, and the third fully connected layer outputs a K-dimensional vector, where K equals the number of parameters in the geometric morphology attribute set. Each component of this output vector represents the predicted change of the corresponding geometric morphology parameter at future time points, such as the change in the length of the equivalent ellipse's major axis, the change in the length of the equivalent ellipse's minor axis, the change in area, and the change in perimeter.

[0042] Step S138: Vector-add the predicted spatial position offset to the spatial position coordinates at the current time point to generate the predicted spatial position coordinates of each pipe wall surface texture primitive fragment at the future time point; extract the value of each geometric morphology parameter from the geometric morphology attribute set at the current time point, and add it to the corresponding parameter change in the predicted geometric morphology attribute change to generate the predicted geometric morphology parameter set for each pipe wall surface texture primitive fragment at the future time point; based on the predicted spatial position coordinates and predicted geometric morphology parameter sets of all pipe wall surface texture primitive fragments, reconstruct the texture primitive spatial configuration of the pipe wall surface to generate the predicted texture primitive evolution diagram of the pipe wall surface at the future time point.

[0043] For each texture primitive fragment on the pipe wall surface, the predicted spatial position offset obtained in step S136 is added to the spatial position coordinates of the fragment at the current time point to obtain the predicted spatial position coordinates at the future time point. The predicted geometric attribute change obtained in step S137 is added to the corresponding parameters in the geometric attribute set of the fragment at the current time point to obtain the predicted geometric parameter set at the future time point. Then, based on the predicted spatial position coordinates and the predicted geometric parameter set of all fragments, the texture primitive spatial configuration of the pipe wall surface is reconstructed. The reconstruction process is as follows: For each fragment, based on the equivalent ellipse major axis length, minor axis length, and fitted curve equation coefficients in its predicted geometric parameter set, a curve segment corresponding to the fragment is generated. The curve segment is placed at the corresponding position in the image plane according to the predicted spatial position coordinates. The curve segments of all fragments are combined to form the texture primitive layout of the pipe wall surface at the future time point, which serves as the predicted texture primitive evolution diagram.

[0044] Step S140: Based on the predicted texture primitive evolution map, perform disease pattern recognition processing to obtain the category identifier of potential diseases in the water supply and drainage network area to be identified and the description information of the expansion trend of potential diseases.

[0045] Step S141: Obtain the set of predicted geometric parameters and predicted spatial coordinates of all pipe wall surface texture primitive fragments in the predicted texture primitive evolution diagram, and extract the texture primitive fragment area parameter and texture primitive fragment boundary perimeter parameter corresponding to each pipe wall surface texture primitive fragment from the set of predicted geometric parameters.

[0046] Traverse each pipe wall surface texture primitive fragment in the predicted texture primitive evolution diagram and read the area value and boundary perimeter value from its predicted geometric morphology parameter set.

[0047] Step S142: Calculate the ratio of the area parameter of the texture primitive fragment to the perimeter parameter of the texture primitive fragment boundary, and generate the area-perimeter ratio parameter corresponding to each texture primitive fragment on the pipe wall surface. The area-perimeter ratio parameter is used to characterize the compactness of the texture primitive fragment on the pipe wall surface. Extract the centroid coordinates of each texture primitive fragment on the pipe wall surface from the predicted spatial position coordinates, and combine the area-perimeter ratio parameter and the centroid coordinates of the texture primitive fragment into a feature vector to construct a compactness spatial distribution vector for each texture primitive fragment on the pipe wall surface.

[0048] For each texture primitive fragment on the pipe wall surface, its area value is divided by its boundary perimeter value to obtain the area-to-perimeter ratio. The area-to-perimeter ratio is combined with the centroid x-coordinate and centroid y-coordinate of the fragment to form a three-dimensional vector, which serves as the compactness spatial distribution vector.

[0049] Step S143: Input the compactness spatial distribution vector of all pipe wall surface texture primitive fragments in the predicted texture primitive evolution diagram into the preset spatial density clustering processing module, and perform spatial neighborhood density calculation processing on the compactness spatial distribution vector through the spatial density clustering processing module to generate a density distribution map of the compactness spatial distribution vector.

[0050] The preset spatial density clustering processing module employs a density-based spatial clustering algorithm. Neighborhood radius and minimum neighbor count parameters are set. For each compact spatial distribution vector, the number of other vectors within its neighborhood radius is calculated. If this number is greater than or equal to the minimum neighbor count, the vector is marked as a core point. Starting from the core point, all density-reachable vectors are grouped into the same cluster. Each cluster corresponds to a region with high density. The density information of all clusters is projected onto a two-dimensional space to generate a density distribution map.

[0051] Step S144: Based on the density distribution gradient change of the compactness spatial distribution vector in the density distribution map, extract the boundaries of continuous spatial regions where the density value exceeds a preset density threshold, and divide the region enclosed by the boundaries of the continuous spatial regions into compactness anomalous clustering regions. Within the compactness anomalous clustering regions, extract the texture principal response direction angle from the predicted geometric morphology parameter set of all pipe wall surface texture primitive fragments, and calculate the difference value of the texture principal response direction angle between any two spatially adjacent pipe wall surface texture primitive fragments within the compactness anomalous clustering regions.

[0052] In the density distribution map, the gradient (first derivative) of the density value at each pixel is calculated. The boundaries of regions where the density gradient direction points towards the direction of increasing density are identified. These boundaries enclose continuous spatial regions where the density value exceeds a preset density threshold; these regions are marked as compactness anomalous clustering regions. For each pipe wall surface texture primitive fragment within the compactness anomalous clustering region, its principal texture response direction angle is read from the predicted geometric morphology parameter set. For any two adjacent fragments within the region, the absolute value of the difference between their principal texture response direction angles is calculated as the direction angle difference value.

[0053] Step S145: Mark the adjacent pipe wall surface texture primitive fragments whose texture main response direction angle difference value exceeds the preset diffusion judgment threshold as direction disorder boundary fragments. Determine the area surrounded by the direction disorder boundary fragments in the compactness abnormal aggregation area as the potential disease core area. Extract the spatial position coordinate sequence of all direction disorder boundary fragments in the potential disease core area. Perform polygon fitting processing on the spatial position coordinate sequence to generate the geometric boundary description polygon of the potential disease core area.

[0054] A diffusion determination threshold is set. For each pair of adjacent pipe wall surface texture primitive fragments within the compactness anomalous clustering region, if their directional angle difference exceeds the threshold, the adjacent relationship is marked as a directional disorder boundary, and the two fragments are marked as directional disorder boundary fragments. Within the compactness anomalous clustering region, the connected region enclosed by the directional disorder boundary fragments is found, and this connected region is taken as the potential disease core region. The spatial coordinate sequence of all directional disorder boundary fragments within this region is extracted, and the coordinate points are polygonally fitted using a convex hull algorithm or a concave hull algorithm to generate a closed polygon, which serves as the geometric boundary description polygon.

[0055] Step S146: Based on the boundary contour morphology attributes of the polygon describing the geometric boundary of the potential disease core area and the numerical distribution characteristics of the compactness spatial distribution vector of the pipe wall surface texture primitive fragments within the potential disease core area, perform pattern similarity retrieval and matching processing in a preset disease category standard pattern library to obtain a disease category identifier that matches the potential disease core area.

[0056] The pre-defined standard pattern library for disease categories stores standard patterns for various disease categories. Each pattern includes standard boundary contour morphological attributes (such as roundness, rectangularity, and concavity / convexity) and statistical features of the standard compactness spatial distribution vector (such as mean, variance, and skewness). The boundary contour morphological attributes of the polygon describing the geometric boundary of the potential disease core area are extracted, including boundary perimeter, area, roundness (area multiplied by 4 times pi divided by the square of the perimeter), and rectangularity (area divided by the area of ​​the circumscribed rectangle). The numerical distribution characteristics of the compactness spatial distribution vector of all pipe wall surface texture primitive fragments within the potential disease core area are extracted, including the mean and variance of each component. These features are compared one by one with the standard patterns in the pattern library, and the Euclidean distance between the feature vectors is calculated. The disease category identifier corresponding to the pattern with the smallest distance is taken as the matching result.

[0057] Step S147: Extract the coordinates of the geometric center point of the geometric boundary description polygon of the potential disease core area. Starting from the coordinates of the geometric center point, extract the compactness spatial distribution vector of the pipe wall surface texture primitive fragments adjacent to the potential disease core area. Calculate the numerical difference between the compactness spatial distribution vector of the adjacent pipe wall surface texture primitive fragments and the compactness spatial distribution vector of the pipe wall surface texture primitive fragments within the potential disease core area. Using the spatial gradient descent direction of the numerical difference as the expansion direction, combine the predicted texture primitive evolution map of multiple future time points predicted by the pipe wall texture primitive evolution prediction network to track the evolution sequence of the geometric boundary description polygon of the potential disease core area at continuous prediction time points, and generate expansion trend description information containing the expansion direction vector sequence and the expansion rate change sequence.

[0058] Calculate the coordinates of the geometric center point of the polygon describing the geometric boundary. These coordinates are the arithmetic mean of the coordinates of all vertices of the polygon. Starting from the geometric center point, search radially outwards for texture primitive fragments on the pipe wall surface adjacent to (i.e., bordering) the core area of ​​the potential disease. Extract the compactness spatial distribution vectors of these adjacent fragments. Calculate the Euclidean distance between the compactness spatial distribution vector of each adjacent fragment and the mean of the compactness spatial distribution vectors of the fragments within the core area, using this distance as the numerical dissimilarity. The direction with the largest numerical dissimilarity is the expansion direction, and the rate of change of numerical dissimilarity with distance along this expansion direction is the spatial gradient. Use this expansion direction as the predicted expansion direction. For the predicted texture primitive evolution maps of multiple future time points generated in step S138, extract the geometric boundary description polygons of the core area of ​​the potential disease at each time point. Calculate the area change and outward expansion distance of the boundary polygons between adjacent time points to obtain the expansion rate. Arrange the expansion directions (represented as direction vectors) of all time points in chronological order to obtain the expansion direction vector sequence; arrange the expansion rates of all time points in chronological order to obtain the expansion rate change sequence. Together, they constitute the extended trend description information.

[0059] Step S150: Generate a pipeline network defect treatment instruction based on the category identifier and the extended trend description information, and send the pipeline network defect treatment instruction to the pipeline network maintenance and dispatch terminal.

[0060] Step S151: Obtain the preset treatment strategy knowledge base corresponding to the category identifier, retrieve the standard treatment process code that has a mapping relationship with the category identifier from the preset treatment strategy knowledge base, and extract the standard treatment resource requirement list associated with the standard treatment process code from the preset treatment strategy knowledge base.

[0061] The pre-defined treatment strategy knowledge base is a relational database containing multiple tables. The first table records the mapping relationship between disease category identifiers and standard treatment process codes. Based on the disease category identifier obtained in step S146, the corresponding standard treatment process code is queried from the first table. The second table records the standard treatment resource requirement list corresponding to each standard treatment process code, including the required number of personnel, equipment models, material types and quantities, and estimated working hours. Based on the queried standard treatment process code, the corresponding standard treatment resource requirement list is extracted from the second table.

[0062] Step S152: Analyze the expansion direction vector sequence and expansion rate change sequence in the expansion trend description information, extract the main expansion direction vector corresponding to the current time from the expansion direction vector sequence, and extract the expansion rate value corresponding to the current time from the expansion rate change sequence.

[0063] Read the first element of the expansion direction vector sequence from the expansion trend description information as the main expansion direction vector at the current moment, and read the first element of the expansion rate change sequence as the expansion rate value at the current moment.

[0064] Step S153: Obtain the geometric center coordinates of the potential disease core area determined in the disease pattern recognition process. Using the geometric center coordinates as the spatial reference starting point, along the spatial direction indicated by the main vector of the expansion direction, and using the expansion rate value as the expansion distance per unit time, perform spatial range extrapolation calculation to generate the predicted spatial coverage range of the potential disease core area in multiple future unit time periods. Map the predicted spatial coverage range onto the pipeline axis to generate the starting coordinates and ending coordinates of the pipeline mileage interval affected by the disease corresponding to each unit time period.

[0065] The geometric center coordinates of the core area of ​​the potential defect are used as the starting point for diffusion. Let the current time be T0, and the future time point T1 be one unit time after T0. The predicted spatial coverage is a semi-circular or fan-shaped area centered on the geometric center coordinates, with a radius equal to the expansion rate multiplied by the unit time length, along the main vector direction of the expansion. Projecting this predicted spatial coverage onto the pipeline's central axis yields the corresponding pipeline mileage coordinate intervals, i.e., the starting and ending coordinates of the pipeline mileage interval affected by the defect. Using the same method, the predicted spatial coverage for multiple future time units such as T2 and T3 is calculated, obtaining the pipeline mileage interval affected by the defect for each time unit.

[0066] Step S154: Calculate the change in the starting coordinate of the pipeline mileage interval affected by the defect between adjacent time units, and simultaneously calculate the change in the ending coordinate of the pipeline mileage interval affected by the defect between adjacent time units to generate the interval length change gradient between adjacent time units. Compare the interval length change gradient with multiple preset emergency level judgment intervals to determine the emergency level judgment interval to which the interval length change gradient belongs, and use the emergency level corresponding to the emergency level judgment interval as the emergency level of the emergency response.

[0067] Calculate the change in the starting coordinate of the (i+1)th time unit by subtracting the starting coordinate of the ith time unit from the starting coordinate of the ith time unit. Similarly, calculate the change in the ending coordinate. Add the absolute values ​​of these two changes or take the maximum value as the interval length change gradient. Multiple urgency level determination intervals are preset, each interval corresponding to a specific urgency level. Compare the calculated interval length change gradient with each of these determination intervals to determine the corresponding interval and thus the urgency level.

[0068] Step S155: Based on the urgency level of the emergency response and the resource items in the standard response resource requirement list, perform resource scheduling priority matching processing, assign resource allocation time sequence weights to each resource item in the standard response resource requirement list according to the urgency level, and generate a response resource scheduling scheme that includes resource allocation time sequence arrangements.

[0069] The higher the urgency level, the higher the priority of resource allocation. Based on the urgency level, each resource item in the standard disposal resource requirement list is assigned a resource allocation sequence weight; the higher the urgency level, the greater the weight. Then, the resource items are sorted from highest to lowest weight to determine the order of resource allocation. Combined with the estimated resource transportation time, a disposal resource scheduling plan is generated, which includes the resource allocation sequence arrangement (i.e., the start time and estimated arrival time of each resource).

[0070] Step S156: Perform data fusion and encapsulation processing on the standard treatment process code and the treatment resource scheduling scheme, and attach the starting coordinates and ending coordinates of the pipeline mileage interval affected by the disease corresponding to each unit time period to the encapsulated data structure. At the same time, attach the pipeline mileage positioning coordinates to generate a pipeline disease treatment instruction containing treatment process identifier, treatment resource list, treatment pipeline mileage interval and treatment response sequence, and send it to the target pipeline maintenance scheduling terminal.

[0071] Create a data structure containing the following fields: Treatment Process Code field, storing the standard treatment process code found in step S151; Treatment Resource List field, storing the treatment resource scheduling plan generated in step S155; Treatment Pipeline Mileage Range field, storing the start and end coordinates of the pipeline mileage range affected by the defect for each unit time period generated in step S153; and Treatment Response Sequence field, storing the urgency level and resource allocation sequence for each unit time period. Serialize this data structure into JSON or binary format and send it to the target pipeline maintenance and scheduling terminal over the network.

[0072] Step S210: After acquiring the in-tube visual detection data set, perform in-tube illumination distribution estimation processing on each in-tube visual image unit in the in-tube visual detection data set. The in-tube illumination distribution estimation processing includes extracting the pixel brightness distribution histogram of the in-tube visual image unit and estimating the main direction angle parameter of the light source and the light source intensity attenuation coefficient based on the distribution pattern of the pixel brightness distribution histogram.

[0073] Step S210 is executed after step S110 and before step S120. For each intra-tube visual image unit in the intra-tube visual detection dataset, it is first converted from the red-green-blue color space to the grayscale color space to obtain a grayscale image. The number of pixels appearing at each brightness level in the grayscale image is counted to obtain a pixel brightness distribution histogram. The horizontal axis of this pixel brightness distribution histogram represents the brightness level, and the vertical axis represents the number of pixels. The distribution pattern of this histogram is analyzed: if the pixel brightness distribution histogram shows a bimodal distribution, the brighter peak corresponds to the light source illumination area, and the darker peak corresponds to the backlight area; the difference in brightness between the two peaks reflects the directionality of the light source. Based on the positions of the two peaks, the principal direction angle parameter of the light source is calculated: the image plane is divided into multiple fan-shaped regions, and the average brightness of the pixels in each fan-shaped region is calculated. The center direction of the fan-shaped region with the highest average brightness is the principal direction angle of the light source. At the same time, the average brightness of pixels at different distances along the principal direction angle of the light source is calculated sequentially, and a curve of brightness decay with distance is fitted to obtain the light source intensity decay coefficient.

[0074] Step S211: Construct an intra-tube illumination distribution field model using the main direction angle parameter of the light source and the intensity attenuation coefficient of the light source. The intra-tube illumination distribution field model is used to describe the illumination intensity value at each pixel position in the intra-tube visual image unit.

[0075] Using the image center as the origin, define the unit direction vector corresponding to the principal direction angle of the light source. For any pixel location in the image, calculate the distance from that pixel location to the incident point of the light source along the principal direction of the light source. Based on the light source intensity attenuation coefficient obtained in step S210, construct a functional relationship between light intensity and distance attenuation, such as an exponential attenuation function or a power-law attenuation function. Apply this function to the entire image plane, so that each pixel location calculates a light intensity value based on its distance along the principal direction of the light source. The light intensity values ​​of all pixel locations constitute a light distribution field matrix with the same size as the original image, i.e., the light distribution field model within the tube.

[0076] Step S212: Calculate the illumination deviation ratio between the illumination intensity value and the standard reference illumination intensity value of each pixel position in the visual image unit of the tube according to the illumination distribution field model inside the tube, generate the illumination compensation coefficient for each pixel position, and use the illumination compensation coefficient to perform pixel-by-pixel multiplication compensation processing on the original pixel value of the corresponding pixel position in the visual image unit of the tube to generate a visual image unit with enhanced illumination consistency inside the tube.

[0077] A standard reference illumination intensity value is set, representing the target brightness under ideal uniform illumination conditions. For each pixel location in the image, the illumination intensity value at that location is obtained from the tube illumination distribution field model constructed in step S211. The ratio of this illumination intensity value to the standard reference illumination intensity value is calculated to obtain the illumination compensation coefficient. The original red, green, and blue channel pixel values ​​at this pixel location are multiplied by the illumination compensation coefficient to obtain the compensated pixel value. After traversing all pixel locations, a tube visual image unit with enhanced illumination consistency is generated.

[0078] Step S213: Replace the original intra-tube visual image unit in the intra-tube visual detection data set with the illumination-enhanced intra-tube visual image unit to form an illumination-enhanced intra-tube visual detection data set, and use the illumination-enhanced intra-tube visual detection data set for the visual texture primitive decomposition processing.

[0079] The illuminated visual image units generated in step S212 are written back to the intra-tube visual detection data set, replacing the original intra-tube visual image units at the corresponding positions. The updated data set serves as the illuminated intra-tube visual detection data set. Subsequent steps S121 to S127 all use this illuminated intra-tube visual detection data set as input.

[0080] Step S310: In the process of constructing the pipe wall texture primitive evolution prediction network, multiple sets of historical data of pipe wall defects are obtained at different collection periods. Each set of historical data of pipe wall defects contains the sequence of pipe wall visual image units corresponding to that collection period.

[0081] Multiple sets of historical visual inspection data of the same pipe segment collected at different time points are read from the historical database. Each set of data corresponds to a collection period and includes a sequence of visual image units collected during that period, as well as the pipeline mileage location coordinates corresponding to each image unit. The above historical data covers the complete process from the health status of the pipe segment to the occurrence and development of defects.

[0082] Step S311: Perform the visual texture primitive decomposition process on each set of historical visual detection data within the tube to generate a texture primitive distribution map and spatial dependency map corresponding to each acquisition period. Use the texture primitive distribution map of adjacent acquisition periods as the evolution prediction input sample and the texture primitive distribution map of the next acquisition period as the evolution prediction supervision label to construct a texture primitive evolution training sample set.

[0083] For each set of historical visual detection data within the tube, steps S120 to S127 are performed to generate a texture primitive distribution map and a spatial dependency map corresponding to that acquisition period. For two temporally adjacent acquisition periods, the texture primitive distribution map from the earlier period is used as the input sample for evolution prediction, and the texture primitive distribution map from the later period is used as the supervision label for evolution prediction. The input samples and supervision labels are paired to form a set of training samples. The set of all training samples constitutes the texture primitive evolution training sample set.

[0084] Step S312: Input the evolution prediction input samples from the texture primitive evolution training sample set into the pipe wall texture primitive evolution prediction network to be trained, and perform forward propagation processing on the evolution prediction input samples through the pipe wall texture primitive evolution prediction network to be trained to generate a predicted texture primitive evolution output map.

[0085] The evolution prediction input samples are input into the pipe wall texture primitive evolution prediction network to be trained, and forward propagation calculations are performed in steps S131 to S138 to output the predicted texture primitive evolution output map.

[0086] Step S313: Calculate the spatial position offset error of each pipe wall surface texture primitive fragment between the predicted texture primitive evolution output map and the evolution prediction supervision label, sum the spatial position offset errors of all pipe wall surface texture primitive fragments, and generate the texture primitive position offset loss function value.

[0087] For each pipe wall surface texture primitive fragment in the evolution prediction supervision label, find the corresponding matching fragment in the predicted texture primitive evolution output map (by nearest neighbor matching or identifier matching). Calculate the Euclidean distance between the centroid coordinates of the matching fragments as the spatial position offset error of that fragment. Sum the offset errors of all fragments to obtain the texture primitive position offset loss function value.

[0088] Step S314: Calculate the error in geometric morphological attribute change of each pipe wall surface texture primitive fragment between the predicted texture primitive evolution output map and the evolution prediction supervision label, sum the errors in geometric morphological attribute change of all pipe wall surface texture primitive fragments, and generate the texture primitive morphological change loss function value.

[0089] For each matched pair of pipe wall surface texture primitive fragments, calculate the absolute value of the difference between each corresponding parameter in the predicted geometric morphology parameter set and the supervised label geometric morphology parameter set. Sum the absolute values ​​of the differences of all parameters to obtain the geometric morphology attribute change error of the fragment. Add the errors of all fragments to obtain the texture primitive morphology change loss function value.

[0090] Step S315: Multiply the texture primitive position offset loss function value with a preset position loss weight coefficient to generate a weighted position loss value; multiply the texture primitive shape change loss function value with a preset shape loss weight coefficient to generate a weighted shape loss value; add the weighted position loss value and the weighted shape loss value to generate a total loss function value; use the total loss function value to perform gradient backpropagation on the pipe wall texture primitive evolution prediction network to be trained; calculate the partial derivative of each network weight parameter in the pipe wall texture primitive evolution prediction network to be trained with respect to the total loss function value; and perform gradient descent update processing on each network weight parameter according to the partial derivative to obtain the updated network weight parameters.

[0091] Set positional loss weights and morphological loss weights, with the sum of the two coefficients being 1. Multiply the texture primitive position offset loss function value by the positional loss weights to obtain the weighted positional loss value. Multiply the texture primitive morphological change loss function value by the morphological loss weights to obtain the weighted morphological loss value. Add the two together to obtain the total loss function value. Use automatic differentiation to calculate the partial derivative of the total loss function value with respect to each trainable weight parameter in the network. Then, according to the gradient descent update formula, subtract the learning rate multiplied by the corresponding partial derivative from each weight parameter to obtain the updated network weight parameters.

[0092] Step S316: Input the next evolution prediction input sample from the texture primitive evolution training sample set into the pipe wall texture primitive evolution prediction network after the network weight parameters have been updated. Repeat the forward propagation process, loss function value calculation process, and network weight parameter update process until the total loss function value converges to the preset training termination condition, and obtain the trained pipe wall texture primitive evolution prediction network. The training termination condition is that the numerical change of the total loss function value in the continuous training iteration process is less than the preset convergence change threshold. Save all the network weight parameters of the pipe wall texture primitive evolution prediction network when the training is completed as the network weight parameters of the pre-constructed pipe wall texture primitive evolution prediction network.

[0093] Input the next sample from the texture primitive evolution training sample set into the updated network, and repeat steps S312 to S315. Check the change in the total loss function value after each iteration. When the decrease in the total loss function value is less than the preset convergence threshold in several consecutive iterations (e.g., 10 iterations), stop training. Save all the weight parameters of the network at this time as the weight parameters of the trained pipe wall texture primitive evolution prediction network. This pipe wall texture primitive evolution prediction network can then be used for step S130.

[0094] Step S410: After generating the predicted texture primitive evolution map, obtain the pipeline laying material information database of the water supply and drainage network area to be identified, and extract the pipeline laying material type identifier corresponding to the pipeline mileage positioning coordinates and the material degradation rate parameter corresponding to the pipeline laying material type identifier from the pipeline laying material information database.

[0095] The pipeline laying material information database is retrieved from the pipeline network management information system. This database records the type of pipeline material used for each pipeline segment corresponding to its mileage coordinates, such as concrete pipe, clay pipe, polyethylene pipe, and steel pipe. Each material type is associated with a material degradation rate parameter, which reflects the aging rate of the pipeline material under normal operating conditions. Based on the current pipeline mileage coordinates to be identified, the corresponding pipeline laying material type identifier and material degradation rate parameter are retrieved from the pipeline laying material information database.

[0096] Step S411: Extract the geometric morphology parameter change amount corresponding to the material degradation rate parameter from the set of predicted geometric morphology parameter changes for each pipe wall surface texture primitive fragment in the predicted texture primitive evolution diagram. Compare the geometric morphology parameter change amount with the material degradation rate parameter and calculate the deviation between them. If the deviation exceeds the preset material degradation deviation tolerance threshold, apply material degradation constraint correction processing to the geometric morphology parameter change amount. The material degradation constraint correction processing includes adjusting the geometric morphology parameter change amount to its weighted average value with the material degradation rate parameter.

[0097] From the predicted texture primitive evolution map, extract the area change or wall thickness change from the predicted geometric morphology parameter change set of each pipe wall surface texture primitive fragment (these quantities are most relevant to material degradation). Compare this change with the material degradation rate parameter obtained in step S410 to calculate the deviation. A preset material degradation deviation tolerance threshold is set. If the deviation exceeds this threshold, it indicates that the predicted change does not conform to the physical laws of material degradation and needs to be corrected. The correction method is to take a weighted average of the original predicted geometric morphology parameter change and the material degradation rate parameter to obtain the corrected change. The weighting coefficient can be dynamically adjusted according to the prediction confidence level.

[0098] Step S412: Add the change in geometric morphology parameters after material degradation constraint correction to the geometric morphology parameter value corresponding to the current time point to generate the predicted value of the geometric morphology parameter after the correction of the texture primitive fragment on the pipe wall surface, and update its predicted geometric morphology parameter set.

[0099] For the corrected change in geometric morphology parameters, add it to the corresponding geometric morphology parameter value at the current time point to obtain the corrected predicted geometric morphology parameter value. Replace the original predicted value in the predicted geometric morphology parameter set with this corrected value to update the predicted geometric morphology parameter set for this segment.

[0100] Step S413: Using the corrected predicted geometric properties of all pipe wall surface texture primitive fragments and the original predicted geometric properties of pipe wall surface texture primitive fragments that do not exceed the deviation tolerance, replace the corresponding predicted geometric properties in the predicted texture primitive evolution diagram to generate a predicted texture primitive evolution diagram corrected for material degradation constraints.

[0101] Iterate through all pipe wall surface texture primitive fragments in the predicted texture primitive evolution diagram. For fragments corrected by steps S411 and S412, replace the original prediction set with the corrected set of predicted geometric morphology parameters; for fragments that do not exceed the deviation tolerance, keep the original prediction unchanged. The updated predicted texture primitive evolution diagram serves as the predicted texture primitive evolution diagram after material degradation constraint correction.

[0102] Step S414: Replace the original predicted texture primitive evolution map with the predicted texture primitive evolution map after the material degradation constraint correction, and use it as the input data for the disease pattern recognition processing.

[0103] In the defect pattern recognition process of step S140, the predicted texture primitive evolution map after material degradation constraint correction generated in step S413 is used instead of the original predicted texture primitive evolution map generated in step S138 as input. This ensures that the defect recognition results conform to the physical degradation law of the pipeline material.

[0104] Step S510: After generating the pipeline network defect treatment instruction and sending the pipeline network defect treatment instruction to the pipeline network maintenance and dispatching terminal, receive the treatment completion status report fed back by the pipeline network maintenance and dispatching terminal after performing the treatment operation. The treatment completion status report includes the treatment completion timestamp and the pipeline mileage positioning coordinate range covered by the treatment operation.

[0105] After completing the repair work, the pipeline maintenance dispatch terminal sends a repair completion status report to the system via the communication interface. This report includes a timestamp (date and time) indicating the repair completion, as well as the pipeline mileage coordinates (starting and ending mileage) covered by the repair operation.

[0106] Step S511: Obtain the set of visual detection data inside the pipe after the treatment operation based on the pipeline mileage positioning coordinate range covered by the treatment operation, and perform the visual texture primitive decomposition processing on the set of visual detection data inside the pipe after the treatment to generate the texture primitive distribution map and the spatial dependency relationship map after the treatment.

[0107] Based on the pipeline mileage and location coordinate range covered by the treatment operation, retrieve the set of in-pipe visual inspection data within that mileage range that was re-acquired after the treatment from the data storage. Perform steps S120 to S127 on this data set to generate a post-treatment texture primitive distribution map and a post-treatment spatial dependency map.

[0108] Step S512: Take the texture primitive distribution map corresponding to the water supply and drainage network area to be identified before processing as the first input, take the texture primitive distribution map after processing as the second input, input the texture primitive recovery effect comparison branch in the pipe wall texture primitive evolution prediction network, and extract the geometric shape recovery deviation of each pipe wall surface texture primitive fragment between the first input and the second input through the texture primitive recovery effect comparison branch.

[0109] A texture primitive recovery effect comparison branch is added to the pipe wall texture primitive evolution prediction network. The structure of this texture primitive recovery effect comparison branch is the same as that of the graph attention feature propagation module in step S134, but the input is two parallel texture primitive distribution maps (before and after treatment). This texture primitive recovery effect comparison branch performs feature alignment on the pipe wall surface texture primitive fragments at the same spatial location in the two distribution maps, and then calculates the change in geometric morphology parameters of each fragment before and after treatment. This change in geometric morphology parameters is the geometric morphology recovery deviation.

[0110] Step S513: Calculate the overall texture restoration degree index of the treatment operation coverage area based on the geometric shape recovery deviation of each pipe wall surface texture primitive fragment and the spatial coverage area of ​​the pipe wall surface texture primitive fragment in the texture primitive distribution map.

[0111] For each texture primitive fragment on the pipe wall surface, the geometric shape restoration deviation is multiplied by its spatial coverage area (fragment area) to obtain the weighted restoration deviation. The weighted restoration deviations of all fragments are summed and then divided by the total area covered by the processing operation to obtain the overall texture primitive restoration degree index. The closer this overall texture primitive restoration degree index is to 1, the better the restoration degree; the closer it is to 0, the worse the restoration degree.

[0112] Step S514: Associate and store the overall texture primitive recovery degree index with the pipeline mileage positioning coordinate range covered by the disposal operation in the disposal completion status report to generate disposal effect evaluation data record.

[0113] The calculated overall texture primitive restoration index is associated with and stored in the treatment completion status report, along with the pipeline mileage positioning coordinate range and treatment process code covered by the treatment operation, forming a treatment effect evaluation data record, which is then stored in the effect evaluation database.

[0114] Step S515: Extract the treatment process code and the corresponding overall texture primitive recovery degree index from the treatment effect evaluation data record, use the overall texture primitive recovery degree index as the treatment efficiency update parameter of the treatment process code, and perform weighted update processing on the historical treatment efficiency record corresponding to the treatment process code in the preset treatment strategy knowledge base to obtain the updated treatment strategy knowledge base.

[0115] The treatment process code and overall texture primitive restoration level index are retrieved from the treatment effect evaluation data records. In the preset treatment strategy knowledge base, the historical treatment effectiveness record corresponding to the treatment process code is located. This historical record records the values ​​of the overall restoration level index from multiple historical treatments. An exponentially weighted moving average method is used to weight and fuse the new overall texture primitive restoration level index with the historical values, updating the treatment effectiveness score of the treatment process code. After the update, the priority of the treatment process code in the knowledge base will be adjusted according to the new effectiveness score.

[0116] For example, the method may also include: Step S610: After obtaining the texture primitive distribution map, perform multi-scale morphological decomposition processing on the region contour defined by the geometric morphological attribute set of each pipe wall surface texture primitive fragment in the texture primitive distribution map. The multi-scale morphological decomposition processing includes performing layer-by-layer morphological peeling processing on the region contour using a preset multi-level morphological decomposition kernel to generate the sub-region contour set and its morphological attributes of the pipe wall surface texture primitive fragment at different morphological scales.

[0117] Step S610 is executed after step S124 and before step S130. For each pipe wall surface texture primitive fragment in the texture primitive distribution map, a binary mask image of its region contour is obtained. A set of multi-level morphological decomposition kernels is preset, each kernel being a circular structuring element with an increasing radius. Starting from the kernel with the smallest radius, a morphological erosion operation is performed on the region contour mask. Each erosion operation removes the boundary layer of the region contour, resulting in a new sub-region mask. The sub-region contour obtained after each erosion layer, as well as the morphological attributes such as area, perimeter, and roundness of the sub-region, are recorded. The sub-region contours and their attributes at all scales constitute the sub-region contour set of the fragment.

[0118] Step S611: Input the sub-region contour set and its morphological attributes at different morphological scales into the preset pipe wall texture primitive damage assessment model, and use the pipe wall texture primitive damage assessment model to fuse multi-scale morphological features to generate the damage assessment index of the pipe wall surface texture primitive fragment.

[0119] The pipe wall texture primitive damage assessment model is a multilayer perceptron. Its input layer receives the morphological attributes (area, perimeter, roundness, etc.) of sub-regions at each scale and concatenates these attributes into a long vector. The first hidden layer maps the input vector to a high-dimensional space, the second hidden layer performs feature cross-combination, and the output layer outputs a scalar value between 0 and 1, which is the damage assessment index. The closer the damage assessment index is to 1, the more severe the damage.

[0120] Step S612: Based on the damage assessment index, perform damage level classification on all pipe wall surface texture primitive fragments in the texture primitive distribution map to obtain the damage level identifier corresponding to each pipe wall surface texture primitive fragment.

[0121] Multiple damage level thresholds are preset, dividing the range of damage assessment index values ​​into multiple intervals. Each interval corresponds to a damage level (e.g., level 0 is no damage, level 1 is mild damage, level 2 is moderate damage, and level 3 is severe damage). Based on the interval in which the damage assessment index of each segment falls, a corresponding damage level label is assigned to it.

[0122] Step S613: The damage level identifier is associated with the geometric morphology attribute set of the corresponding pipe wall surface texture primitive fragment in the texture primitive distribution map as additional information, and an enhanced texture primitive distribution map containing the damage level identifier is generated.

[0123] In the data record of each texture primitive fragment on the pipe wall surface of the texture primitive distribution map, an additional field is added to store the damage level identifier of that fragment. The updated texture primitive distribution map serves as an enhanced texture primitive distribution map.

[0124] Step S614: Replace the original texture primitive distribution map with the enhanced texture primitive distribution map and use it as the input data of the pipe wall texture primitive evolution prediction network, so that the pipe wall texture primitive evolution prediction network can use the damage level identifier to constrain the evolution trend of the texture primitive fragments on the pipe wall surface during the temporal evolution prediction process.

[0125] When the pipe wall texture primitive evolution prediction network is invoked in step S130, the enhanced texture primitive distribution map generated in step S613 is used instead of the original texture primitive distribution map as input. In the graph attention feature propagation module of the network, the damage level identifier is used as an additional node feature, which participates in message passing and aggregation along with other features, so that the evolution prediction result is affected by the damage level. Fragments with high damage levels will show a faster morphological deterioration trend.

[0126] Step S710: After generating the predicted texture primitive evolution map, the pipe wall stress field inversion process is performed based on the predicted spatial position coordinate changes and predicted geometric morphology parameter changes of each pipe wall surface texture primitive fragment in the predicted texture primitive evolution map.

[0127] Step S710 is executed after step S138 and before step S140. For each pipe wall surface texture primitive fragment in the predicted texture primitive evolution diagram, its current spatial position coordinates and predicted spatial position coordinates are obtained, and the difference between the two constitutes a displacement vector field. At the same time, the geometric morphology parameters (such as the equivalent ellipse major axis length) and predicted geometric morphology parameters at the current time point are obtained, and the difference between the two constitutes a strain change field. The above displacement and strain information reflects the deformation mode of the pipe wall surface.

[0128] Step S711: The change in the predicted spatial position coordinates of each pipe wall surface texture primitive fragment at the predicted time point relative to its current spatial position coordinates is used as the predicted displacement field data at the pipe wall surface texture primitive fragment. This data is then used as the boundary condition of the pipe wall surface displacement field and input into the preset pipe wall mechanical constitutive inversion model. The stress field is solved in reverse through the pipe wall mechanical constitutive inversion model to generate the equivalent nodal stress tensor field of the pipe wall region where the pipe wall surface texture primitive fragment is located.

[0129] The displacement vector of each texture element fragment on the pipe wall surface obtained in step S710 is used as a boundary condition and input into the preset pipe wall mechanical constitutive inversion model. This model is based on the finite element method, discretizing the pipe wall surface into triangular mesh elements. The displacement at each mesh node is known (obtained by displacement interpolation of the texture element fragments). The model employs a linear elastic constitutive relation, where stress and strain satisfy Hooke's law. By solving the equilibrium equations, the stress tensor (containing normal and shear stress components) at each mesh node is derived. The stress tensors of all mesh nodes constitute an equivalent nodal stress tensor field.

[0130] Step S712: Extract the principal stress direction angle of each node position from the equivalent node stress tensor field, extract the principal response direction angle of the corresponding pipe wall surface texture primitive fragment from the texture primitive distribution map on which the spatial dependency graph is based, calculate the angle deviation between the two, and generate the stress texture direction deviation angle associated with each pipe wall surface texture primitive fragment.

[0131] For each mesh node, the stress tensor obtained in step S711 is subjected to eigenvalue decomposition to calculate the two principal stresses (maximum principal stress and minimum principal stress) and their corresponding directions. The direction angle of the maximum principal stress is taken as the principal stress direction angle of the node. The texture primitive fragment on the pipe wall surface at the location of the node is found from the texture primitive distribution map, and the principal response direction angle of the fragment is obtained. The absolute value of the difference between the principal stress direction angle and the principal response direction angle is calculated to obtain the stress-texture direction deviation angle.

[0132] Step S713: Extract the pipe wall surface texture primitive fragments whose stress texture direction deviation angle exceeds the preset stress texture mismatch threshold, and mark the extracted pipe wall surface texture primitive fragments as stress texture anomalous fragments. Perform region connectivity processing on spatially adjacent stress texture anomalous fragments to generate stress texture anomalous connected domains.

[0133] Set a stress texture mismatch threshold (e.g., 30 degrees). Traverse all texture primitive fragments on the pipe wall surface. If the stress texture direction deviation angle of a fragment is greater than the threshold, mark it as a stress texture anomalous fragment. Use a connected component analysis algorithm to cluster spatially adjacent (shared boundaries or corners) stress texture anomalous fragments into connected regions, with each connected region serving as a stress texture anomalous connected component.

[0134] Step S714: Perform ellipse fitting on the geometric boundary of the stress texture abnormal connected domain to generate the stress concentration elliptical region corresponding to the stress texture abnormal connected domain, and extract the major axis direction angle, minor axis direction angle and major axis-minor axis length ratio of the stress concentration elliptical region.

[0135] For each stress texture anomaly connected domain, the coordinate sequence of its boundary points is extracted. An ellipse is fitted to the boundary points using the least squares method, aiming to find an ellipse that minimizes the sum of the squared algebraic distances from all boundary points to the ellipse. The fitted ellipse yields five parameters: center point coordinates, major axis radius, minor axis radius, and rotation angle. The major axis angle (rotation angle), minor axis angle (rotation angle plus 90 degrees), and the ratio of the major axis radius to the minor axis radius are recorded as the major-minor axis length ratio.

[0136] Step S715: Take the angle of the major axis of the stress concentration elliptical region as the main direction of potential crack propagation, take the ratio of the major axis to the minor axis as the stress concentration degree factor, and take the coordinates of the geometric center point of the stress concentration elliptical region as the coordinates of the potential crack initiation point.

[0137] From the ellipse fitting results in step S714, the angle of the major axis direction is taken as the main direction of potential crack propagation, the ratio of the major axis to the minor axis length is taken as the stress concentration factor (the larger the ratio, the more significant the stress concentration), and the coordinates of the center point of the ellipse are taken as the coordinates of the potential crack initiation point.

[0138] Step S716: Input the potential crack propagation direction, the stress concentration factor, and the coordinates of the potential crack initiation point into a preset pipe wall crack propagation path prediction model. Use the pipe wall crack propagation path prediction model to generate a crack propagation path segment by segment along the potential crack initiation point coordinates, generating a pipe wall potential crack propagation path prediction result containing a sequence of crack path node coordinates.

[0139] The pipe wall crack propagation path prediction model is based on the Paris formula and the maximum circumferential stress criterion. Starting from the coordinates of the potential crack initiation point, an initial crack segment is generated along the main direction of potential crack propagation. Then, at each crack tip, the stress intensity factor is recalculated to determine the next direction of crack propagation. This process is repeated to gradually generate the crack path. The coordinates of the crack segment endpoints at each step are recorded in the path node coordinate sequence. The final generated crack path node coordinate sequence is the prediction result of the potential crack propagation path of the pipe wall.

[0140] Step S717: Overlay the predicted path of the potential crack in the pipe wall onto the predicted texture primitive evolution map to generate an enhanced predicted texture primitive evolution map containing crack propagation path annotations, and replace the original predicted texture primitive evolution map with the enhanced predicted texture primitive evolution map for the disease pattern recognition processing.

[0141] On the image plane of the predicted texture primitive evolution map, the crack path nodes generated in step S716 are connected sequentially to form a curve, and this curve is drawn with a striking color (such as red). The predicted texture primitive evolution map labeled with crack paths is used as the enhanced predicted texture primitive evolution map. In the disease pattern recognition processing of step S140, this enhanced predicted texture primitive evolution map is used instead of the original predicted texture primitive evolution map as input.

[0142] Step S718: In the disease pattern recognition process, the prediction result of the potential crack propagation path of the pipe wall is used as an additional input feature, and together with the boundary contour morphology attribute of the geometric boundary description polygon of the determined potential disease core area, it participates in the pattern similarity retrieval and matching process in the preset disease category standard pattern library to generate a disease category identifier that integrates stress field inversion information.

[0143] In the pattern similarity retrieval and matching process of step S146, in addition to using the geometric boundary to describe the boundary contour morphology attributes of the polygon and the numerical distribution characteristics of the compactness spatial distribution vector, the features of the crack path prediction results (such as the total crack length and the crack direction change rate) are also used as additional feature vectors. These features are then matched with the standard patterns in the disease category standard pattern library to obtain the disease category identifier that integrates stress field inversion information.

[0144] Step S719: Update the treatment process code in the pipeline network defect treatment instruction according to the defect category identifier of the fused stress field inversion information, replace the original treatment process code with the updated treatment process code, and generate a pipeline network defect treatment instruction based on the fused stress inversion analysis results.

[0145] In the pipeline defect treatment instruction generated in step S156, the original treatment process code is replaced with the treatment process code corresponding to the defect category identifier obtained in step S718, which incorporates the stress field inversion information. The updated instruction includes the results of the stress field inversion analysis, which can guide more targeted maintenance measures.

[0146] Step S810: After obtaining the texture primitive distribution map and the spatial dependency relationship map, a dynamic association map structure of pipe segment texture primitives is constructed using the pipe wall surface texture primitive fragments at each pipe mileage positioning coordinate as nodes; wherein, the adjacency relationship defined in the spatial dependency relationship map constitutes the spatial dependency edge between nodes within the same pipe mileage positioning coordinate; based on the spatial proximity relationship of texture primitive fragments in the texture primitive distribution map at different pipe mileage positioning coordinates, cross-point spatial dependency edges are established between nodes at adjacent pipe mileage positioning coordinates.

[0147] Step S810 is executed after step S127 and before step S130. For each pipeline mileage location coordinate, each pipe wall surface texture primitive fragment in its texture primitive distribution map is taken as a node. Within the same mileage coordinate, according to the adjacency relationship defined in the spatial dependency graph, edges are added between corresponding nodes as spatial dependency edges. For adjacent pipeline mileage location coordinates, the spatial distance between texture primitive fragments at the two mileages is calculated, and edges are added between fragment pairs with a distance less than a preset threshold as cross-point spatial dependency edges. All nodes and edges constitute the dynamic association graph structure of the pipe segment texture primitives.

[0148] Step S811: Extract the motion offset descriptor of the pipe wall surface texture primitive fragment at the same pipeline mileage positioning coordinate in the time series of the single-point visual image at consecutive time points, perform time series clustering on the direction angle offset in the motion offset descriptor, and generate the motion direction evolution mode identifier of the pipe wall surface texture primitive fragment.

[0149] For each pipe wall surface texture primitive segment, the direction angle offset sequence is extracted from the motion offset descriptor generated in step S123. This sequence is then clustered using the K-means clustering algorithm, with a preset number of three clusters, corresponding to the direction angle offsets concentrated around 0 degrees, around 90 degrees positive, and around 90 degrees negative, respectively. After clustering, the direction angle offset at each time point is assigned a category label. The frequency of the category label at all time points for this segment is counted, and the category with the highest frequency is taken as the motion direction evolution mode identifier for this segment.

[0150] Step S812: Extract the change in offset distance from the motion offset descriptor, perform time series trend decomposition on the change in offset distance, and generate the motion amplitude evolution trend component and motion amplitude periodic fluctuation component of the texture primitive fragment on the pipe wall surface.

[0151] The sequence of offset distance changes is extracted from the motion offset descriptor. A locally weighted regression method is used to decompose this sequence into trend and periodic components. The trend component reflects the long-term direction of change (increasing or decreasing) of the offset distance change, while the periodic component reflects the fluctuation pattern of the offset distance change over time. The trend and periodic components are stored separately.

[0152] Step S813: Perform feature cross-combination processing on the motion direction evolution mode identifier and the motion amplitude evolution trend component to generate a motion evolution mode feature vector corresponding to each pipe wall surface texture primitive segment. The motion evolution mode feature vector is used to characterize the long-term motion behavior mode of the pipe wall surface texture primitive segment.

[0153] The motion direction evolution modal identifier (discrete value) is converted into a one-hot encoded vector, which is then concatenated with the trend slope value of the motion amplitude evolution trend component to form a feature vector. This feature vector serves as the motion evolution pattern feature vector, used to characterize the long-term motion behavior pattern of the texture primitive fragment.

[0154] Step S814: Perform spatial coordination consistency analysis on the motion evolution mode feature vectors of all pipe wall surface texture primitive fragments within the same pipeline mileage positioning coordinates, calculate the cosine value of the vector angle between the motion evolution mode feature vectors of any two spatially adjacent pipe wall surface texture primitive fragments, and generate the motion evolution mode coordination degree matrix.

[0155] For two spatially adjacent texture primitive fragments within the same mileage coordinate system, calculate the cosine of the angle between their motion evolution pattern feature vectors. A cosine value close to 1 indicates a high degree of coordination between the two fragments' motion evolution patterns, a cosine value close to 0 indicates low coordination, and a negative cosine value indicates opposite motion directions. Fill a matrix with the cosine values ​​of all adjacent fragment pairs, where the row and column indices of the matrix represent the fragment numbers. This matrix serves as the motion evolution pattern coordination matrix.

[0156] Step S815: Perform spectral clustering analysis on the motion evolution mode coordination degree matrix, and divide the pipe wall surface texture primitive fragments with similar motion evolution mode feature vectors into the same motion coordination region, generating the pipe wall surface texture primitive motion coordination region division result.

[0157] Spectral clustering is performed on the motion evolution pattern coordination matrix. First, the Laplacian matrix is ​​calculated. Then, the eigenvectors corresponding to the first k smallest eigenvalues ​​of the Laplacian matrix are solved. These eigenvectors are combined into a matrix, and finally, K-means clustering is performed on the row vectors of this matrix to obtain the clustering results. Each cluster corresponds to a motion coordination region, and the texture primitive fragments within the region have similar motion behavior patterns.

[0158] Step S816: Extract the region boundary of each motion-coordinated region in the result of the motion-coordinated region division of the surface texture primitives of the pipe wall, perform spatial superposition and comparison processing on the region boundary and the edge line set of the pipe wall structure, and calculate the boundary overlap parameter between the region boundary and the edge line set of the pipe wall structure.

[0159] Extract the outer boundary contour of each region from the motion-coordinated region segmentation results. Obtain the set of edge lines of pipe wall structural components (e.g., weld locations, joint locations, support ring locations) from the pipe design drawings or structural models. Calculate the ratio of the coincidence length between the region boundary and the edge line of the structural component to the total length of the region boundary, as the boundary coincidence parameter.

[0160] Step S817: Mark the motion coordination region with the boundary overlap parameter lower than the preset boundary overlap threshold as the cross-structural component coordinated motion abnormal region, extract the motion evolution mode feature vector of all pipe wall surface texture primitive fragments in the cross-structural component coordinated motion abnormal region, and input the motion evolution mode feature vector of the pipe wall surface texture primitive fragments in the cross-structural component coordinated motion abnormal region into the disease pattern recognition processing.

[0161] Set a boundary overlap threshold, for example, 0.3. If the boundary overlap of a certain motion coordination region is lower than this threshold, it indicates that the region crosses the structural component boundary and its coordinated motion behavior is abnormal. This region is then marked as an abnormal cross-structural component coordinated motion region. Extract the motion evolution pattern feature vectors of all texture primitive fragments within this region and input them as additional features into the disease pattern recognition processing in step S140.

[0162] Step S818: In the disease pattern recognition processing, for each pipe wall surface texture primitive fragment in the cross-structural component cooperative motion abnormal area, its motion evolution mode feature vector is spliced ​​with its own compactness spatial distribution vector to generate a multi-dimensional fused disease feature vector of the pipe wall surface texture primitive fragment; based on the multi-dimensional fused disease feature vector of all fragments in the area, regional feature aggregation is performed, and the aggregated regional features are matched with the standard disease feature template in the preset disease category standard pattern library to generate a disease category identifier that integrates motion cooperative analysis information.

[0163] In the pattern similarity retrieval and matching process of step S146, for texture primitive fragments within the cross-structural component cooperative motion anomaly region, their motion evolution pattern feature vector and compactness spatial distribution vector are concatenated to form a higher-dimensional feature vector, which serves as the multi-dimensional fused disease feature vector for that fragment. The multi-dimensional fused disease feature vectors of all fragments within the region are averaged to obtain the region feature aggregation vector. This aggregation vector is then matched with the standard disease feature templates in the disease category standard pattern library to obtain the disease category identifier of the fused motion cooperative analysis information.

[0164] Step S819: Based on the defect category identifier of the fused motion coordination analysis information and the pipeline mileage positioning coordinate range corresponding to the regional boundary of the cross-structural component motion coordination anomaly area, generate a pipeline defect treatment instruction with additional cross-structural component motion coordination anomaly annotation.

[0165] In the pipeline defect handling instruction generated in step S156, a field is added to indicate whether the defect involves cross-structural component coordinated motion anomalies. The pipeline mileage location coordinate range of the cross-structural component coordinated motion anomaly area is also attached. The updated instruction serves as the pipeline defect handling instruction with the added cross-structural component coordinated motion anomaly annotation.

[0166] Based on the same inventive concept, please refer to Figure 2This diagram illustrates a schematic block diagram of a deep learning-based water supply and drainage network defect identification system provided in an embodiment of this application. The system includes a central processing unit (CPU), a system memory comprising random access memory (RAM) and read-only memory (ROM), and a system bus connecting the system memory and the CPU. The deep learning-based water supply and drainage network defect identification system also includes a basic input / output system to facilitate information transmission between various devices within the computer, and a large-capacity storage device for storing the operating system, applications, and other program modules.

[0167] A basic input / output system includes a display for showing information and input devices such as a mouse and keyboard for user input. Both the display and the input devices are connected to the central processing unit via an input / output controller connected to the system bus. The basic input / output system may also include an input / output controller for receiving and processing input from multiple other devices such as a keyboard, mouse, or electronic stylus. Similarly, the input / output controller also provides output to a display screen, printer, or other types of output devices.

[0168] A mass storage device is connected to the central processing unit via a mass storage controller connected to the system bus. The mass storage device and its associated computer-readable medium provide non-volatile storage for deep learning-based identification of defects in water supply and drainage networks. Without loss of generality, the computer-readable medium may include computer storage media and communication media. Computer storage media includes volatile and non-volatile, removable and non-removable media implemented using any method or technology for storing information such as computer-readable instructions, data structures, program modules, or other data. According to various embodiments of this application, deep learning-based identification of defects in water supply and drainage networks can also be performed by connecting to a remote computer on a network, such as the Internet. That is, deep learning-based identification of defects in water supply and drainage networks can be connected to a network via a network interface unit connected to the system bus, or alternatively, a network interface unit can be used to connect to other types of networks or remote computer systems.

[0169] In addition, in the specific embodiments of this application, data such as user information are involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0170] The above are merely exemplary embodiments of this application and are not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application shall be included within the protection scope of this application.

Claims

1. A method for identifying defects in water supply and drainage pipe networks based on deep learning, characterized in that, The method includes: Obtain the pipe-in-pipe visual detection data set corresponding to the water supply and drainage pipe network area to be identified. The pipe-in-pipe visual detection data set includes continuous frames of pipe-in-pipe visual image units and the pipe mileage positioning coordinates corresponding to each pipe-in-pipe visual image unit. The visual texture primitive decomposition process is performed on the visual detection data set inside the pipe to obtain the texture primitive distribution map of the pipe wall surface in the visual image unit inside the pipe and the spatial dependency relationship map between the texture primitives on the pipe wall surface. The pre-built pipe wall texture primitive evolution prediction network is invoked to perform temporal evolution prediction processing on the texture primitive distribution map and the spatial dependency relationship map, generating a predicted texture primitive evolution map of the pipe wall surface at future time points; Based on the predicted texture primitive evolution map, disease pattern recognition processing is performed to obtain the category identifier of potential diseases in the water supply and drainage network area to be identified and the description information of the expansion trend of potential diseases. A pipeline network defect treatment instruction is generated based on the category identifier and the extended trend description information, and the pipeline network defect treatment instruction is sent to the pipeline network maintenance and dispatch terminal. The method further includes: After acquiring the in-tube visual detection data set, the in-tube illumination distribution estimation process is performed on each in-tube visual image unit in the in-tube visual detection data set. The in-tube illumination distribution estimation process includes extracting the pixel brightness distribution histogram of the in-tube visual image unit and estimating the main direction angle parameter of the light source and the light source intensity attenuation coefficient based on the distribution pattern of the pixel brightness distribution histogram. An illumination distribution field model inside the tube is constructed using the principal direction angle parameter of the light source and the intensity attenuation coefficient of the light source. The illumination distribution field model inside the tube is used to describe the illumination intensity value of each pixel position in the visual image unit inside the tube. The illumination deviation ratio between the illumination intensity value and the standard reference illumination intensity value of each pixel position in the visual image unit of the tube is calculated based on the illumination distribution field model inside the tube. An illumination compensation coefficient is generated for each pixel position. The original pixel value of the corresponding pixel position in the visual image unit of the tube is multiplied and compensated pixel by pixel using the illumination compensation coefficient to generate a visual image unit of the tube with enhanced illumination consistency. The original intra-tube visual image units in the intra-tube visual detection data set are replaced with the intra-tube visual image units with enhanced illumination consistency to form an illumination-enhanced intra-tube visual detection data set, and the illumination-enhanced intra-tube visual detection data set is used for the visual texture primitive decomposition processing. The method further includes: After generating the predicted texture primitive evolution map, the pipeline laying material information database of the water supply and drainage network area to be identified is obtained, and the pipeline laying material type identifier corresponding to the pipeline mileage positioning coordinates and the material degradation rate parameter corresponding to the pipeline laying material type identifier are extracted from the pipeline laying material information database. From the set of predicted geometric morphology parameter changes for each pipe wall surface texture primitive fragment in the predicted texture primitive evolution diagram, extract the geometric morphology parameter changes corresponding to the material degradation rate parameter, compare them numerically with the material degradation rate parameter, calculate the deviation between the two, and if the deviation exceeds a preset material degradation deviation tolerance threshold, apply a material degradation constraint correction process to the geometric morphology parameter changes. The material degradation constraint correction process includes adjusting the geometric morphology parameter changes to their weighted average with the material degradation rate parameter. The change in geometric morphology parameters after material degradation constraint correction is added to the geometric morphology parameter value at the current time point to generate the predicted value of the geometric morphology parameter after the correction of the texture primitive fragment on the pipe wall surface, and the predicted geometric morphology parameter set is updated. Using the corrected predicted geometric properties of all pipe wall surface texture primitive fragments and the original predicted geometric properties of pipe wall surface texture primitive fragments that do not exceed the deviation tolerance, the corresponding predicted geometric properties in the predicted texture primitive evolution diagram are replaced to generate a predicted texture primitive evolution diagram corrected for material degradation constraints. The original predicted texture primitive evolution map is replaced by the predicted texture primitive evolution map after the material degradation constraint correction, and used as the input data for the disease pattern recognition processing; For example, the method may also include: After obtaining the texture primitive distribution map, the region contour defined by the geometric morphological attribute set of each pipe wall surface texture primitive fragment in the texture primitive distribution map is subjected to multi-scale morphological decomposition processing. The multi-scale morphological decomposition processing includes using a preset multi-level morphological decomposition kernel to perform layer-by-layer morphological peeling processing on the region contour to generate the sub-region contour set and its morphological attributes of the pipe wall surface texture primitive fragment at different morphological scales. The sub-region contour set and its morphological attributes at different morphological scales are input into a preset pipe wall texture primitive damage assessment model. The multi-scale morphological features are fused through the pipe wall texture primitive damage assessment model to generate a damage assessment index for the pipe wall surface texture primitive fragment. Based on the damage assessment index, all pipe wall surface texture primitive fragments in the texture primitive distribution map are classified into damage levels to obtain the damage level identifier corresponding to each pipe wall surface texture primitive fragment. The damage level identifier is associated with the geometric morphology attribute set of the corresponding pipe wall surface texture primitive fragment in the texture primitive distribution map as additional information, thereby generating an enhanced texture primitive distribution map containing the damage level identifier. The method further includes: after generating the predicted texture primitive evolution map, performing pipe wall stress field inversion processing based on the predicted spatial position coordinate changes and predicted geometric morphology parameter changes of each pipe wall surface texture primitive fragment in the predicted texture primitive evolution map; The change in the predicted spatial coordinates of each pipe wall surface texture primitive fragment at the predicted time point relative to its current spatial coordinates is used as the predicted displacement field data at the pipe wall surface texture primitive fragment. This data is then used as the boundary condition of the pipe wall surface displacement field and input into a preset pipe wall mechanical constitutive inversion model. The stress field is then solved in reverse through the pipe wall mechanical constitutive inversion model to generate the equivalent nodal stress tensor field of the pipe wall region where the pipe wall surface texture primitive fragment is located. Extract the principal stress direction angle of each node position from the equivalent node stress tensor field, extract the principal response direction angle of the corresponding pipe wall surface texture primitive fragment from the texture primitive distribution map on which the spatial dependency graph is based, calculate the angle deviation between the two, and generate the stress texture direction deviation angle associated with each pipe wall surface texture primitive fragment. Extract the pipe wall surface texture primitive fragments whose stress texture direction deviation angle exceeds the preset stress texture mismatch threshold, and mark the extracted pipe wall surface texture primitive fragments as stress texture anomalous fragments. Perform region connectivity processing on spatially adjacent stress texture anomalous fragments to generate stress texture anomalous connected domains. Ellipse fitting is performed on the geometric boundary of the stress texture abnormal connected domain to generate the stress concentration elliptical region corresponding to the stress texture abnormal connected domain, and the major axis direction angle, minor axis direction angle and major axis-minor axis length ratio of the stress concentration elliptical region are extracted. The angle of the major axis of the stress concentration elliptical region is taken as the main direction of potential crack propagation, and the ratio of the major axis to the minor axis is taken as the stress concentration degree factor. The coordinates of the geometric center point of the stress concentration elliptical region are taken as the coordinates of the potential crack initiation point. The potential crack propagation direction, the stress concentration factor, and the coordinates of the potential crack initiation point are input into a preset pipe wall crack propagation path prediction model. The pipe wall crack propagation path prediction model is used to generate a crack propagation path segment by segment along the potential crack initiation point coordinates along the potential crack propagation direction, generating a pipe wall potential crack propagation path prediction result containing a sequence of crack path node coordinates. The prediction results of the potential crack propagation path of the pipe wall are superimposed on the prediction texture primitive evolution map to generate an enhanced prediction texture primitive evolution map containing crack propagation path annotations. The enhanced prediction texture primitive evolution map replaces the original prediction texture primitive evolution map for the disease pattern recognition processing. In the disease pattern recognition process, the prediction result of the potential crack propagation path of the pipe wall is used as an additional input feature. Together with the boundary contour morphology attribute of the geometric boundary description polygon of the determined potential disease core area, it participates in the pattern similarity retrieval and matching process in the preset disease category standard pattern library to generate a disease category identifier that integrates stress field inversion information. The treatment process code in the pipeline network disease treatment instruction is updated based on the disease category identifier of the integrated stress field inversion information. The updated treatment process code replaces the original treatment process code, and a pipeline network disease treatment instruction based on the integrated stress inversion analysis results is generated.

2. The method for identifying defects in water supply and drainage pipe networks based on deep learning according to claim 1, characterized in that, The step of performing visual texture primitive decomposition on the visual detection data set inside the pipe to obtain the texture primitive distribution map of the pipe wall surface in the visual image unit inside the pipe and the spatial dependency relationship map between the texture primitives on the pipe wall surface includes: Extract the pipeline mileage positioning coordinates corresponding to each pipeline visual image unit in the pipeline visual detection data set. Arrange all pipeline visual image units with the same pipeline mileage positioning coordinates in the order of acquisition time to construct a single-point visual image time series corresponding to each pipeline mileage positioning coordinate. For the first pipeline visual image unit in each single-point visual image time series, use a preset texture primitive perception filter group to perform multi-directional texture response calculation processing on the first pipeline visual image unit to generate a multi-directional texture response vector for each pixel position in the first pipeline visual image unit. For each pixel location, the multi-directional texture response vector is subjected to directional main peak detection processing to extract the texture main response direction and texture main response intensity of each pixel location. Based on the texture main response direction, adjacent pixels with similar texture main response directions are aggregated to generate the initial texture primitive region set in the first intra-tube visual image unit. The region shape regularization processing is performed on each initial texture primitive region in the initial texture primitive region set, the region skeleton line of each initial texture primitive region is extracted, and the region skeleton line is subjected to equal curvature segment fitting processing along the extension direction of the region skeleton line to generate the tube wall surface texture primitive fragment corresponding to each initial texture primitive region. The remaining intra-pipe visual image units in the single-point visual image time series are sequentially input into the texture primitive perception filter group. Combined with the position information of the pipe wall surface texture primitive fragments extracted from the previous intra-pipe visual image units, the current intra-pipe visual image unit is subjected to texture primitive fragment position-guided texture primitive tracking and extraction processing to generate the spatiotemporal position sequence of the pipe wall surface texture primitive fragments in each intra-pipe visual image unit. Based on the spatiotemporal position sequence of each pipe wall surface texture primitive fragment, the spatial offset vector of the pipe wall surface texture primitive fragment at adjacent time points is extracted, and the direction angle offset and offset distance change of the spatial offset vector are extracted to generate the motion offset descriptor of each pipe wall surface texture primitive fragment. All pipe wall surface texture primitive fragments at the same time point in the single-point visual image time series are spatially arranged according to their spatial location attributes to generate a texture primitive distribution map of the pipe wall surface corresponding to each time point. The texture primitive distribution map records the geometric morphological attribute set and spatial location coordinates of each pipe wall surface texture primitive fragment. For any two spatially adjacent pipe wall surface texture primitive fragments in the texture primitive distribution map, calculate the angle difference of the main texture response directions of the two pipe wall surface texture primitive fragments as the directional angle, calculate the relative distance and direction between the boundaries of the two pipe wall surface texture primitive fragment regions as the boundary relative position offset, and construct the spatial dependency vector between the two pipe wall surface texture primitive fragments based on the directional angle and the boundary relative position offset. The spatial dependency vectors of all spatially adjacent pipe wall surface texture primitive fragment pairs in the texture primitive distribution map are summarized and organized according to the spatial adjacency order to generate a spatial dependency graph between the pipe wall surface texture primitives. The spatial dependency graph includes the topological structure information of nodes and edges and the spatial dependency vector corresponding to each edge. The texture primitive distribution map and spatial dependency graph corresponding to each pipeline mileage positioning coordinate are combined to form the output data of the visual texture primitive decomposition processing. The texture primitive distribution map in the output data contains the complete geometric morphological attributes of the texture primitive fragments on the pipe wall surface at the pipeline mileage positioning coordinate, and the spatial dependency graph in the output data contains the complete spatial constraint relationship between the texture primitive fragments on the pipe wall surface at the pipeline mileage positioning coordinate.

3. The method for identifying defects in water supply and drainage pipe networks based on deep learning according to claim 1, characterized in that, The step of calling a pre-built pipe wall texture primitive evolution prediction network to perform temporal evolution prediction processing on the texture primitive distribution map and the spatial dependency relationship map, generating a predicted texture primitive evolution map of the pipe wall surface at future time points, including: The geometric morphological attribute set of each texture primitive fragment on the pipe wall surface is extracted from the texture primitive distribution map, and the geometric morphological attribute set is encoded into a fixed-dimensional texture primitive geometric embedding vector. Simultaneously, the spatial position coordinates of each texture primitive fragment on the pipe wall surface are extracted from the texture primitive distribution map, and the spatial position coordinates are encoded into a fixed-dimensional texture primitive position embedding vector. The texture primitive geometric embedding vector and the texture primitive position embedding vector of each texture primitive fragment on the pipe wall surface are concatenated to generate a texture primitive comprehensive embedding vector corresponding to each texture primitive fragment on the pipe wall surface. The texture primitive comprehensive embedding vector contains a joint representation of the geometric morphological information and spatial position information of the texture primitive fragment on the pipe wall surface. The spatial dependency vector between each pair of pipe wall surface texture primitive fragments in the spatial dependency relationship diagram is input into a preset spatial relationship encoding module. The spatial dependency vector is then subjected to nonlinear mapping processing by the spatial relationship encoding module to generate a fixed-dimensional spatial relationship embedding vector. Using the texture primitive composite embedding vector as graph node features and the spatial relationship embedding vector as graph edge features, a texture primitive spatial dependency graph structure is constructed with the texture primitive fragments on the pipe wall surface as nodes. The initial feature of each node in the texture primitive spatial dependency graph structure is the texture primitive composite embedding vector of the corresponding texture primitive fragment. Using the graph attention feature propagation module in the pipe wall texture primitive evolution prediction network, the texture primitive spatial dependency graph structure is subjected to multiple graph attention message passing processes. In each graph attention message passing process, the message passing attention weight is calculated based on the spatial relationship embedding vector between the target node and its neighboring nodes. Based on the message passing attention weight, the texture primitive comprehensive embedding vectors of the neighboring nodes are aggregated into the texture primitive comprehensive embedding vector of the target node to generate the context-aware texture primitive feature vector of each node. The context-aware texture primitive feature vector of each pipe wall surface texture primitive fragment and the motion offset descriptor of the pipe wall surface texture primitive fragment are input into the temporal state update module of the pipe wall texture primitive evolution prediction network. The historical motion state of the motion offset descriptor is encoded by the gated recurrent processing unit in the temporal state update module to generate the initial motion state vector. The initial motion state vector and the context-aware texture primitive feature vector are subjected to feature-gated fusion processing. The gating mechanism of the gated loop processing unit selectively fuses the initial motion state vector and the context-aware texture primitive feature vector to generate a fused time-dependent texture primitive dynamic evolution feature vector. This fused feature vector is then input into the spatial coordinate regression branch of the pipe wall texture primitive evolution prediction network. The spatial coordinate regression branch performs spatial position offset regression calculation on the texture primitive dynamic evolution feature vector to generate the predicted spatial position offset of each pipe wall surface texture primitive fragment at future time points. The dynamic evolution feature vector of the texture primitive is simultaneously input into the geometric morphology regression branch in the pipe wall texture primitive evolution prediction network. The geometric morphology regression branch performs regression calculation on the dynamic evolution feature vector of the texture primitive to generate the predicted geometric morphology attribute change of each pipe wall surface texture primitive fragment at future time points. The predicted spatial position offset is vector-added with the spatial position coordinates at the current time point to generate the predicted spatial position coordinates of each pipe wall surface texture primitive fragment at the future time point; the value of each geometric morphology parameter is extracted from the geometric morphology attribute set at the current time point and added to the corresponding parameter change in the predicted geometric morphology attribute change to generate the predicted geometric morphology parameter set for each pipe wall surface texture primitive fragment at the future time point; based on the predicted spatial position coordinates and predicted geometric morphology parameter sets of all pipe wall surface texture primitive fragments, the texture primitive spatial configuration of the pipe wall surface is reconstructed to generate the predicted texture primitive evolution diagram of the pipe wall surface at the future time point.

4. The method for identifying defects in water supply and drainage pipe networks based on deep learning according to claim 1, characterized in that, The disease pattern recognition processing based on the predicted texture primitive evolution map yields the category identifiers of potential diseases within the water supply and drainage network area to be identified, as well as descriptions of the potential disease expansion trends, including: Obtain the set of predicted geometric parameters and predicted spatial coordinates of all pipe wall surface texture primitive fragments in the predicted texture primitive evolution diagram, and extract the texture primitive fragment area parameter and texture primitive fragment boundary perimeter parameter corresponding to each pipe wall surface texture primitive fragment from the set of predicted geometric parameters. The ratio of the area parameter of the texture primitive fragment to the perimeter parameter of the texture primitive fragment boundary is calculated to generate the area-perimeter ratio parameter corresponding to each texture primitive fragment on the pipe wall surface. The area-perimeter ratio parameter is used to characterize the compactness of the texture primitive fragment on the pipe wall surface. The centroid coordinates of each texture primitive fragment on the pipe wall surface are extracted from the predicted spatial position coordinates. The area-perimeter ratio parameter and the centroid coordinates of the texture primitive fragment are combined into a feature vector to construct a compactness spatial distribution vector for each texture primitive fragment on the pipe wall surface. The compactness spatial distribution vector of all pipe wall surface texture primitive fragments in the predicted texture primitive evolution diagram is input into a preset spatial density clustering processing module. The spatial density clustering processing module performs spatial neighborhood density calculation on the compactness spatial distribution vector to generate a density distribution map of the compactness spatial distribution vector. Based on the density distribution gradient change of the compactness spatial distribution vector in the density distribution map, the boundaries of continuous spatial regions with density values ​​exceeding a preset density threshold are extracted, and the regions enclosed by the boundaries of the continuous spatial regions are divided into compactness anomalous clustering regions. Within the compactness anomalous clustering regions, the texture principal response direction angle is extracted from the predicted geometric morphology parameter set of all pipe wall surface texture primitive fragments, and the difference value of the texture principal response direction angle between any two spatially adjacent pipe wall surface texture primitive fragments within the compactness anomalous clustering regions is calculated. Adjacent pipe wall surface texture primitive fragments whose texture main response direction angle difference value exceeds the preset diffusion judgment threshold are marked as direction disorder boundary fragments. The area surrounded by the direction disorder boundary fragments in the compactness abnormal aggregation area is determined as the potential disease core area. The spatial position coordinate sequence of all direction disorder boundary fragments in the potential disease core area is extracted. Polygon fitting processing is performed on the spatial position coordinate sequence to generate the geometric boundary description polygon of the potential disease core area. Based on the boundary contour morphology attributes of the polygon describing the geometric boundary of the potential disease core area and the numerical distribution characteristics of the compactness spatial distribution vector of the pipe wall surface texture primitive fragments within the potential disease core area, a pattern similarity retrieval and matching process is performed in a preset disease category standard pattern library to obtain a disease category identifier that matches the potential disease core area. The geometric center coordinates of the geometric boundary description polygon of the potential disease core area are extracted. Starting from the geometric center coordinates, the compactness spatial distribution vectors of the pipe wall surface texture primitive fragments adjacent to the potential disease core area are extracted. The numerical difference between the compactness spatial distribution vectors of the adjacent pipe wall surface texture primitive fragments and the compactness spatial distribution vectors of the pipe wall surface texture primitive fragments within the potential disease core area are calculated. Taking the spatial gradient descent direction of the numerical difference as the expansion direction, and combining the predicted texture primitive evolution diagrams of multiple future time points predicted by the pipe wall texture primitive evolution prediction network, the evolution sequence of the geometric boundary description polygon of the potential disease core area at continuous prediction time points is tracked to generate expansion trend description information including the expansion direction vector sequence and the expansion rate change sequence.

5. The method for identifying defects in water supply and drainage pipe networks based on deep learning according to claim 1, characterized in that, The step of generating a pipeline network defect treatment instruction based on the category identifier and the extended trend description information, and sending the pipeline network defect treatment instruction to the pipeline network maintenance dispatch terminal, includes: Obtain the preset treatment strategy knowledge base corresponding to the category identifier, retrieve the standard treatment process code that has a mapping relationship with the category identifier from the preset treatment strategy knowledge base, and extract the standard treatment resource requirement list associated with the standard treatment process code from the preset treatment strategy knowledge base; The expansion direction vector sequence and expansion rate change sequence in the expansion trend description information are analyzed. The main expansion direction vector corresponding to the current time is extracted from the expansion direction vector sequence, and the expansion rate value corresponding to the current time is extracted from the expansion rate change sequence. Obtain the geometric center coordinates of the potential disease core area determined in the disease pattern recognition process. Using the geometric center coordinates as the spatial reference starting point, along the spatial direction indicated by the main vector of the expansion direction, and using the expansion rate value as the expansion distance per unit time, perform spatial range extrapolation calculation to generate the predicted spatial coverage range of the potential disease core area in multiple future unit time periods. Map the predicted spatial coverage range onto the pipeline axis to generate the starting coordinates and ending coordinates of the pipeline mileage interval affected by the disease corresponding to each unit time period. The change in the starting coordinate of the pipeline mileage interval affected by the defect between adjacent time periods is calculated, and the change in the ending coordinate of the pipeline mileage interval affected by the defect between adjacent time periods is also calculated. An interval length change gradient between adjacent time periods is generated. The interval length change gradient is compared with multiple preset emergency level judgment intervals to determine the emergency level judgment interval to which the interval length change gradient belongs. The emergency level corresponding to the emergency level judgment interval is used as the emergency level of the emergency response. Based on the urgency level of the emergency response and the resource items in the standard response resource requirement list, resource scheduling priority matching is performed. According to the order of the urgency level, resource allocation time sequence weights are assigned to each resource item in the standard response resource requirement list to generate a response resource scheduling scheme that includes resource allocation time sequence arrangements. The standard treatment process code and the treatment resource scheduling scheme are fused and encapsulated. The starting coordinates and ending coordinates of the pipeline mileage interval affected by the disease for each unit time period are added to the encapsulated data structure. At the same time, the pipeline mileage positioning coordinates are added to generate a pipeline disease treatment instruction containing treatment process identifier, treatment resource list, treatment pipeline mileage interval and treatment response sequence, and then sent to the target pipeline maintenance scheduling terminal.

6. The method for identifying defects in water supply and drainage pipe networks based on deep learning according to claim 1, characterized in that, The method further includes: In the process of constructing the pipe wall texture primitive evolution prediction network, multiple sets of historical data of pipe wall defects are obtained at different collection periods. Each set of historical data of pipe wall defects contains the sequence of pipe wall visual image units corresponding to that collection period. The visual texture primitive decomposition process is performed on each set of historical visual detection data in the tube to generate a texture primitive distribution map and a spatial dependency map corresponding to each acquisition period. The texture primitive distribution map of adjacent acquisition periods is used as the evolution prediction input sample, and the texture primitive distribution map of the next acquisition period is used as the evolution prediction supervision label to construct a texture primitive evolution training sample set. The evolution prediction input samples from the texture primitive evolution training sample set are input into the pipe wall texture primitive evolution prediction network to be trained. The evolution prediction input samples are then processed by the pipe wall texture primitive evolution prediction network to be trained to generate a predicted texture primitive evolution output map. Calculate the spatial position offset error of each pipe wall surface texture primitive fragment between the predicted texture primitive evolution output map and the evolution prediction supervision label, sum the spatial position offset errors of all pipe wall surface texture primitive fragments, and generate the texture primitive position offset loss function value. Calculate the error in geometric morphological attribute change of each pipe wall surface texture primitive fragment between the predicted texture primitive evolution output map and the evolution prediction supervision label, sum the errors in geometric morphological attribute change of all pipe wall surface texture primitive fragments, and generate the texture primitive morphological change loss function value. The texture primitive position offset loss function value is multiplied with a preset position loss weight coefficient to generate a weighted position loss value. The texture primitive shape change loss function value is multiplied with a preset shape loss weight coefficient to generate a weighted shape loss value. The weighted position loss value and the weighted shape loss value are added together to generate a total loss function value. The total loss function value is used to perform gradient backpropagation on the pipe wall texture primitive evolution prediction network to be trained. The partial derivative of each network weight parameter in the pipe wall texture primitive evolution prediction network to be trained with respect to the total loss function value is calculated. The partial derivative is then used to update each network weight parameter with gradient descent to obtain the updated network weight parameters. The next evolution prediction input sample from the texture primitive evolution training sample set is input into the pipe wall texture primitive evolution prediction network after the network weight parameters have been updated. The forward propagation process, loss function value calculation process, and network weight parameter update process are repeatedly executed until the total loss function value converges to the preset training termination condition, and the trained pipe wall texture primitive evolution prediction network is obtained. The training termination condition is that the numerical change of the total loss function value in the continuous training iteration process is less than the preset convergence change threshold. All network weight parameters of the pipe wall texture primitive evolution prediction network at the time of training completion are saved as the network weight parameters of the pre-constructed pipe wall texture primitive evolution prediction network.

7. The method for identifying defects in water supply and drainage pipe networks based on deep learning according to claim 5, characterized in that, The method further includes: After generating the pipeline network defect treatment instruction and sending the pipeline network defect treatment instruction to the pipeline network maintenance and dispatch terminal, the system receives a treatment completion status report fed back by the pipeline network maintenance and dispatch terminal after performing the treatment operation. The treatment completion status report includes a treatment completion timestamp and the pipeline mileage positioning coordinate range covered by the treatment operation. Based on the pipeline mileage positioning coordinate range covered by the treatment operation, obtain the set of visual detection data inside the pipe that is collected again after the treatment. Perform the visual texture primitive decomposition processing on the set of visual detection data inside the pipe that is collected again after the treatment to generate a texture primitive distribution map and a spatial dependency relationship map after the treatment. The texture primitive distribution map corresponding to the water supply and drainage network area to be identified before processing is used as the first input, and the texture primitive distribution map after processing is used as the second input. The texture primitive recovery effect comparison branch in the pipe wall texture primitive evolution prediction network is input, and the geometric shape recovery deviation of each pipe wall surface texture primitive fragment between the first input and the second input is extracted through the texture primitive recovery effect comparison branch. Based on the geometric shape recovery deviation of each pipe wall surface texture primitive fragment and the spatial coverage area of ​​the pipe wall surface texture primitive fragment in the texture primitive distribution map, calculate the overall texture primitive recovery degree index of the treatment operation coverage area; The overall texture primitive recovery degree index is associated with the pipeline mileage positioning coordinate range covered by the disposal operation in the disposal completion status report and stored to generate disposal effect evaluation data records. Extract the treatment process code and the corresponding overall texture primitive recovery degree index from the treatment effect evaluation data record. Use the overall texture primitive recovery degree index as the treatment efficiency update parameter for the treatment process code. Perform weighted update processing on the historical treatment efficiency records corresponding to the treatment process code in the preset treatment strategy knowledge base to obtain the updated treatment strategy knowledge base.

8. A deep learning-based system for identifying defects in water supply and drainage pipe networks, characterized in that, include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; The processor is configured to execute the deep learning-based water supply and drainage network defect identification method according to any one of claims 1 to 7 by executing the machine-executable instructions.

Images