CN-121982000-A - Pipeline construction monitoring method and system based on image recognition

CN121982000ACN 121982000 ACN121982000 ACN 121982000ACN-121982000-A

Abstract

The invention discloses a pipeline construction monitoring method and a system based on image recognition, which belong to the technical field of pipeline construction and comprise the steps of projecting a light spot array to a monitoring scene containing welding seams, analyzing the displacement of the light spot array in an image sequence to analyze affine transformation parameters to generate a global motion flow field, utilizing the global motion flow field to carry out reverse compensation on the image sequence to generate a physical calibration sequence, carrying out signal analysis on each pixel in a time dimension on the physical calibration sequence to generate a time domain modulation intensity map, carrying out morphological expansion and connected domain analysis on the time domain modulation intensity map, and realizing double interference of high-concentration metal dust and equipment high-frequency vibration in underground shield tunnel construction, realizing jitter reverse compensation through analyzing the motion parameters of the light spot array, eliminating dust shielding influence by combining the time domain analysis and texture reconstruction technology, and avoiding the problem that the defect characteristics are erased by mistake in the noise removal process in the traditional method.

Inventors

MA DEYI
BAI WENJIANG
Liu Rongshuang
RAO JIANG
CHENG RUIRUI
HAN SIYU

Assignees

中交二公局铁路建设有限公司

Dates

Publication Date: 20260505
Application Date: 20260127

Claims (10)

1. The pipeline construction monitoring method based on image recognition is characterized by comprising the following steps of: projecting a light spot array to a monitoring scene containing a welding line, analyzing the displacement of the light spot array in an image sequence to analyze affine transformation parameters and generating a global motion flow field; Performing reverse compensation on the image sequence by using the global motion flow field to generate a physical calibration sequence; Performing signal analysis on each pixel in the time dimension on the physical calibration sequence to generate a time domain modulation intensity map, and performing morphological expansion and connected domain analysis on the time domain modulation intensity map to generate a binary dust mask; Marking invalid data in a physical calibration sequence by utilizing a binarization dust mask, reconstructing textures of a marked area by solving an anisotropic diffusion equation based on unmarked valid pixel information and geometric prior of a welding line, and outputting an image sequence after texture reconstruction; calculating an inter-frame difference map on the image sequence after texture reconstruction, establishing a normal deformation prediction model based on historical data of a global motion flow field, subtracting the model from the inter-frame difference map to obtain an abnormal motion residual map, and dividing the abnormal motion residual map to obtain a preliminary defect region; Screening based on the historical distribution of the binarized dust mask and the topological relation of the preliminary defect area relative to the welding line, and outputting a defect marking map.
2. The image recognition-based pipeline construction monitoring method according to claim 1, wherein the generation of the global motion flow field comprises: Projecting a composite light spot pattern to the monitoring scene, wherein the composite light spot pattern is formed by superposing a main frequency sinusoidal stripe and a Morle circular spot; Carrying out local micropattern analysis on the moire circular spot area, and calculating the absolute phase value of each circular spot through two-dimensional phase unwrapping operation; Matching between continuous frames based on absolute phase values, predicting a search area according to motion information of a preamble frame, matching in the search area, and verifying matching point pairs based on phase continuity and motion consistency to filter abnormal matching so as to obtain a displacement observation set; And (3) parallelly fitting a plurality of motion models by using the pure displacement observation set, calculating the space-time distribution characteristics of residual errors of the plurality of motion models, and selecting a final model from the plurality of motion models based on the fitting goodness of each motion model and the space-time distribution characteristics of the residual errors to generate a global motion flow field.
3. The image recognition-based pipe construction monitoring method of claim 2, wherein the generation of the physical calibration sequence comprises: generating a self-adaptive sampling grid according to the model type of the global motion flow field, reversely mapping the grid point to an original image according to the global motion flow field, and interpolating; Analyzing residual fluctuation of each position of the remapped image on a time sequence based on a residual distribution diagram corresponding to the global motion flow field, calculating pixel-level motion calibration confidence and marking an absolute reliable anchor point; Performing an anisotropic diffusion process constrained by pixel-level motion calibration confidence with an absolute reliable anchor point and pixels with confidence above a first predetermined threshold as boundaries, repairing areas with confidence below a second predetermined threshold; and comparing pixel value differences of overlapping areas of pixels with confidence degrees higher than a first preset threshold value of adjacent frames, performing time sequence smoothing filtering on areas with discontinuous jumps, and outputting a physical calibration sequence.
4. The image recognition-based pipe construction monitoring method of claim 3, wherein the generation of the time-domain modulation intensity map comprises: Preprocessing the pixel time-course signals based on the pixel-level motion calibration confidence coefficient to extract differential time-course signals containing abnormal fluctuation, and carrying out matching tracking on the differential time-course signals to obtain a group of time-frequency atoms; calculating the distribution entropy of time-frequency atoms in a time-frequency domain and the energy aggregation degree of the time-frequency atoms, and generating a pixel-level modulation intensity scalar in a fusion way; And taking the modulation intensity scalar as an anchor point, and interpolating under the constraint of the image gradient and the confidence map to generate a time domain modulation intensity map.
5. The image recognition-based pipe construction monitoring method of claim 4, wherein the generation of the binarized dust mask comprises: Based on the time domain modulation intensity map, performing local threshold segmentation by taking pixel-level motion calibration confidence as an adjustment basis to generate a preliminary binary drawing; Performing time domain persistence verification and morphological purification on the connected region in the preliminary binary drawing; And performing cross-frame fusion on the purified regional mask, and trimming the fusion result according to the confidence level diagram to generate a binarization dust mask.
6. The image recognition-based pipe construction monitoring method of claim 5, wherein reconstructing textures of the marked region by solving an anisotropic diffusion equation, outputting a texture reconstructed image sequence, comprising: calculating a structure tensor field in an effective area of the physical calibration sequence, and performing anisotropic smoothing; performing tensor voting on the binarized dust mask area by taking the smoothed structure tensor field as a source to generate an reasoning structure tensor field covering the whole graph; Constructing an anisotropic diffusion equation based on the rational structure tensor field, carrying out iterative solution by taking the pixel value of the effective area as a boundary condition, and reconstructing a texture substrate of the ineffective area; And calculating dense optical flow fields of the current frame and the adjacent frames on the effective area, projecting texture details of the adjacent frames to the ineffective area of the current frame based on the dense optical flow fields, combining pixel-level motion calibration confidence and closing errors of optical flow projection, and outputting an image sequence after texture reconstruction.
7. The image recognition-based pipe construction monitoring method of claim 6, wherein the generating of the abnormal motion residual map comprises: Performing space-time decomposition on the historical global motion flow field sequence, extracting a spatial mode and a time evolution rule thereof, synthesizing a periodic predictive deformation field and a trend predictive deformation field, and superposing to generate a predictive global deformation field; Anisotropic modulation is carried out on the predicted global deformation field in a welding line area by utilizing the geometric prior of the welding line, and a physical constraint deformation predicted field is generated; And carrying out multi-resolution pyramid analysis on the physical constraint deformation prediction field and the inter-frame difference map, and generating an abnormal motion residual map by fusing all scales of residuals in a trans-scale manner.
8. The image recognition-based pipe construction monitoring method of claim 7, wherein the acquiring of the preliminary defect area comprises: determining an abnormal seed point based on the abnormal motion residual map, and performing competitive region growth by taking the abnormal seed point as a starting point to generate candidate defect regions and competition strength scores; Based on the competition strength score, verifying the space-time consistency of each candidate defect region in the abnormal motion residual diagram of the continuous multiframe, and screening to obtain a preliminary defect region.
9. The image recognition-based pipe construction monitoring method of claim 8, wherein the generation of the defect marking map comprises: Comprehensively verifying candidate defect areas from three dimensions of dust coverage history, geometric form and topological relation and time-sequence causal consistency; integrating the comprehensive verification result and the competition strength score to generate a defect marking map marked with defect types and detection confidence.
10. The image recognition-based pipeline construction monitoring system applied to the image recognition-based pipeline construction monitoring method as claimed in any one of claims 1 to 9, characterized by comprising: the flow field analysis module is used for projecting a light spot array to a monitoring scene containing a welding seam, analyzing the displacement of the light spot array in an image sequence to analyze affine transformation parameters and generating a global motion flow field; The image calibration module is used for carrying out reverse compensation on the image sequence by utilizing the global motion flow field to generate a physical calibration sequence; The mask manufacturing module is used for performing signal analysis on each pixel in the time dimension on the physical calibration sequence to generate a time domain modulation intensity map, and performing morphological expansion and connected domain analysis on the time domain modulation intensity map to generate a binary dust mask; the texture restoration module is used for marking invalid data in the physical calibration sequence by utilizing the binarization dust mask, reconstructing textures of a marked area by solving an anisotropic diffusion equation based on unlabeled valid pixel information and geometric prior of a welding line, and outputting an image sequence after texture reconstruction; The defect primary screening module calculates an inter-frame difference map on the image sequence after texture reconstruction, establishes a normal deformation prediction model based on historical data of a global motion flow field, deducts the model from the inter-frame difference map to obtain an abnormal motion residual map, and divides the abnormal motion residual map to obtain a primary defect region; And the defect fine screening module is used for screening based on the historical distribution of the binarized dust mask and the topological relation of the primary defect area relative to the welding line, and outputting a defect marking map.

Description

Pipeline construction monitoring method and system based on image recognition Technical Field The invention relates to the technical field of pipeline construction, in particular to a pipeline construction monitoring method and system based on image recognition. Background The image recognition technology has the advantages of non-contact, high efficiency and automatic detection, is widely applied to defect detection of weld cracks, air holes and the like, gradually replaces the traditional manual visual mode, and is also applied to pipeline construction weld detection, however, the underground shield tunnel construction environment is special, high-concentration metal dust generated in some interferences is easy to adhere to a lens of acquisition equipment, so that the image is covered with noise points to mask the outline and fine defect characteristics of the weld, and high-frequency vibration in shield machine propulsion can cause equipment shake to cause dislocation and distortion between frames of an image sequence. The existing denoising scheme mostly adopts a strong spatial filtering algorithm, and although the method can effectively eliminate dust interference, fine defect characteristics such as cracks, air holes and the like are easy to erase at the same time, so that the problem of omission is caused; in the aspect of jitter compensation, the mainstream technology relies on inter-frame feature point matching to realize image alignment, however dust noise points are often misjudged as effective feature points, so that the alignment accuracy is seriously reduced, negative effects caused by jitter cannot be eliminated, and distortion effects caused by noise are aggravated, under the dual interference, if the image resolution is simply improved to capture microscopic defects, not only motion artifacts are synchronously amplified, but also alignment distortion caused by the dust noise points is aggravated, and finally defect signals are completely covered by noise. Disclosure of Invention Aiming at the problems existing in the prior art, the invention aims to provide a pipeline construction monitoring method and system based on image recognition, which can inhibit double interference of high-concentration metal dust and equipment high-frequency vibration in underground shield tunnel construction, realize jitter reverse compensation by analyzing motion parameters through a light spot array, eliminate dust shielding influence by combining time domain analysis and texture reconstruction technology, and avoid the problem that defect features are erased by mistake in the noise removal process of the traditional method. In order to solve the problems, the invention adopts the following technical scheme: in a first aspect, a method for monitoring construction of a pipeline based on image recognition, the method comprising: projecting a light spot array to a monitoring scene containing a welding line, analyzing the displacement of the light spot array in an image sequence to analyze affine transformation parameters and generating a global motion flow field; Performing reverse compensation on the image sequence by using the global motion flow field to generate a physical calibration sequence; Performing signal analysis on each pixel in the time dimension on the physical calibration sequence to generate a time domain modulation intensity map, and performing morphological expansion and connected domain analysis on the time domain modulation intensity map to generate a binary dust mask; Marking invalid data in a physical calibration sequence by utilizing a binarization dust mask, reconstructing textures of a marked area by solving an anisotropic diffusion equation based on unmarked valid pixel information and geometric prior of a welding line, and outputting an image sequence after texture reconstruction; calculating an inter-frame difference map on the image sequence after texture reconstruction, establishing a normal deformation prediction model based on historical data of a global motion flow field, subtracting the model from the inter-frame difference map to obtain an abnormal motion residual map, and dividing the abnormal motion residual map to obtain a preliminary defect region; Screening based on the historical distribution of the binarized dust mask and the topological relation of the preliminary defect area relative to the welding line, and outputting a defect marking map. Further, the generating of the global motion flow field includes: Projecting a composite light spot pattern to the monitoring scene, wherein the composite light spot pattern is formed by superposing a main frequency sinusoidal stripe and a Morle circular spot; Carrying out local micropattern analysis on the moire circular spot area, and calculating the absolute phase value of each circular spot through two-dimensional phase unwrapping operation; Matching between continuous frames based on absolute phase values, predicting a search are