CN-122023317-A - Method, storage medium and system for identifying cutting defects of building steel

CN122023317ACN 122023317 ACN122023317 ACN 122023317ACN-122023317-A

Abstract

The invention provides a method, a storage medium and a system for identifying cutting defects of building steel, wherein the method comprises the steps of calculating a curvature tensor field of a first image, constructing a curvature-driven diffusion evolution model, setting the diffusion strength of the model to be in direct proportion to curvature gradient, enabling the diffusion direction to be along the main direction of curvature, iteratively executing a diffusion process until the curvature tensor field reaches an equilibrium state, identifying curvature extremum points as candidate defect seed points, generating a second image, carrying out image enhancement processing on the second image, generating a third image, calculating a coherent group of the third image, recording the quantity of connected components and the quantity of holes, carrying out morphological expansion and corrosion operation under the constraint of the coherent group, generating a fourth image, carrying out space continuity verification on the fourth image, screening out a connected region meeting preset requirements as a cutting defect region, and finally generating a cutting defect identification result, so that under the interference of complex optical phenomena, the cutting defects of the building steel can still be accurately identified.

Inventors

WU ZHUOFAN
HUANG HAICHENG
DENG QIANHUA
LIAO ZILIANG
DENG YICHENG
LI JIAWEI

Assignees

广东众城天弘建设有限公司

Dates

Publication Date: 20260512
Application Date: 20260127

Claims (10)

1. A method for identifying cutting defects of a construction steel, comprising: Acquiring an original image of a cutting surface of a building steel to be identified, and carrying out normalization processing on the original image to obtain a first image; Calculating a curvature tensor field of the first image, constructing a curvature-driven diffusion evolution model, setting the diffusion strength of the model to be in direct proportion to a curvature gradient, and iteratively executing a diffusion process until the curvature tensor field reaches an equilibrium state, identifying a curvature extremum point of the curvature tensor field in the equilibrium state as a candidate defect seed point, and generating a second image containing the candidate defect seed point; Performing image enhancement processing on the second image to generate a third image, calculating a coherent group of the third image, recording the number of connected components and the number of holes, and performing morphological expansion and corrosion operation under the constraint of the coherent group to generate a fourth image; and carrying out space continuity verification on the fourth image, screening out a connected region meeting preset requirements as a cutting defect region, generating a cutting defect positioning map, and generating a cutting defect identification result according to the cutting defect positioning map.
2. The method of claim 1, wherein normalizing the original image to obtain a first image comprises: Acquiring motion trail parameters of cutting equipment, converting the motion trail parameters into a space coordinate sequence, and extracting sequence direction vectors as guiding references; Constructing an illumination compensation field guided by a cutting track, decomposing pixel-by-pixel brightness components of the original image along a guide reference based on the illumination compensation field, generating tangential illumination components and normal illumination components, performing local contrast suppression processing on the normal illumination components, and re-fusing the tangential illumination components and the normal illumination components after the suppression processing to obtain a first image.
3. The method of claim 2, wherein said constructing a cutting trajectory-guided illumination compensation field comprises: Constructing an illumination field matrix with the same size as the original image, and initializing all element values of the illumination field matrix into unit values; carrying out spline interpolation on the space coordinate sequence to generate a continuous cutting track curve, and calculating tangential direction angles at each point on the curve; And constructing a mapping relation table of the tangential direction angle and the pixel coordinates, and assigning a value to the illumination field matrix according to the mapping relation table, so that the value of each position in the illumination field matrix is associated with the cosine value of the corresponding tangential direction angle, and generating an illumination compensation field.
4. The method of claim 1, wherein the determination that the curvature tensor field reaches an equilibrium state comprises: updating the curvature tensor field in each iteration, and calculating the gradient modular length of the latest curvature tensor field; And stopping iteration when the gradient modular length is smaller than a preset threshold value, and judging that the curvature tensor field reaches an equilibrium state when the iteration is stopped.
5. The method of claim 1, wherein performing image enhancement processing on the second image to generate a third image comprises: Performing multi-scale decomposition on the second image, and calculating the response intensity of the characteristic channel on each scale; And sorting the characteristic channels according to the response intensity, reserving channels with preset proportion, which are sorted in front, as main channels, exponentially amplifying response values of the main channels, performing power attenuation on response values of non-main channels, and accumulating all processed channel response images pixel by pixel to generate a third image.
6. The method of claim 1, wherein the performing morphological dilation and erosion operations under the coherent population constraint generates a fourth image comprising: calculating the zero-dimensional Betty number and the one-dimensional Betty number of the third image, wherein the zero-dimensional Betty number corresponds to the number of connected components, the one-dimensional Betty number corresponds to the number of holes, and the Betty number before reconstruction is recorded; constructing an adaptive structural element, wherein the shape of the structural element is adaptively selected according to the local curvature of the third image; Performing morphological opening operation on the third image based on the self-adaptive structural element, performing expansion operation after performing corrosion operation, recording coordinate set of removed pixel points in the corrosion process, preferentially recovering pixel points in the recorded coordinate set in the expansion process, and ensuring that the recovery operation does not change the quantity of connected components, or Performing morphological closing operation on the third image, performing corrosion operation after performing expansion operation, marking newly added pixels in the expansion process, and reserving marked pixels in contact with the hole boundaries in the corrosion process to ensure that the number of holes is not reduced; And recalculating the Betty number of the third image after each morphological operation, comparing the Betty numbers before and after the operation, if the Betty number is changed, backing back the operation, and re-executing the operation by selecting smaller structural element sizes until the Betty number before and after the operation is kept unchanged, so as to generate a fourth image.
7. The method according to claim 1, wherein the performing spatial continuity verification on the fourth image, and screening out the connected region meeting the preset requirement as the cutting defect region, includes: performing binarization processing on the fourth image to generate a fifth image; Extracting all foreground communication areas in the fifth image, performing skeleton extraction and iterative deletion of boundary pixels on each foreground communication area, and keeping connectivity unchanged to obtain a skeleton line with single pixel width; Detecting branch points and end points on a skeleton line, wherein the branch points are pixel points with the number of neighbors being greater than two in a pixel neighborhood, the end points are pixel points with the number of neighbors being equal to one, and counting the total number of the skeleton branch points and the total number of the end points of each connected region; And calculating the ratio of the total number of branch points to the total number of endpoints, comparing the ratio with a preset interval, and screening a communication area with the ratio falling in the preset interval as a cutting defect area meeting the requirement of space continuity.
8. The method of claim 1, wherein generating a cutting defect identification result from the cutting defect localization map comprises: And matching the cutting defect positioning map with a building steel standard template, calculating the geometric attribute of a cutting defect region, classifying and marking the cutting defect according to the geometric attribute, and generating a cutting defect identification result, wherein the geometric attribute comprises an area, a perimeter and a shape.
9. A system for identifying cutting defects in a construction steel, comprising: the acquisition module is used for acquiring an original image of a cutting surface of the building steel to be identified, and carrying out normalization processing on the original image to obtain a first image; The computing module is used for computing a curvature tensor field of the first image, constructing a curvature-driven diffusion evolution model, setting the diffusion strength of the model to be in direct proportion to a curvature gradient and the diffusion direction to be along the main direction of curvature, iteratively executing a diffusion process until the curvature tensor field reaches an equilibrium state, identifying a curvature extreme point of the curvature tensor field as a candidate defect seed point in the equilibrium state, and generating a second image containing the candidate defect seed point; The image enhancement module is used for carrying out image enhancement processing on the second image to generate a third image, calculating a coherent group of the third image, recording the quantity of connected components and the quantity of holes, and carrying out morphological expansion and corrosion operation under the constraint of the coherent group to generate a fourth image; The generation module is used for carrying out space continuity verification on the fourth image, screening out a connected region meeting preset requirements as a cutting defect region, generating a cutting defect positioning map, and generating a cutting defect identification result according to the cutting defect positioning map.
10. A storage medium having stored thereon a computer program which, when executed by a processor, implements the method for identifying a cutting defect of a construction steel material according to any one of claims 1 to 8.

Description

Method, storage medium and system for identifying cutting defects of building steel Technical Field The invention relates to the technical field of cutting quality detection of building steel, in particular to a method, a storage medium and a system for identifying cutting defects of building steel. Background The cutting processing quality of the building steel serving as a core bearing material of modern steel structure buildings, bridges and major infrastructures is directly related to the process stability of subsequent welding and assembly and the safety performance of the whole structure. The presence of the cut surface defects will significantly reduce the effective fusion area of the weld, induce stress concentrations, and even lead to early failure of the structural member. Along with the promotion of building industrialization and intelligent manufacturing, the traditional quality management method relying on manual visual sampling inspection has difficulty in meeting the requirements of production line beats and precision, and an automatic detection technology based on machine vision becomes a necessary development direction. At present, aiming at the detection of the cutting defects of the metal plate, the detection method is mainly based on an image detection mode of edge detection and threshold segmentation, and has certain applicability in the detection of a relatively flat and uniform-illumination thin plate, but for the building steel with large thickness and complex reflection characteristics, the cutting surface of the building steel always presents complex optical phenomena of strong specular reflection, strong grain directivity, uneven local illumination and the like, so that the false recognition rate of the cutting defects of the building steel under the threshold segmentation is obviously increased. Disclosure of Invention The invention provides a method, a storage medium and a system for identifying cutting defects of building steel, which can still ensure that the cutting defects of the building steel are accurately identified under the interference of complex optical phenomena. In order to solve the problems, the invention adopts the following technical scheme: The invention provides a method for identifying cutting defects of building steel, which comprises the following steps: Acquiring an original image of a cutting surface of a building steel to be identified, and carrying out normalization processing on the original image to obtain a first image; Calculating a curvature tensor field of the first image, constructing a curvature-driven diffusion evolution model, setting the diffusion strength of the model to be in direct proportion to a curvature gradient, and iteratively executing a diffusion process until the curvature tensor field reaches an equilibrium state, identifying a curvature extremum point of the curvature tensor field in the equilibrium state as a candidate defect seed point, and generating a second image containing the candidate defect seed point; Performing image enhancement processing on the second image to generate a third image, calculating a coherent group of the third image, recording the number of connected components and the number of holes, and performing morphological expansion and corrosion operation under the constraint of the coherent group to generate a fourth image; and carrying out space continuity verification on the fourth image, screening out a connected region meeting preset requirements as a cutting defect region, generating a cutting defect positioning map, and generating a cutting defect identification result according to the cutting defect positioning map. Preferably, the normalizing the original image to obtain a first image includes: Acquiring motion trail parameters of cutting equipment, converting the motion trail parameters into a space coordinate sequence, and extracting sequence direction vectors as guiding references; Constructing an illumination compensation field guided by a cutting track, decomposing pixel-by-pixel brightness components of the original image along a guide reference based on the illumination compensation field, generating tangential illumination components and normal illumination components, performing local contrast suppression processing on the normal illumination components, and re-fusing the tangential illumination components and the normal illumination components after the suppression processing to obtain a first image. Preferably, the constructing the illumination compensation field guided by the cutting track includes: Constructing an illumination field matrix with the same size as the original image, and initializing all element values of the illumination field matrix into unit values; carrying out spline interpolation on the space coordinate sequence to generate a continuous cutting track curve, and calculating tangential direction angles at each point on the curve; And constructing a mapping relation tab