An image detection-based urban drainage pipe network endoscopic detection optimization method

Abstract

The application discloses a kind of urban drainage pipe network endoscopy detection optimization methods based on image detection, it is related to pipe network image detection technical field, including: collecting urban drainage pipe network endoscopy video, extract continuous original frame image, establish pipe wall development image and coordinate mapping relationship;Establish pipe wall isotope frame group, calculate continuous visible area, edge clear degree and interference area proportion, determine preferred interpretation fragment and candidate interpretation fragment;Determine candidate abnormal area boundary, candidate abnormal area area and candidate category, write candidate abnormal area into corresponding pipe wall isotope frame group;Generate candidate abnormal area isotope actual measurement boundary and three kinds of attribution reference boundary, calculate pipe wall attachment attribution difference, reflection shielding interference attribution difference and water surface dynamic interference attribution difference, determine attribution type and trigger candidate interpretation fragment recheck, generate severity representation value, detection content, exclusion record or review content;Endoscopy detection credibility and recheck effectiveness are improved.

Description

Technical Field

[0001] This invention relates to the field of pipeline network image detection technology, and in particular to an optimized method for endoscopic detection of urban drainage pipeline networks based on image detection. Background Technology

[0002] Endoscopic inspection of urban drainage pipe networks typically uses pipe robots or push-rod camera equipment to acquire continuous video. Combined with mileage recording, lens calibration, distortion correction, and grayscale edge analysis, it identifies locations such as pipe walls, joints, pipe bottoms, and water surfaces. Through manual interpretation or image algorithms, it extracts abnormal features such as cracks, misalignments, leaks, deposits, and obstacles, which are used for pipe network maintenance assessment, defect location, and repair decisions.

[0003] However, conventional methods still have two limitations in complex drainage environments. On the one hand, conventional methods usually use single frames or local continuous frames as the basis for interpretation, and lack cross-frame co-location and selection mechanisms for the same actual pipe wall location, making them susceptible to fluctuations in visible area and edge clarity. On the other hand, they are not precise enough in distinguishing between the boundaries of interference such as reflections, lens water, sludge obstruction, and water surface ripples and the boundaries of abnormal pipe wall adhesion, resulting in relatively insufficient basis for re-inspection. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an optimized method for endoscopic inspection of urban drainage pipe networks based on image detection, which solves the problems of insufficient stability in the selection of segments for judging the same actual pipe wall location and the easy confusion between the attribution of interference boundaries and abnormal pipe wall attachment boundaries in the existing technology.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] This invention provides an optimized method for endoscopic inspection of urban drainage pipe networks based on image detection. The method includes: acquiring endoscopic video of the urban drainage pipe network, extracting continuous original frame images, identifying pipe inner wall boundaries, interface loops, pipe bottom direction, visible interference boundaries, and dynamic interference boundaries, and establishing a mapping relationship between the unfolded pipe wall image and coordinates; establishing a pipe wall co-position frame group based on the unfolded pipe wall image and coordinate mapping relationship, determining the visible pipe wall area based on the interference area formed by the visible interference boundaries, calculating the continuous visible area, edge clarity, and interference area ratio to form a sequence of interpretation segments, and determining the preferred interpretation segment and candidate interpretation segments; and extracting boundary candidates and region candidates based on the preferred interpretation segment, and classifying the boundary... Candidate and regional candidates are merged to generate candidate anomaly areas. The boundaries, areas, and categories of candidate anomaly areas are determined. Candidate anomaly areas are written into the corresponding pipe wall co-location frame group to establish a binding relationship. Based on the candidate anomaly area boundaries, candidate categories, pipe wall location numbers, preferred interpretation segment frame numbers, pipe wall co-location frame groups, interpretation segment sequences, and coordinate mapping relationships, the measured co-location boundaries and three types of attribution reference boundaries of candidate anomaly areas are generated. The differences in pipe wall attachment attribution, reflection shading interference attribution, and water surface dynamic interference attribution are calculated. The attribution type is determined based on the unique minimum value of the attribution difference, and the re-examination of candidate interpretation segments is triggered. The severity characterization value, detection content, exclusion records, or review content are generated.

[0008] As a preferred embodiment of the image detection-based optimization method for endoscopic inspection of urban drainage pipe networks according to the present invention, the step of establishing the pipe wall unfolded image and coordinate mapping relationship includes: acquiring endoscopic video of urban drainage pipe networks, synchronously recording the acquisition time and travel distance, performing distortion correction on the original frame image to generate a corrected frame image, and generating a grayscale image and a corrected color image from the corrected frame image; performing boundary recognition and connectivity analysis on the grayscale image and the corrected color image to obtain the pipe inner wall boundary, pipe center position, pipe bottom direction, water surface boundary, floating object boundary, water flow ripple boundary, reflective area boundary, and sludge-covered area boundary; and performing pipe wall projection unfolding based on the pipe inner wall boundary, pipe center position, pipe bottom direction, and travel distance to generate the pipe wall unfolded image, coordinate mapping relationship, interface loop, and lens water-holding area boundary.

[0009] As a preferred embodiment of the image detection-based optimization method for endoscopic inspection of urban drainage pipe networks according to the present invention, the establishment of the pipe wall co-position frame group includes: according to the pipe wall unfolded image and coordinate mapping relationship, the image segments corresponding to the same actual pipe wall position in different correction frame images are grouped into the pipe wall co-position frame group.

[0010] As a preferred embodiment of the image detection-based optimization method for endoscopic inspection of urban drainage pipe networks according to the present invention, the calculation of continuous visible area, edge clarity, and interference area proportion includes: mapping the boundaries of reflective areas, lens-mounted water areas, sludge-covered areas, and water surfaces to the expanded image coordinates of each image segment in the pipe wall co-position frame group according to coordinate mapping relationships, merging them to form interference areas, subtracting interference areas from the complete range of image segments to generate visible pipe wall areas, performing connectivity analysis on the visible pipe wall areas to generate continuous visible pipe wall areas and segments without visible pipe wall areas; performing area statistics on the continuous visible pipe wall areas to generate continuous visible areas, taking the median of the grayscale change intensity of pixels within the visible pipe wall areas to generate edge clarity, and performing area statistics on interference areas and the complete range of image segments to generate interference area proportion.

[0011] As a preferred embodiment of the image detection-based optimization method for endoscopic inspection of urban drainage pipe networks according to the present invention, the step of determining the preferred interpretation segment and the candidate interpretation segment includes: sorting image segments with visible pipe wall areas in lexicographical order according to continuous visible area, edge clarity, proportion of interference area, and acquisition time; appending segments without visible pipe wall areas to the end of the sorting to generate a sequence of interpretation segments; reading the sequence of interpretation segments, determining the image segment ranked first as the preferred interpretation segment, and determining the image segments ranked second and after the second as candidate interpretation segments.

[0012] As a preferred embodiment of the image detection-based endoscopic detection optimization method for urban drainage pipe networks described in this invention, the step of extracting boundary candidates and region candidates based on the preferred interpretation segment includes: calculating grayscale changes within the candidate extraction range based on the preferred interpretation segment, extracting pixels with abrupt grayscale changes and skeletonizing them to generate an edge skeleton, analyzing the topological structure of the edge skeleton and assigning boundary attributes to form boundary candidates, and performing connectivity analysis on the grayscale image and the corrected color image within the candidate extraction range for bright areas, dark areas, and color anomaly areas to generate region candidates.

[0013] As a preferred embodiment of the image detection-based optimization method for endoscopic inspection of urban drainage pipe networks according to the present invention, the determination of the candidate anomaly zone boundary, candidate anomaly zone area, and candidate category includes: geometrically merging boundary candidates and region candidates under the coordinates of the unfolded pipe wall image to generate candidate anomaly zones; extracting the outer edge of the candidate anomaly zone as the candidate anomaly zone boundary; and calculating the candidate anomaly zone area. Based on the interface loop, pipe bottom direction, water surface boundary, floating object boundary, and candidate anomaly zone boundary, the candidate anomaly zones are judged one by one according to the order of interface-related candidates, pipe bottom-related candidates, and pipe wall surface candidates to determine the candidate category. Based on the pipe wall location number to which the candidate anomaly zone belongs, the pipe wall co-position frame group is read, and the candidate anomaly zone boundary, candidate anomaly zone area, candidate category, and preferred judgment segment frame number are written into the pipe wall co-position frame group to establish a binding relationship.

[0014] As a preferred embodiment of the image detection-based optimization method for endoscopic inspection of urban drainage pipe networks according to the present invention, the generation of the measured boundary of the candidate anomaly area and the three types of attribution reference boundaries includes: reading the pipe wall co-location frame group and the judgment segment sequence according to the pipe wall location number; mapping the candidate anomaly area boundary to each frame of the correction frame image within the same pipe wall co-location frame group according to the preferred judgment segment frame number and coordinate mapping relationship to generate the mapped boundary; re-extracting boundary candidates within the same pipe wall location range to generate the measured boundary of the candidate anomaly area; uniformly mapping the measured boundary of the candidate anomaly area to the coordinates of the unfolded pipe wall image; using the mapped boundary as the pipe wall attachment attribution reference boundary, mapping the boundary of the reflective area, the boundary of the lens water-hanging area, and the boundary of the sludge-covered area to the coordinates of the unfolded pipe wall image to generate the reflective occlusion interference attribution reference boundary; mapping the water surface boundary, the floating object boundary, and the water flow ripple boundary to the coordinates of the unfolded pipe wall image to generate the water surface dynamic interference attribution reference boundary, thus forming the three types of attribution reference boundaries.

[0015] As a preferred embodiment of the image detection-based optimization method for endoscopic inspection of urban drainage pipe networks according to the present invention, the step of determining the attribution type based on the unique minimum value of the attribution difference includes: calculating the attribution difference of pipe wall attachment, the attribution difference of reflective occlusion interference, and the attribution difference of water surface dynamic interference based on the measured boundary of the candidate anomaly area and the three types of attribution reference boundaries; comparing the attribution differences of pipe wall attachment, reflective occlusion interference, and water surface dynamic interference, and when the attribution difference of pipe wall attachment is the unique minimum value, the candidate anomaly area is determined to be a pipe wall attachment anomaly; when the attribution difference of reflective occlusion interference is the unique minimum value, the candidate anomaly area extraction is re-executed by calling the candidate interpretation segment, and the measured boundary generation of the candidate anomaly area is performed again on the re-extracted candidate anomaly area. The process includes generating reference boundaries for three categories of classification, calculating differences among the three categories of classification, and determining the classification type. When the difference in classification of water surface dynamic disturbance is the only minimum value and the candidate category is scum or obstacle, the candidate anomaly area is retained and recorded as a functional anomaly under the classification of water surface dynamic disturbance. When the difference in classification of water surface dynamic disturbance is the only minimum value and the candidate category is crack, rupture, misalignment, disconnection, or leakage, the candidate anomaly area extraction is re-executed by calling the candidate interpretation segment. The re-extracted candidate anomaly area is then subjected to the same measured boundary generation, generation of reference boundaries for three categories of classification, calculation of differences among the three categories of classification, and determination of the classification type. When there is no unique minimum value for the differences in classification of pipe wall adhesion, reflective shading interference, and water surface dynamic disturbance, the review content is generated.

[0016] As a preferred embodiment of the image detection-based endoscopic inspection optimization method for urban drainage pipe networks described in this invention, the generation of severity characterization values, detection content, exclusion records, or review content includes: calculating severity characterization values ​​for candidate abnormal areas determined to be pipe wall adhesion abnormalities and candidate abnormal areas recorded as functional abnormalities under water surface dynamic interference, and generating detection content by combining candidate categories, axial positions, circumferential positions, and attribution types; generating exclusion records for candidate abnormal areas that do not reappear after re-examination by alternative interpretation segments; and generating review content for candidate abnormal areas without alternative interpretation segments, where the attribution difference does not have a unique minimum value, and where pipe wall adhesion abnormalities still cannot be formed after re-examination by alternative interpretation segments.

[0017] The beneficial effects of this invention are as follows: by forming a sequence of interpretation segments and determining the preferred interpretation segment and the candidate interpretation segment, cross-frame optimal interpretation of the same actual pipe wall location is achieved, improving the stability of the candidate anomaly area extraction basis; by generating the measured boundary of the candidate anomaly area and three types of attribution reference boundaries, and triggering the re-examination of the candidate interpretation segment according to the attribution difference, the attribution determination of pipe wall adhesion anomalies, reflective occlusion interference and water surface dynamic interference is achieved, generating detection content, exclusion records or review content, improving the reliability of endoscopic detection and the effectiveness of re-examination. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of an optimized method for endoscopic inspection of urban drainage pipe networks based on image detection.

[0020] Figure 2 A flowchart for the pipe wall unfolding mapping.

[0021] Figure 3 A flowchart for selecting the best segment for interpretation.

[0022] Figure 4 This is a flowchart for the re-examination of attribution determination.

[0023] Figure 5 A diagram showing the optimal result for interpreting the segment.

[0024] Figure 6 This is a stability diagram of the isotopic boundary. Detailed Implementation

[0025] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0026] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0027] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0028] Reference Figures 1-6 This is one embodiment of the present invention, which provides an optimized method for endoscopic detection of urban drainage pipe networks based on image detection, including the following steps:

[0029] S1. Collect endoscopic video of urban drainage pipe network, extract continuous raw frame images, identify pipe inner wall boundary, interface loop, pipe bottom direction, visible interference boundary and dynamic interference boundary, and establish pipe wall unfolding image and coordinate mapping relationship.

[0030] It should be noted that the visible interference boundary includes the boundary of the reflective area, the boundary of the lens water-covered area, the boundary of the mud-covered area, and the water surface boundary. The visible interference boundary is used to form the interference area and deduct the area in the complete range of the image segment that is not suitable for pipe wall interpretation. The dynamic interference boundary includes the water surface boundary, the floating object boundary, and the water ripple boundary. The dynamic interference boundary is used to generate the reference boundary for the dynamic interference on the water surface. Among them, the water surface boundary is used to determine the water surface occlusion area in the visible interference boundary, and the water surface boundary is used to characterize the dynamic interference reference formed by the change of the water surface position with the continuous frame images in the dynamic interference boundary.

[0031] The system collects endoscopic video of the urban drainage network, simultaneously recording the acquisition time and travel distance. Distortion correction is performed on the original frame images to generate corrected frame images. Grayscale and corrected color images are then generated from the corrected frame images. Boundary recognition and connectivity analysis are performed on the grayscale and corrected color images to obtain the boundaries of the pipe inner wall, pipe center position, pipe bottom direction, water surface boundary, floating object boundary, water flow ripple boundary, reflective area boundary, and sludge-covered area boundary.

[0032] Furthermore, before testing, the endoscopic camera is calibrated using a calibration board to obtain the camera's intrinsic parameters and distortion parameters.

[0033] During the inspection process, the endoscopic device is moved along the pipe section to be inspected from the starting inspection well to the ending inspection well, continuously acquiring endoscopic videos of the urban drainage pipe network, and simultaneously recording the acquisition time and travel distance corresponding to each frame of the image.

[0034] Furthermore, consecutive raw frame images are extracted in chronological order of acquisition time, and each raw frame image is bound and saved with acquisition time, travel distance, camera intrinsic parameters, and distortion parameters.

[0035] Furthermore, distortion correction is performed on each original frame image using camera intrinsic parameters and distortion parameters to obtain a corrected frame image; a grayscale image is generated from the corrected frame image, while the corrected color image is retained.

[0036] Furthermore, gray-level abrupt change edges are extracted in each frame of gray-level image. According to the eight-neighborhood connectivity rule, gray-level abrupt change edge pixels that are eight neighbors to each other are assigned to the same connected boundary. The connected boundaries where the first and last edge pixels are eight neighbors to each other are recorded as candidate closed contours. The candidate closed contour that includes the geometric center of the image and has the largest enclosed area is selected as the inner wall boundary of the pipe.

[0037] It should be noted that for frames that cannot form a complete closed contour, the identified visible inner wall arc segments are extracted, and the visible inner wall arc segments and the pipe inner wall boundary points within the corresponding angle range in the previous and next frames are used together as fitting points. The complete pipe inner wall boundary is obtained by least squares ellipse fitting. The geometric center of the area enclosed by the pipe inner wall boundary is calculated as the pipe center position.

[0038] Furthermore, on the inner wall boundary of the pipe, find the boundary point with the lowest vertical coordinate in the image, and determine the direction from the center of the pipe to the lowest boundary point as the pipe bottom direction.

[0039] It should be noted that if the lowest position of the inner wall boundary of the pipe cannot be directly determined by the visible inner wall edge of the pipe, or if the boundary point at the lowest position is not the true edge of the pipe wall, then an ellipse is fitted along the visible arc segments on both sides of the inner wall boundary of the pipe, and the pipe bottom direction is determined according to the point with the lowest vertical coordinate on the fitted ellipse.

[0040] Furthermore, within the image region corresponding to the bottom of the pipe, connectivity analysis is performed on the edges of gray-scale abrupt changes, and the longest continuous boundary that intersects with both sides of the inner wall boundary of the pipe and spans the bottom region of the pipe is recorded as the water surface boundary.

[0041] On the side of the water surface boundary closer to the bottom of the pipe, the connected boundary that is adjacent to the water surface boundary and whose outer edge is independently closed or semi-closed is recorded as the floating object boundary.

[0042] Furthermore, within the water surface area, the elongated, continuous, alternating light and dark, slender, connected boundary extending along the pipe axis is recorded as the water flow ripple boundary.

[0043] In correcting color and grayscale images, the grayscale image is treated as an eight-bit grayscale image with grayscale values ​​ranging from 0 to 255. Based on the grayscale value distribution of the current frame, the maximum inter-class variance method is used to obtain the brightness segmentation boundary within the grayscale value range. Pixels with grayscale values ​​higher than the brightness segmentation boundary are grouped into bright pixels. Connectivity analysis is performed on the bright pixels, and the connected regions formed by the bright pixels are taken as bright connected regions.

[0044] Furthermore, taking the outer edge of the bright connected region as the starting boundary, the set of pixels outside the bright connected region that have eight neighboring regions that are in contact with the bright connected region but do not belong to the bright connected region is determined as the extended neighborhood; the median gray-level gradient inside the bright connected region and in the extended neighborhood are calculated respectively; when the median gray-level gradient inside the bright connected region is less than the median gray-level gradient in the extended neighborhood, the bright connected region is recorded as a reflective region, and the outer edge of the bright connected region is extracted as the boundary of the reflective region.

[0045] Furthermore, a ray is drawn from the center of the pipe towards the bottom, and the area along the ray and the lower half of the pipe wall boundary on both sides is defined as the lower pipe wall area. Connectivity analysis is then performed on the grayscale image and the corrected color image within the lower pipe wall area.

[0046] Furthermore, connected regions that have gray-level or color boundaries with adjacent pipe wall regions and whose median gray-level gradient is less than that of adjacent pipe wall regions are recorded as sludge-covered regions, and their outer edges are extracted as the boundaries of the sludge-covered regions.

[0047] If the sludge-covered area is in contact with the water surface boundary, the adjacency relationship between the boundary of the sludge-covered area and the water surface boundary is preserved.

[0048] Based on the inner wall boundary of the pipe, the center position of the pipe, the direction of the bottom of the pipe, and the travel distance, the pipe wall projection is unfolded to generate the unfolded image of the pipe wall, the coordinate mapping relationship, the interface loop, and the boundary of the water-holding area of ​​the lens.

[0049] Furthermore, before establishing the unfolded image of the pipe wall, imaging rays are established for the pixels in the corrected frame image based on the camera intrinsic parameters and distortion parameters; based on the inner wall boundary of the pipe and the center position of the pipe, the elliptical contour corresponding to the inner wall boundary of the pipe is back-projected into the pipe coordinates of the current frame, and the back-projected inner wall contour of the pipe is fitted into a local cylindrical surface.

[0050] Furthermore, the axial direction of the local cylindrical surface is normalized and determined as the unit vector of the pipe axis, the radius of the local cylindrical surface is determined as the radius of the inner wall of the pipe, and the optical center of the camera is determined as the position of the optical center in the pipe coordinates of the current frame; the bottom direction of the pipe is projected onto a cross section perpendicular to the unit vector of the pipe axis and normalized as the unit vector of the bottom direction of the pipe; and a transverse unit vector perpendicular to the unit vector of the bottom direction of the pipe is constructed in the same cross section.

[0051] Furthermore, for any pixel in the calibration frame image, the pixel is back-projected into an imaging ray originating from the optical center according to the camera intrinsic parameters. The intersection of the imaging ray and the local cylindrical surface is then determined, and the intersection point is identified as the three-dimensional projection point of the pixel on the inner wall of the pipe.

[0052] Furthermore, based on the offset of the 3D projection point relative to the optical center position in the axial direction of the pipe, the axial distance of the pixel is determined; based on the arc distance of the 3D projection point relative to the unit vector of the pipe bottom direction, the circumferential position of the pixel on the pipe wall is determined, thus forming the unfolded image of the pipe wall and the coordinate mapping relationship.

[0053] Specifically, the axial distance of the pipe and the circumferential position of the pipe wall are expressed as follows: ; ;

[0054] in, Indicates the first In-frame correction of frame image coordinates The corresponding axial distance of the pipeline, in units consistent with the travel distance. This indicates the original frame number; the corrected frame image uses the corresponding frame number from the original frame image. This represents the horizontal coordinate of a pixel in the corrected frame image. This represents the vertical coordinate of a pixel in the corrected frame image. Indicates the first The travel distance corresponding to the frame. Indicates the first In-frame correction of frame image coordinates The three-dimensional projection points obtained by the intersection of the pixel points along the imaging ray and the local cylindrical surface. Indicates the first The position of the optical center of the frame camera in the current frame pipeline coordinates. Indicates the first The unit vector of the axial direction of the local cylindrical surface in the frame. Indicates the first In-frame correction of frame image coordinates The corresponding circumferential position of the pipe wall, expressed in units of length. Indicates the first The radius of the local cylindrical surface in the frame. This represents the modulo operation. This represents the bivariate arctangent operator. Indicates the first The unit vector of the pipe bottom direction formed in the cross-section perpendicular to the pipe axis is the pipe bottom direction. Indicates the first The lateral unit vector perpendicular to the unit vector in the direction of the tube bottom within the same cross-section of the frame. Represents pi (π). Represents the vector dot product. This indicates a scalar multiplication operation.

[0055] Furthermore, while creating the unfolded image of the pipe wall, the coordinate mapping relationship is saved. The coordinate mapping relationship includes the original frame number, the coordinates of the corrected frame image, the axial distance of the pipe, the circumferential position of the pipe wall, and the reverse mapping record.

[0056] Furthermore, the reverse mapping record is a lookup table for the coordinates of the unfolded tube wall image to the coordinates of the correction frame image. When one unfolded tube wall image coordinate corresponds to multiple correction frame image coordinates, the correction frame image coordinate with the shortest projection distance is retained. When one unfolded tube wall image coordinate does not directly correspond to a correction frame image coordinate, it is determined by interpolation based on the correction frame image coordinates corresponding to adjacent unfolded tube wall image coordinates.

[0057] In the unfolded image of the pipe wall, connectivity analysis is performed on the edges of gray-scale abrupt changes to extract the strip-shaped regions extending along the circumferential direction; for multiple strip-shaped regions existing in the same axial neighborhood, the strip-shaped region with the longest continuous circumferential length is recorded as the interface loop line in the axial neighborhood; if multiple interface loop lines exist at different axial positions, they are recorded separately according to the axial distance from the starting inspection well to the ending inspection well.

[0058] Furthermore, in the continuous original frame images, we find the boundary with continuous brightness change, closed or semi-closed outer edge, and covering the underlying pipe wall texture. We calculate the change in the geometric center of the boundary in the original image coordinates. By unfolding the pipe wall image, we calculate the change in the pipe wall texture in the unfolded pipe wall coordinates during the same time period.

[0059] When the change in the geometric center of the boundary in the original image coordinates is less than the change in the pipe wall texture in the pipe wall unfolding coordinates during the same period, the area enclosed by the boundary is recorded as the lens water-hanging area, and the outer edge of the lens water-hanging area is recorded as the lens water-hanging area boundary.

[0060] S2. Based on the unfolded image of the pipe wall and the coordinate mapping relationship, establish a pipe wall co-position frame group. Based on the interference area formed by the visible interference boundary, determine the visible pipe wall area, calculate the continuous visible area, edge clarity and interference area ratio, form a sequence of interpretation segments, and determine the preferred interpretation segment and the candidate interpretation segment.

[0061] Based on the unfolded pipe wall image and coordinate mapping relationship, image segments corresponding to the same actual pipe wall position in different calibration frame images are grouped into pipe wall co-position frame groups. According to the coordinate mapping relationship, the boundaries of the reflective area, the lens water-hanging area, the sludge-covered area, and the water surface are mapped to the pipe wall unfolded image coordinates of each image segment in the pipe wall co-position frame group, and merged to form interference areas. The interference areas are deducted from the complete range of the image segments to generate visible pipe wall areas. Connectivity analysis is performed on the visible pipe wall areas to generate continuous visible pipe wall areas and segments without visible pipe wall areas. The area of ​​the continuous visible pipe wall areas is statistically analyzed to generate continuous visible areas. The median of the grayscale change intensity of pixels in the visible pipe wall areas is taken to generate edge sharpness. The area of ​​the interference areas and the complete range of the image segments is statistically analyzed to generate the interference area ratio.

[0062] Furthermore, based on the pipe wall unfolding image and coordinate mapping relationship, and using the pipe axial distance and pipe wall circumferential position in the pipe wall unfolding image as the positioning basis, image segments corresponding to the same actual pipe wall position in different correction frame images are grouped into the same pipe wall co-position frame group.

[0063] Specifically, the axial and circumferential sampling step sizes of the pipe wall unfolding image are determined based on the coordinate mapping relationship. The difference in axial distance between two adjacent axial sampling positions is determined as the axial distance interval, and the difference in circumferential position between two adjacent circumferential sampling positions is determined as the circumferential position interval. The pipe wall unfolding image is divided into discrete pipe wall regions according to the axial distance interval and the circumferential position interval, and each discrete pipe wall region is assigned a unique pipe wall position number. Based on the coordinate mapping relationship, the image segment corresponding to each pipe wall position number in different calibration frame images is found. Image segments with the same pipe wall position number are grouped into a pipe wall co-position frame group. Each image segment synchronously saves the original frame number, acquisition time, travel distance, pipe wall unfolding image range, and coordinate mapping relationship.

[0064] Furthermore, for each image segment in each pipe wall co-position frame group, the boundaries of the reflective area, the lens water-receiving area, the sludge-covered area, and the water surface are read. Based on the coordinate mapping relationship, the boundaries of the reflective area, the lens water-receiving area, the sludge-covered area, and the water surface are mapped to the pipe wall unfolded image coordinates. Under the pipe wall unfolded image coordinates, the reflective area enclosed by the reflective area boundary, the lens water-receiving area enclosed by the lens water-receiving area boundary, the sludge-covered area enclosed by the sludge-covered area boundary, and the water surface-covered area enclosed by the water surface boundary towards the bottom of the pipe are merged into an interference area.

[0065] Furthermore, based on the coordinate mapping relationship, the complete range of each image segment in the unfolded image of the pipe wall is determined, and the interference area is subtracted from the complete range of the image segment to obtain the visible pipe wall area.

[0066] It should be noted that if the visible pipe wall region is split into multiple unconnected regions, the boundaries and areas of each unconnected region are saved separately; if the visible pipe wall region does not exist, the image segment is recorded as a segment without a visible pipe wall region and retained in the pipe wall co-position frame group.

[0067] Furthermore, connectivity analysis is performed on the visible pipe wall regions. If there are multiple unconnected visible pipe wall regions within an image segment, the visible pipe wall region with the highest area value is selected as the continuous visible pipe wall region; the area of ​​the continuous visible pipe wall region is taken as the continuous visible area.

[0068] Specifically, the number of pixels in the continuously visible pipe wall area is combined with the area of ​​a single pixel in the unfolded image of the pipe wall to obtain the continuously visible area; the continuously visible area is used to represent the range of the pipe wall that can be continuously observed for image detection in the current image segment at the same actual pipe wall location; the higher the continuously visible area, the more complete the morphology of the candidate anomaly region.

[0069] It should be noted that segments with no visible pipe wall area are not included in the calculation of continuous visible area, nor are they included in the calculation of edge sharpness; after the image segments with visible pipe wall area are sorted in lexicographical order, the segments with no visible pipe wall area are arranged at the end of the interpretation segment sequence according to the acquisition time.

[0070] Furthermore, the grayscale image corresponding to the visible pipe wall area in the unfolded pipe wall image is read, and the grayscale changes are calculated along the axial distance direction of the pipe and the circumferential position direction of the pipe wall. For any pixel, the grayscale value of the pixel, the grayscale values ​​of the axially adjacent pixels, and the grayscale values ​​of the circumferentially adjacent pixels are read. The absolute value of the grayscale difference between the pixel and the axially adjacent pixels is determined as the axial grayscale difference, and the absolute value of the grayscale difference between the pixel and the circumferentially adjacent pixels is determined as the circumferential grayscale difference. The maximum value of the axial grayscale difference and the circumferential grayscale difference is determined as the grayscale change intensity of the pixel. When the grayscale image is an eight-bit grayscale image, the value range of the grayscale change intensity is 0 to 255. All grayscale change intensities in the visible pipe wall area are arranged in numerical order, and the median is taken as the edge sharpness.

[0071] It should be noted that the median is used as the edge sharpness because local reflective edges of the pipe wall, sludge particle edges, and watermark edges can cause a small number of extreme grayscale changes. The median is derived from the grayscale distribution within the current image segment and can stably express the distinguishability of interface edges, pipe wall textures, and potential defect boundaries. Segments without visible pipe wall areas are not included in the edge sharpness calculation and are arranged after image segments with visible pipe wall areas in the segment sequence sorting.

[0072] Furthermore, area statistics were performed on the interference area and the complete range of the image segment. The area occupied by the reflective area, the area covered by the lens water, the area covered by mud, and the area covered by the water surface within the complete range of the image segment was taken as the interference area. The ratio between the interference area and the area of ​​the complete range of the image segment was taken as the interference area percentage.

[0073] It should be noted that the value of the interference area percentage ranges from zero to one. A value close to zero indicates that the image segment is less affected by reflections, lens water, mud, and water surface obstruction; a value close to one indicates that the image segment is mainly occupied by the interference area.

[0074] Based on the continuous visible area, edge clarity, proportion of interference area, and acquisition time, image segments with visible pipe wall areas are sorted lexicographically. Segments without visible pipe wall areas are appended to the end of the sorting to generate a sequence of interpretation segments. The sequence of interpretation segments is read, and the image segment ranked first is determined as the first interpretation segment, while the image segments ranked second and after are determined as candidate interpretation segments.

[0075] Furthermore, after obtaining the continuous visible area, edge clarity, and proportion of interference area, the image segments with visible pipe wall regions in the same pipe wall co-position frame group are sorted lexicographically to form a sequence of interpretation segments.

[0076] Specifically, the sequence of segments to be interpreted is represented as follows: ;

[0077] in, Indicates the first A sequence of segments for determining the location of each pipe wall. Indicates the pipe wall location number. Indicates containing the first The set of frame numbers corresponding to the correction frame images at each pipe wall location. Indicates to An operator that sorts the inner frame numbers lexicographically from largest to smallest according to the data within parentheses. Indicates the first Frame correction in the first frame image The continuous visible area at each pipe wall location Indicates the first Frame correction in the first frame image The clarity of the edges at each pipe wall location, Indicates the first Frame correction in the first frame image The percentage of interference area at each pipe wall location Indicates the first The acquisition time corresponding to the frame.

[0078] It should be noted that the segment sequence is sorted in lexicographical order, and the four types of data correspond to different meanings of image interpretation. The continuous visible area is used to ensure that there is a complete observable area at the same actual pipe wall location, which is the primary basis for subsequent image detection to extract complete candidate anomaly areas. The edge clarity is used to ensure that the pipe wall texture, interface edges, and potential anomaly boundaries can be identified, which is the second sorting criterion when the continuous visible area is the same. The proportion of interference area is used to reduce the impact of reflective areas, lens water-covered areas, sludge-covered areas, and water-covered areas on image detection, which is the third sorting criterion when the continuous visible area and edge clarity are the same. The acquisition time is only used to maintain the original video order when the first three items are the same, so as to avoid uncertainty in the sorting under the same conditions.

[0079] Furthermore, the sorting process is performed sequentially. Specifically, first, the continuous visible area is compared, and the image segment with the higher continuous visible area value is ranked first. If the continuous visible areas are the same, the edge sharpness is compared, and the image segment with the higher edge sharpness value is ranked first. If the edge sharpness is the same, the interference area ratio is compared, and the image segment with the lower interference area ratio value is ranked first. If the interference area ratio is the same, the acquisition time is compared, and the image segment acquired earlier is ranked first. After the image segments with visible pipe wall areas are sorted, the segments without visible pipe wall areas are appended to the end of the interpretation segment sequence according to the acquisition time.

[0080] It should be noted that the reason for using a negative sign in the formula for judging the segment sequence is as follows: higher values ​​for continuous visible area and edge clarity are more beneficial to image detection and should be ranked first in lexicographical order from largest to smallest; lower values ​​for interference area ratio are more beneficial to image detection and are therefore ranked in the opposite direction; the acquisition time is used to maintain the original video sequence when the first three items are exactly the same, so it is ranked in the opposite direction, so that image segments with earlier acquisition times are ranked first. The negative sign is only used to determine the ranking direction and does not indicate weight superposition or change the original data of the image segments.

[0081] Furthermore, for each co-located frame group of the pipe wall, the sequence of interpretation segments is read, and the image segment ranked first in the sequence of interpretation segments is determined as the preferred interpretation segment.

[0082] The image segments ranked second and subsequent in the sequence of interpreted segments are sequentially identified as candidate interpreted segments.

[0083] It should be noted that if there is only one image segment in the segment sequence, the candidate segment is empty.

[0084] S3. Based on the preferred reading segment, extract boundary candidates and region candidates, merge the boundary candidates and region candidates to generate candidate anomaly regions, determine the boundary of the candidate anomaly region, the area of ​​the candidate anomaly region and the candidate category, write the candidate anomaly region into the corresponding pipe wall co-position frame group, and establish a binding relationship.

[0085] Based on the preferred interpretation segment, grayscale changes are calculated within the candidate extraction range. Pixels with abrupt grayscale changes are extracted and skeletonized to generate an edge skeleton. The topological structure of the edge skeleton is analyzed and boundary attributes are assigned to form boundary candidates. Connectivity analysis of bright areas, dark areas, and color anomaly areas is performed on the grayscale image and the corrected color image within the candidate extraction range to generate region candidates. Boundary candidates and region candidates are geometrically merged under the coordinates of the unfolded image of the tube wall to generate candidate anomaly areas. The outer edge of the candidate anomaly area is extracted as the boundary of the candidate anomaly area, and the area of ​​the candidate anomaly area is calculated.

[0086] Furthermore, the preferred interpretation segment is read, and the pipe wall position number to which the preferred interpretation segment belongs, the frame number of the preferred interpretation segment, the range of the expanded pipe wall image, the coordinate mapping relationship, the grayscale image, and the corrected color image are read simultaneously.

[0087] The complete image range of the preferred interpretation segment is used as the candidate extraction range. The candidate extraction range does not deduct reflective areas, areas covered by water from the lens, areas obscured by mud, and areas obscured by the water surface.

[0088] Furthermore, the grayscale image is read within the candidate extraction range, and the grayscale change intensity of each pixel within the candidate extraction range is determined according to the method for determining the grayscale change intensity, thus forming a grayscale change image.

[0089] Furthermore, based on the distribution of gray-level change intensity within the preferred segment, the Otsu's method is used to generate edge segmentation boundaries. Pixels with gray-level change intensity higher than the edge segmentation boundaries are identified as gray-level abrupt change pixels. Connectivity analysis is performed on the gray-level abrupt change pixels according to the eight-neighborhood connectivity rule to obtain gray-level abrupt change connected regions. Skeletonization processing is then performed on the gray-level abrupt change connected regions to obtain the edge skeleton.

[0090] Furthermore, the number of adjacent skeleton pixels for each skeleton pixel in the edge skeleton is counted. Specifically, a skeleton pixel with only one adjacent skeleton pixel is denoted as an endpoint, and a skeleton pixel with at least three adjacent skeleton pixels is denoted as a bifurcation point. Boundary attributes are generated for each edge skeleton. When the edge skeleton has no endpoints, a closed attribute is added to the edge skeleton. When the edge skeleton has endpoints, an open attribute is added to the edge skeleton. When the edge skeleton has bifurcation points, a bifurcation attribute is added to the edge skeleton. When the edge skeleton intersects with or is adjacent to the interface loop, an interface-related attribute is added to the edge skeleton. The same edge skeleton is allowed to have multiple boundary attributes at the same time.

[0091] Furthermore, the edge skeletons with boundary attributes are recorded as boundary candidates, and the boundary coordinates, boundary length, endpoint position, bifurcation point position, boundary attributes, and the corresponding pipe wall position number are saved.

[0092] Furthermore, grayscale images and corrected color images are read within the candidate extraction range; based on the grayscale value distribution of the grayscale image, the maximum inter-class variance method is used to generate brightness segmentation boundaries, and pixels with grayscale values ​​higher than the brightness segmentation boundaries are identified as bright area pixels, while pixels with grayscale values ​​lower than the brightness segmentation boundaries are identified as dark area pixels.

[0093] Furthermore, a color difference image is formed based on the color difference between each pixel in the corrected color image and the median value of the color within the candidate extraction range. The corrected color image is represented using red, green, and blue channels, with each channel having a value range of 0 to 255. The median value of the color within the candidate extraction range is a color vector composed of the median values ​​of the red, green, and blue channels. The three-channel Euclidean distance between the color vector of each pixel and the median value of the color within the candidate extraction range is determined as the color difference of the pixel. Based on the color difference value distribution of the color difference image, the Otsu's method is used to generate the color difference segmentation boundary, and pixels with color difference values ​​higher than the color difference segmentation boundary are identified as pixels in the color anomaly region.

[0094] Furthermore, connectivity analysis is performed on bright area pixels, dark area pixels, and color anomaly area pixels respectively, and each connected region is recorded as a region candidate.

[0095] It should be noted that bright connected areas, dark connected areas, and color-abnormal connected areas are recorded as area candidates respectively; the area candidates retain the area boundary, area area, area source type, and the pipe wall location number to which they belong. The area source type includes bright connected areas, dark connected areas, and color-abnormal connected areas.

[0096] Furthermore, boundary candidates and region candidates are geometrically merged under the same pipe wall unfolded image coordinates. Specifically, when boundary candidates and region candidates have an intersection point, they are merged into one candidate anomaly region; when a boundary candidate is completely located inside a region candidate, they are merged into one candidate anomaly region; when multiple boundary candidates have a common endpoint or intersection point, they are merged into one candidate anomaly region; when multiple region candidates have a common boundary point, they are merged into one candidate anomaly region.

[0097] After merging, the outer edge of each candidate anomaly region is extracted as the candidate anomaly region boundary; the number of pixels within the area enclosed by the candidate anomaly region boundary is counted, and combined with the area of ​​a single pixel in the expanded image of the pipe wall, the area of ​​the candidate anomaly region is generated.

[0098] Based on the interface loop, pipe bottom direction, water surface boundary, floating object boundary, and candidate anomaly zone boundary, the positional relationship and morphological characteristics of each candidate anomaly zone are determined in the order of interface-related candidates, pipe bottom-related candidates, and pipe wall surface candidates to determine the candidate category. According to the pipe wall location number to which the candidate anomaly zone belongs, the pipe wall co-location frame group is read, and the candidate anomaly zone boundary, candidate anomaly zone area, candidate category, and preferred interpretation segment frame number are written into the pipe wall co-location frame group to establish a binding relationship.

[0099] Furthermore, candidate categories are determined sequentially based on the interface loop line, pipe bottom direction, water surface boundary, floating object boundary, candidate anomaly zone boundary, and candidate anomaly zone morphology.

[0100] Specifically, for each candidate anomaly area, the positional relationship between the boundary of the candidate anomaly area and the interface loop line, water surface boundary, floating object boundary and pipe bottom direction is first determined, and the candidate category is determined in the order of interface-related candidates, pipe bottom-related candidates, and pipe wall surface candidates.

[0101] Specifically, when there are open boundary candidates on both sides of the interface loop, and the endpoints of the open boundary candidates are misaligned in the circumferential position, the candidate category is recorded as misalignment; when the interface loop is divided into discontinuous segments by the boundary of the candidate anomaly area, and two opposing open boundary candidates are formed at the dividing position, the candidate category is recorded as disconnection; when the candidate anomaly area contains boundary candidates with bifurcation attributes, and the starting point of the bifurcation boundary candidates is located on the interface loop, the candidate category is recorded as root intrusion; when the candidate anomaly area is a regional candidate and the regional boundary intersects with the interface loop and extends away from the interface loop along the pipe axis, the candidate category is recorded as leakage; when the candidate anomaly area contains boundary candidates with closure attributes, and the closure boundary candidates intersect with the interface loop and extend into the pipe, the candidate category is recorded as branch pipe concealed connection.

[0102] When a candidate anomaly area does not enter the interface-related candidate judgment and is located in the corresponding area in the pipe bottom direction, it enters the pipe bottom-related candidate judgment. Specifically, when the outer edge of the candidate anomaly area is connected to the inner wall boundary of the pipe in the pipe bottom direction and forms an upper boundary that extends continuously along the pipe axis, the candidate category is recorded as sediment. When the candidate anomaly area contains a boundary candidate with closed properties, or when the outer edge of the candidate anomaly area forms an unclosed area boundary but the endpoints are all located on the inner wall boundary of the pipe in the pipe bottom direction, and the candidate anomaly area protrudes from the pipe bottom direction into the water passage space, the candidate category is recorded as obstacle. When the boundary of the candidate anomaly area intersects with the boundary of floating matter or the boundary of the candidate anomaly area shares an edge with the water surface boundary, the candidate category is recorded as scum.

[0103] When a candidate anomaly area does not enter the interface-related candidate judgment or the pipe bottom-related candidate judgment, it enters the pipe wall surface candidate judgment. Specifically, when the candidate anomaly area consists only of boundary candidates with open attributes, and the boundary candidates with open attributes extend continuously along the pipe wall unfolding direction, the candidate category is recorded as crack; when the candidate anomaly area contains both regional candidates and boundary candidates with closed attributes, and the boundary candidates with closed attributes are located inside the regional candidates or intersect with the boundaries of the regional candidates, and there are gray-scale abrupt connected regions inside the regional candidates, the candidate category is recorded as rupture; when the candidate anomaly area is formed by at least one of the regional candidates, such as bright connected regions, dark connected regions, or color-abnormal connected regions, and the candidate anomaly area does not intersect with the interface loop line, water surface boundary, or the corresponding region in the pipe bottom direction, the candidate category is recorded as corrosion; when the boundary of the candidate anomaly area deviates locally from the boundary of the pipe inner wall and replaces the original pipe inner wall boundary to form a new inner wall contour segment, the candidate category is recorded as deformation.

[0104] When the candidate anomaly area cannot be classified according to the interface-related candidates, pipe bottom-related candidates, and pipe wall surface candidates, the candidate category is recorded as an anomaly to be proven by contradiction, and the generation of the measured boundary of the candidate anomaly area, the generation of the three types of attribution reference boundaries, the calculation of the three types of attribution differences, and the determination of the attribution type are continued.

[0105] Furthermore, based on the pipe wall location number to which the candidate anomaly area belongs, the pipe wall co-location frame group is read, and the candidate anomaly area boundary, candidate anomaly area area, candidate category, and first-choice segment frame number are written into the corresponding pipe wall co-location frame group, forming a binding relationship between the candidate anomaly area and the pipe wall co-location frame group. The binding relationship is used to map the candidate anomaly area boundary to other correction frame images within the same pipe wall co-location frame group; the candidate anomaly area area is used to record the initial occupied range of the candidate anomaly area in the first-choice segment, and is saved as the initial area information of the candidate anomaly area when generating detection content, excluding records, or reviewing content.

[0106] S4. Based on the candidate anomaly zone boundary, candidate category, pipe wall location number, first choice interpretation segment frame number, pipe wall co-location frame group, interpretation segment sequence, and coordinate mapping relationship, generate the candidate anomaly zone co-location measured boundary and three types of attribution reference boundary, calculate the pipe wall attachment attribution difference, reflective shading interference attribution difference, and water surface dynamic interference attribution difference, determine the attribution type based on the unique minimum value of the attribution difference and trigger the candidate interpretation segment re-examination, and generate severity characterization value, detection content, exclusion record or review content.

[0107] Read the pipe wall co-position frame group and the judgment segment sequence according to the pipe wall location number. Based on the first judgment segment frame number and coordinate mapping relationship, map the candidate anomaly area boundary to each frame of the correction frame image in the same pipe wall co-position frame group to generate the mapping boundary. Re-extract the boundary candidate within the same pipe wall location range to generate the candidate anomaly area co-position measured boundary. Map the candidate anomaly area co-position measured boundary to the pipe wall unfolded image coordinates.

[0108] Based on the mapping boundary, the mapping boundary is used as the reference boundary for pipe wall attachment. The boundary of the reflective area, the boundary of the lens water-hanging area, and the boundary of the sludge-covered area are mapped to the coordinates of the unfolded pipe wall image to generate the reference boundary for reflective occlusion interference. The boundary of the water surface, the boundary of floating objects, and the boundary of water flow ripples are mapped to the coordinates of the unfolded pipe wall to generate the reference boundary for dynamic water surface interference, thus forming three types of reference boundaries.

[0109] Furthermore, the candidate anomaly zone boundary, candidate anomaly zone area, candidate category, corresponding pipe wall location number, and preferred interpretation segment frame number are read. Based on the corresponding pipe wall location number, the pipe wall co-located frame group and interpretation segment sequence are read.

[0110] Furthermore, the coordinate mapping relationship, the unfolded image of the pipe wall, the boundary of the reflective area, the boundary of the lens water-covered area, the boundary of the sludge-covered area, the water surface boundary, the boundary of the floating objects, and the boundary of the water flow ripples are read.

[0111] Furthermore, based on the preferred interpretation segment frame number and coordinate mapping relationship, the candidate anomaly region boundary is mapped to each corrected frame image within the same pipe wall co-position frame group, generating a mapping boundary.

[0112] Furthermore, in each corrected frame image, boundary candidates are re-extracted within the same pipe wall location range according to the boundary candidate extraction method; under the coordinates of the unfolded pipe wall image, the boundary overlap length, the boundary non-overlapping part length, and the boundary merging part length between each boundary candidate and the mapped boundary are calculated, and the boundary candidate with the non-zero boundary overlap length and the smallest ratio between the boundary non-overlapping part length and the boundary merging part length is recorded as the measured boundary of the candidate anomaly area.

[0113] If there are no candidate boundaries with a non-zero overlap length in the same correction frame image, they are recorded as empty boundaries in the same position of the candidate anomaly area. The empty boundaries in the same position of the candidate anomaly area are boundary records with a boundary length of zero. They are used to indicate that the candidate anomaly area is not reproduced in the same correction frame image and participate in the calculation of the attribution difference according to the empty boundary calculation rules in the attribution difference calculation process.

[0114] Furthermore, a reference boundary for pipe wall attachment is generated. Specifically, based on the coordinate mapping relationship, the boundary of the candidate anomaly area is mapped from the preferred interpretation segment to each corrected frame image within the pipe wall co-position frame group to obtain the mapped boundary, which is then used as the reference boundary for pipe wall attachment.

[0115] Furthermore, a reference boundary for attributing reflective occlusion interference is generated. Specifically, the boundaries of the reflective area, the lens water-laden area, and the sludge-occluded area in each corrected frame image are mapped to the coordinates of the unfolded pipe wall image. Boundaries located within the pipe wall position of the candidate anomaly area or intersecting with the measured boundary of the candidate anomaly area are retained. When multiple retained boundaries exist, the boundary with the smallest difference from the measured boundary of the candidate anomaly area is selected as the reference boundary for attributing reflective occlusion interference. When no retained boundaries exist, the difference in attribution of reflective occlusion interference is not calculated.

[0116] To generate a reference boundary for water surface dynamic disturbance attribution, specifically, the water surface boundary, floating object boundary, and water flow ripple boundary in each correction frame image are mapped to the coordinates of the unfolded pipe wall image. Boundaries located within the pipe wall position of the candidate anomaly area or intersecting with the measured boundary of the candidate anomaly area are retained. When multiple retained boundaries exist, the boundary with the smallest difference from the measured boundary of the candidate anomaly area is selected as the reference boundary for water surface dynamic disturbance attribution. When no retained boundaries exist, the difference in water surface dynamic disturbance attribution is not calculated.

[0117] It should be noted that if a certain calibration frame image does not have a reference boundary for the attribution of reflective occlusion interference, then the same calibration frame image will not participate in the calculation of the difference in the attribution of reflective occlusion interference; if a certain calibration frame image does not have a reference boundary for the attribution of dynamic water surface interference, then the same calibration frame image will not participate in the calculation of the difference in the attribution of dynamic water surface interference.

[0118] Based on the measured boundary of the candidate anomaly area and the three types of reference boundaries, calculate the differences in pipe wall attachment, reflective shading interference, and water surface dynamic interference. Compare the differences in pipe wall attachment, reflective shading interference, and water surface dynamic interference. When the difference in pipe wall attachment is the only minimum value, the candidate anomaly area is determined to be a pipe wall attachment anomaly.

[0119] When the difference in attribution due to reflective occlusion interference is the only minimum value, the candidate anomaly region extraction is re-executed by calling the alternative interpretation segment. The re-extracted candidate anomaly region is then subjected to the following processes: generation of the measured boundary of the candidate anomaly region, generation of the three types of attribution reference boundaries, calculation of the three types of attribution differences, and determination of the attribution type.

[0120] When the difference in the classification of dynamic disturbances on the water surface is the only minimum value and the candidate category is scum or obstacle, the candidate anomaly area is retained and recorded as a functional anomaly under the classification of dynamic disturbances on the water surface.

[0121] When the difference in the dynamic disturbance attribution of the water surface is the only minimum value and the candidate category is crack, rupture, misalignment, disconnection or leakage, the candidate anomaly area extraction is re-executed by calling the alternative interpretation segment. The re-extracted candidate anomaly area is then subjected to the following re-execution processes: generation of the measured boundary of the candidate anomaly area, generation of the three-category reference boundary, calculation of the difference in the three-category attribution, and determination of the attribution type.

[0122] When there is no unique minimum value for the differences in pipe wall adhesion attribution, the differences in the attribution of reflective shading interference, and the differences in the attribution of dynamic water surface interference, the review content is generated.

[0123] Furthermore, the attribution differences are calculated based on the measured boundaries of the candidate anomaly regions and the three types of attribution reference boundaries.

[0124] Specifically, the attribution difference is represented by the ratio between the non-overlapping boundary portion and the merged boundary portion, expressed as: ;

[0125] in, Indicates candidate anomaly region In terms of attribution type The difference in attribution, Indicates the candidate anomaly region number. Indicates the attribution type flag, with a value of , or , Indicates the attribution of pipe wall attachments. Indicates the attribution of reflective obstruction interference. Indicates the attribution of dynamic disturbances on the water surface. Indicates candidate anomaly region In the Frame correction of candidate anomaly regions in frame images with corresponding measured boundaries. Indicates candidate anomaly region In the Frame correction of the type of the frame image The corresponding attribution reference boundary, Indicates candidate anomaly region In terms of attribution type The set of frame numbers that participate in the calculation Indicates the non-overlapping boundary portion. Indicates the boundary merging part. This indicates the length of the boundary in the coordinates of the unfolded image of the pipe wall.

[0126] It should be noted that the attribution type The set of frame numbers participating in the calculation includes frame numbers where the measured boundary of the candidate anomaly region is not empty and the assigned reference boundary is not empty, as well as frame numbers where the measured boundary of the candidate anomaly region is empty and the assigned reference boundary is not empty; Attribution type The set of frame numbers participating in the calculation does not include frame numbers with empty home reference boundaries; Home type When the set of frame numbers participating in the calculation is empty, no attribution type is generated. The differences in attribution, attribution type It does not participate in the unique minimum value comparison.

[0127] It should be noted that when the measured boundary of the candidate anomaly region is empty and the assigned reference boundary is not empty, the non-overlapping part of the boundary is the assigned reference boundary, and the merged part of the boundary is the assigned reference boundary, with the single-frame boundary difference recorded as one; when the measured boundary of the candidate anomaly region is empty and the assigned reference boundary is empty, the same calibration frame image is not included in the attribution type. The set of frame numbers participating in the calculation; when the measured boundary of the candidate anomaly region is not empty and the assigned reference boundary is empty, the same corrected frame image is not included in the attribution type. The set of frame numbers that participate in the calculation.

[0128] The three types of attribution differences adopt the same boundary difference structure to allow for comparison of pipe wall adhesion attribution, reflective shading interference attribution, and water surface dynamic interference attribution on the same scale. The numerator represents the non-overlapping boundary length between the measured boundary of the candidate anomaly area and the attribution reference boundary, and the denominator represents the boundary length after merging the measured boundary of the candidate anomaly area and the attribution reference boundary. The attribution difference value ranges from zero to one. An attribution difference value close to zero indicates that the measured boundary of the candidate anomaly area and the attribution reference boundary are similar in position and shape; a value close to one indicates that the measured boundary of the candidate anomaly area and the attribution reference boundary are significantly different in position and shape.

[0129] The reason for adopting the boundary difference structure is that the boundaries of reflective areas, lens-mounted water areas, sludge-covered areas, water surfaces, floating objects, and water ripples in the endoscopic images of urban drainage pipe networks may all form boundary morphologies similar to defects. Simply comparing the detection confidence or single-frame area cannot distinguish between real pipe wall anomalies and image interference. The attribution difference formula directly compares which type of reference boundary the measured boundary of the candidate anomaly area is closer to, thereby determining whether the candidate anomaly area is more likely to reproduce with the pipe wall or change with reflective occlusion interference or dynamic water surface interference. The reason for using the median is that there may be local occlusion, single-frame reflection, or missing boundary extraction within the same frame group of the pipe wall. The median can reduce the impact of a single anomaly frame on the attribution judgment.

[0130] Furthermore, by comparing the three types of attribution differences involved in the calculation, when the pipe wall attachment attribution difference is the only minimum value, the candidate abnormal area is determined to be a pipe wall attachment abnormality, the candidate category is retained, and it enters the severity characterization value generation.

[0131] Furthermore, when the difference in the attribution of reflective occlusion interference is the only minimum value, the candidate anomaly area is determined to be attributed to reflective occlusion interference. The next candidate segment in the interpretation segment sequence is called, and the candidate anomaly area extraction is re-executed at the same pipe wall location. If the re-extracted candidate anomaly area overlaps with the original candidate anomaly area in the coordinates of the unfolded pipe wall image, the candidate anomaly area with the largest overlap area is selected, and the generation of the candidate anomaly area's in-situ measured boundary, the generation of the three types of attribution reference boundaries, the calculation of the three types of attribution differences, and the determination of the attribution type are re-executed. If no candidate anomaly area with overlapping areas is obtained through re-extraction, an exclusion record is generated. If there is no candidate segment in the interpretation segment sequence, the review content is generated.

[0132] Furthermore, when the difference in the classification of dynamic disturbances on the water surface is the only minimum value, candidate categories are read; when the candidate category is scum or obstacle, candidate anomaly areas are retained and recorded as functional anomalies under the classification of dynamic disturbances on the water surface; when the candidate category is crack, rupture, misalignment, disconnection or leakage, the next candidate judgment segment in the judgment segment sequence is called, and candidate anomaly area extraction is re-executed at the same pipe wall location.

[0133] When the re-extracted candidate anomaly area overlaps with the original candidate anomaly area in the coordinates of the unfolded pipe wall image, the candidate anomaly area with the largest overlapping area is selected to re-execute the on-site measured boundary generation, the three-category reference boundary generation, the three-category difference calculation, and the classification determination of the candidate anomaly area. When no candidate anomaly area with overlapping areas can be obtained by re-extraction, an exclusion record is generated, and the reason for exclusion is water surface dynamic interference classification. When there is no candidate interpretation segment in the interpretation segment sequence, the review content is generated.

[0134] Furthermore, when there is no unique minimum value among the attribution differences involved in the calculation, verification content is generated.

[0135] It should be noted that when comparing the differences in pipe wall adhesion attribution, the differences in ...

[0136] For candidate abnormal areas identified as pipe wall adhesion abnormalities and candidate abnormal areas recorded as functional abnormalities under the water surface dynamic disturbance classification, a severity characterization value is calculated. Combined with the candidate category, axial position, circumferential position, and classification type, detection content is generated. For candidate abnormal areas that do not reappear after re-examination of the alternative interpretation segment, an exclusion record is generated. For candidate abnormal areas without alternative interpretation segments, where the classification difference does not have a unique minimum value, and where pipe wall adhesion abnormalities still cannot be formed after re-examination of the alternative interpretation segment, review content is generated.

[0137] Furthermore, for candidate abnormal areas identified as pipe wall adhesion abnormalities and candidate abnormal areas recorded as functional abnormalities under the water surface dynamic interference classification, the area occupancy ratio of the candidate abnormal areas in each corrected frame image of the pipe wall co-position frame group is calculated, and the median of the area occupancy ratio is used as the severity characterization value.

[0138] Specifically, the severity characterization value expresses the degree to which an image occupies space, and is represented as: ;

[0139] in, Indicates candidate anomaly region The severity rating value, Candidate anomaly region In the The region in the frame-corrected image defined by the measured boundary of the candidate anomaly region. For the area element in the unfolded image of the pipe wall, Indicates the first Candidate anomaly regions in frame correction frame image The range of the image segment corresponding to the pipe wall location.

[0140] It should be noted that the severity rating ranges from zero to one. A value close to zero indicates that the candidate abnormal area occupies a low degree within the corresponding pipe wall location; a value close to one indicates that the candidate abnormal area occupies a high degree within the corresponding pipe wall location.

[0141] The severity characterization value is the ratio between the area of ​​the candidate anomaly region and the area of ​​the image segment of its corresponding pipe wall location. This ratio expresses the stable occupancy of the candidate anomaly region within its corresponding pipe wall location. The numerator is derived from the area determined by the measured boundary of the candidate anomaly region in each corrected frame image within the same pipe wall frame group, while the denominator is derived from the image segment range of the pipe wall location to which the candidate anomaly region belongs in the same corrected frame image. The reason for using the area ratio is that the absolute pixel area of ​​the candidate anomaly region cannot be directly compared under different pipe diameters, shooting distances, and unfolding scales. The area ratio can convert the candidate anomaly region into its occupancy relative to the range of its corresponding pipe wall location, allowing the severity of candidate anomalies such as cracks, ruptures, sediments, obstacles, and scum to be measured at a uniform scale. The reason for using the median is that the same pipe wall location may have local occlusion or edge loss in multiple frames. The median can reflect the stable occupancy across multiple frames, avoiding the deviation of the severity characterization value from the true image evidence caused by the area of ​​anomalies in a single frame.

[0142] Furthermore, for abnormal pipe wall adhesion, detection content is formed. Specifically, the detection content includes candidate category, axial position, circumferential position, severity characterization value, preferred interpretation segment frame number, and pipe wall adhesion attribution mark. The axial position and circumferential position are determined by the median position of the candidate abnormal area boundary in the pipe wall unfolded image coordinates.

[0143] Furthermore, for candidate anomaly areas recorded as functional anomalies under the water surface dynamic disturbance classification, detection content is generated. The detection content includes candidate category, axial position, circumferential position, severity characterization value, preferred interpretation segment frame number, and functional anomaly marker under the water surface dynamic disturbance classification.

[0144] Furthermore, for candidate abnormal areas that do not reappear after re-examination of the alternative interpretation segment, an exclusion record is formed. The exclusion record includes the axial position, circumferential position, candidate category, the type of classification that triggered the re-examination, and the reason for exclusion.

[0145] Furthermore, for candidate abnormal areas where there are no alternative interpretation segments, the attribution difference does not have a unique minimum value, and the alternative interpretation segments still cannot form abnormal pipe wall adhesion after re-examination, review content is formed. The review content includes axial position, circumferential position, candidate category, attribution type participating in the comparison, and review reason.

[0146] Furthermore, in this embodiment, the beneficial effects of the present invention are verified through simulation experiments. Specifically, an endoscopic detection simulation environment for urban drainage pipe networks is adopted. The simulation environment includes a circular drainage pipe, an interface loop, the pipe bottom direction, the water surface boundary, the floating object boundary, the water flow ripple boundary, the reflective area boundary, the lens water-covered area boundary, and the sludge-covered area boundary.

[0147] The simulation features a circular drainage pipe with an inner diameter of 800 mm, a pipe section length of 45 m, an interface spacing of 2.5 m, an image resolution of 1280 pixels × 720 pixels, an inter-frame travel distance of 0.18 m, a camera eccentricity of 18 mm, a camera pitch angle of 2 degrees, a motion blur of 0.14, and an image noise intensity of 0.032. The simulation materials include abnormal image materials related to pipe wall adhesion, reflective occlusion interference, dynamic water surface interference, and functional abnormal image materials under the dynamic water surface interference category. Abnormal image materials related to pipe wall adhesion include cracks, misalignments, and leaks. Reflective occlusion interference includes reflective areas, lens water-laden areas, and sludge-covered areas. Dynamic water surface interference includes water surface boundaries, water flow ripple boundaries, and floating object boundaries. Functional abnormal image materials under the dynamic water surface interference category include scum and obstacles.

[0148] Furthermore, Figure 5The improvement effect of the interpretation segment sequence on the basic stability of candidate anomaly region extraction was verified. Specifically, during the simulation, six calibration frame images were continuously generated at the same actual pipe wall location. Reflective areas, lens-mounted water areas, sludge-covered areas, and water-covered areas with different area proportions were superimposed on the six calibration frame images. For each calibration frame image, a corresponding image segment was extracted, recording the continuous visible area, edge sharpness, interference area proportion, and acquisition time. An interpretation segment sequence was formed according to the following order: continuous visible area from largest to smallest, edge sharpness from highest to lowest, interference area proportion from lowest to highest, and acquisition time from earliest to latest. The image segment ranked first was selected as the primary interpretation segment.

[0149] Figure 5 Simultaneously, the extraction completeness rate of candidate abnormal regions corresponding to the most recently acquired segments, random segments, and highlighted priority segments is recorded; specifically, Figure 5 The results show that when the proportion of interference area increases from 10% to 60%, the curve corresponding to the preferred interpretation segment is always higher than the curves corresponding to the most recently acquired segment, random segment, and highlight priority segment. This indicates that the interpretation segment sequence can select more suitable image segments for candidate anomaly area extraction from multiple image frames, so that the interpretation basis of the same actual pipe wall location no longer depends on a single image frame, thus improving the stability of the candidate anomaly area extraction process.

[0150] Figure 6 The study verified the role of the measured boundary of the candidate anomaly region in distinguishing between real anomalies and interference boundaries. Specifically, during the simulation, four types of image materials were arranged in the circular drainage pipe: pipe wall adhesion anomaly, reflection occlusion interference, water surface dynamic interference, and functional anomaly under the water surface dynamic interference. Continuous correction frame images were generated at six levels with interference area ratios ranging from 10% to 60%.

[0151] Furthermore, a pipe wall co-position frame group is first established based on the pipe wall unfolding image and coordinate mapping relationship. Then, the candidate anomaly area boundary in the first interpretation segment is mapped to each correction frame image in the same pipe wall co-position frame group. Boundary candidates are re-extracted within the same actual pipe wall position range to generate the candidate anomaly area co-position measured boundary. Figure 6 Record the proportion of overlapping lengths of co-located boundaries corresponding to different real-world scene types. The proportion of overlapping lengths of co-located boundaries is the percentage of the length of the overlapping boundary between the measured co-located boundary and the mapped boundary in the candidate anomaly area in the length of the mapped boundary.

[0152] Figure 6The data shows that the curves corresponding to pipe wall adhesion anomalies are generally higher than those corresponding to reflective shading interference, water surface dynamic interference, and functional anomalies under the water surface dynamic interference classification. This indicates that pipe wall adhesion anomalies such as cracks, misalignments, and leaks have stronger cross-frame reproducibility at the same actual pipe wall location. The curves corresponding to reflective shading interference and water surface dynamic interference are lower than those corresponding to pipe wall adhesion anomalies, indicating that reflections, lens water, sludge shading, water surface ripples, and floating object boundaries are difficult to stably reproduce at the same actual pipe wall location. Therefore, the boundaries corresponding to different classification types exhibit different reproduction characteristics within the same frame group. The measured boundaries of the candidate anomaly area can provide stable data for calculating the differences in pipe wall adhesion classification, reflective shading interference classification, and water surface dynamic interference classification, thereby improving the reliability of classification type determination.

[0153] In summary, this invention achieves cross-frame optimal interpretation of the same actual pipe wall location by forming a sequence of interpretation segments and determining the preferred interpretation segment and the candidate interpretation segment, thereby improving the stability of the candidate anomaly area extraction basis. By generating the measured boundary of the candidate anomaly area and three types of attribution reference boundaries, and triggering the re-examination of the candidate interpretation segment based on the attribution difference, it realizes the attribution determination of pipe wall adhesion anomalies, reflective occlusion interference, and water surface dynamic interference, generates detection content, exclusion records, or review content, thereby improving the reliability of endoscopic detection and the effectiveness of re-examination.

[0154] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An optimized method for endoscopic inspection of urban drainage pipe networks based on image detection, characterized in that, include: Acquire endoscopic videos of urban drainage pipe networks, extract continuous raw frame images, identify pipe inner wall boundaries, interface loops, pipe bottom direction, visible interference boundaries, and dynamic interference boundaries, and establish the relationship between pipe wall unfolded images and coordinate mapping; Based on the unfolded image of the pipe wall and the coordinate mapping relationship, a pipe wall co-position frame group is established. Based on the interference area formed by the visible interference boundary, the visible pipe wall area is determined. The continuous visible area, edge clarity and interference area ratio are calculated to form a sequence of interpretation segments. The preferred interpretation segment and the candidate interpretation segment are determined. Based on the preferred reading segment, boundary candidates and region candidates are extracted, and the boundary candidates and region candidates are merged to generate candidate anomaly regions. The boundaries, areas and categories of candidate anomaly regions are determined, and candidate anomaly regions are written into the corresponding pipe wall co-position frame group to establish a binding relationship. Based on the candidate anomaly zone boundary, candidate category, pipe wall location number, first choice interpretation segment frame number, pipe wall co-location frame group, interpretation segment sequence, and coordinate mapping relationship, the candidate anomaly zone co-location measured boundary and three types of attribution reference boundary are generated. The pipe wall attachment attribution difference, reflective shading interference attribution difference, and water surface dynamic interference attribution difference are calculated. The attribution type is determined based on the unique minimum value of the attribution difference and the candidate interpretation segment re-examination is triggered. The severity characterization value, detection content, exclusion record or review content are generated.

2. The optimized method for endoscopic inspection of urban drainage pipe networks based on image detection as described in claim 1, characterized in that, The establishment of the tube wall unfolding image and coordinate mapping relationship includes: The system collects endoscopic video of the urban drainage network, records the collection time and travel distance simultaneously, performs distortion correction on the original frame images, generates corrected frame images, and generates grayscale and corrected color images from the corrected frame images. Boundary recognition and connectivity analysis are performed on grayscale images and corrected color images to obtain the boundaries of the pipe inner wall, pipe center position, pipe bottom direction, water surface boundary, floating object boundary, water flow ripple boundary, reflective area boundary, and sludge-covered area boundary. Based on the inner wall boundary of the pipe, the center position of the pipe, the direction of the bottom of the pipe, and the travel distance, the pipe wall projection is unfolded to generate the unfolded image of the pipe wall, the coordinate mapping relationship, the interface loop, and the boundary of the water-holding area of ​​the lens.

3. The optimized method for endoscopic inspection of urban drainage pipe networks based on image detection as described in claim 2, characterized in that, The establishment of the pipe wall co-location frame group includes: Based on the unfolded tube wall image and coordinate mapping relationship, image segments corresponding to the same actual tube wall position in different calibration frame images are grouped into tube wall co-position frame groups.

4. The optimized method for endoscopic inspection of urban drainage pipe networks based on image detection as described in claim 1, characterized in that, The calculation of continuously visible area, edge sharpness, and interference area ratio includes: Based on the coordinate mapping relationship, the boundaries of the reflective area, the lens water-covered area, the sludge-covered area, and the water surface are mapped to the pipe wall unfolded image coordinates of each image segment in the pipe wall co-position frame group, and merged to form the interference area. The interference area is deducted from the complete range of the image segment to generate the visible pipe wall area. The visible pipe wall area is then subjected to connectivity analysis to generate continuous visible pipe wall area and no visible pipe wall area segments. The area of ​​the continuously visible pipe wall region is statistically analyzed to generate the continuously visible area. The median of the grayscale change intensity of the pixels within the visible pipe wall region is taken to generate the edge sharpness. The area of ​​the interference region and the complete range of the image segment are statistically analyzed to generate the interference area ratio.

5. The image detection-based endoscopic inspection optimization method for urban drainage pipe networks as described in claim 4, characterized in that, The determination of the preferred and alternative reading segments includes: Based on the continuous visible area, edge clarity, proportion of interference area, and acquisition time, image segments with visible pipe wall areas are sorted in lexicographical order, and segments without visible pipe wall areas are appended to the end of the sorting to generate a sequence of interpretation segments. Read the sequence of image segments to be interpreted, determine the image segment ranked first as the primary image segment to be interpreted, and determine the image segments ranked second and after the first as candidate image segments to be interpreted.

6. The optimized method for endoscopic inspection of urban drainage pipe networks based on image detection as described in claim 1 or 5, characterized in that, The step of extracting boundary candidates and region candidates based on the preferred read segment includes: Based on the preferred interpretation segment, grayscale changes are calculated within the candidate extraction range. Pixels with abrupt grayscale changes are extracted and skeletonized to generate an edge skeleton. The topological structure of the edge skeleton is analyzed and boundary attributes are assigned to form boundary candidates. Connectivity analysis of bright areas, dark areas, and color anomaly areas is performed on the grayscale image and the corrected color image within the candidate extraction range to generate region candidates.

7. The optimized method for endoscopic inspection of urban drainage pipe networks based on image detection as described in claim 6, characterized in that, The determination of the candidate anomaly region boundary, candidate anomaly region area, and candidate category includes: Boundary candidates and region candidates are geometrically merged under the coordinates of the unfolded pipe wall image to generate candidate anomaly regions. The outer edge of the candidate anomaly region is extracted as the boundary of the candidate anomaly region, and the area of ​​the candidate anomaly region is calculated. Based on the interface loop, pipe bottom direction, water surface boundary, floating object boundary, and candidate anomaly zone boundary, the positional relationship and morphological characteristics of each candidate anomaly zone are determined in the order of interface-related candidates, pipe bottom-related candidates, and pipe wall surface candidates to determine the candidate category. Based on the pipe wall location number to which the candidate anomaly area belongs, read the pipe wall co-location frame group, and write the candidate anomaly area boundary, candidate anomaly area area, candidate category, and preferred interpretation segment frame number into the pipe wall co-location frame group to establish a binding relationship.

8. The optimized method for endoscopic inspection of urban drainage pipe networks based on image detection as described in claim 1, characterized in that, The generated candidate anomaly region isotopic measured boundary and three types of attribution reference boundaries include: Read the pipe wall co-position frame group and the judgment segment sequence according to the pipe wall position number. Based on the first judgment segment frame number and coordinate mapping relationship, map the candidate anomaly area boundary to each frame of the correction frame image in the same pipe wall co-position frame group to generate the mapping boundary. Re-extract the boundary candidate within the same pipe wall position range to generate the candidate anomaly area co-position measured boundary. Map the candidate anomaly area co-position measured boundary to the pipe wall unfolded image coordinates. Based on the mapping boundary, the mapping boundary is used as the reference boundary for pipe wall attachment. The boundary of the reflective area, the boundary of the lens water-hanging area, and the boundary of the sludge-covered area are mapped to the coordinates of the unfolded pipe wall image to generate the reference boundary for reflective occlusion interference. The boundary of the water surface, the boundary of floating objects, and the boundary of water flow ripples are mapped to the coordinates of the unfolded pipe wall to generate the reference boundary for dynamic water surface interference, thus forming three types of reference boundaries.

9. The optimized method for endoscopic inspection of urban drainage pipe networks based on image detection as described in claim 8, characterized in that, The method of determining the attribution type based on the unique minimum value of the attribution difference includes: Based on the measured boundary of the candidate anomaly area and the three types of reference boundary, calculate the differences in pipe wall attachment, reflection shading interference, and water surface dynamic interference. Compare the differences in pipe wall attachment attribution, the differences in ... When the difference in the classification of reflective occlusion interference is the only minimum value, the candidate anomaly region extraction is re-executed by calling the alternative interpretation segment. The candidate anomaly region is then re-executed to generate the measured boundary of the candidate anomaly region, generate the three types of classification reference boundaries, calculate the three types of classification differences, and determine the classification type. When the difference in the attribution of dynamic disturbances on the water surface is the only minimum value and the candidate category is scum or obstacle, the candidate anomaly area is retained and recorded as a functional anomaly under the attribution of dynamic disturbances on the water surface. When the difference in the dynamic disturbance attribution of the water surface is the only minimum value and the candidate category is crack, rupture, misalignment, disconnection or leakage, the candidate anomaly area extraction is re-executed by calling the alternative interpretation segment. The candidate anomaly area is then re-executed to generate the measured boundary of the candidate anomaly area, generate the reference boundary of the three categories, calculate the difference of the three categories, and determine the attribution type. When there is no unique minimum value for the differences in pipe wall adhesion attribution, the differences in the attribution of reflective shading interference, and the differences in the attribution of dynamic water surface interference, the review content is generated.

10. The optimized method for endoscopic inspection of urban drainage pipe networks based on image detection as described in claim 9, characterized in that, The generated severity characterization value, detection content, exclusion record, or review content includes: For candidate abnormal areas identified as pipe wall adhesion abnormalities and candidate abnormal areas recorded as functional abnormalities under the water surface dynamic disturbance classification, a severity characterization value is calculated. Combined with the candidate category, axial position, circumferential position, and classification type, detection content is generated. For candidate abnormal areas that do not reappear after re-examination of the alternative interpretation segment, an exclusion record is generated. For candidate abnormal areas without alternative interpretation segments, where the classification difference does not have a unique minimum value, and where pipe wall adhesion abnormalities still cannot be formed after re-examination of the alternative interpretation segment, review content is generated.

Images