CN-121236214-B - Post-processing optimization and quantification method, device and equipment for fracture segmentation image

CN121236214BCN 121236214 BCN121236214 BCN 121236214BCN-121236214-B

Abstract

The application discloses a post-processing optimization and quantization method, a device and equipment for a crack segmentation image, and relates to the technical field of computer vision and digital image processing, wherein the method comprises the steps of obtaining an initial binarization crack mask image corresponding to an image to be processed, and performing geometric form filtering operation to remove non-crack areas so as to obtain a preliminary filtered mask image; and performing breakpoint connection processing on the mask map of the region of interest to repair discontinuous crack fragments, and generating a continuous optimized crack mask map to calculate geometric size parameters of the cracks. The application can obviously improve the accuracy and the integrity of the fracture segmentation result and realize the automation and the accurate quantification of key parameters.

Inventors

ZHAO XUNGANG
LI OU
JIANG YU
ZHANG MANGMANG
QU JINYU
WANG LONG
SUN HEMING
LIU SHIYANG
ZHOU QIANG
WANG ZIYU
WANG JIANGSHENG
ZHANG DUANYANG
ZHONG JIWEI
WANG BO
YAO JINXIN
Duan Jingxiang
WU JUFENG
WU XIANZHI

Assignees

中铁大桥局集团有限公司
国能新朔铁路有限责任公司大准铁路分公司
中铁大桥科学研究院有限公司

Dates

Publication Date: 20260512
Application Date: 20251128

Claims (5)

1. A post-processing optimizing and quantifying method for a slit segmentation image, characterized in that the post-processing optimizing and quantifying method for a slit segmentation image comprises: obtaining an initial binary crack mask image corresponding to an image to be processed, and performing geometric form filtering operation to remove a non-crack region to obtain a mask image after preliminary filtering; Performing contour detection on the mask map after preliminary filtering, and identifying and cutting to obtain a mask map of an interested region containing cracks; Performing breakpoint connection processing on the region of interest mask map to repair discontinuous crack fragments, and generating a continuous optimized crack mask map to calculate geometric size parameters of cracks; the geometric form filtering operation is performed to remove the non-crack area, and specifically comprises the following steps: Analyzing the initial binarization crack mask map, extracting a crack contour to obtain crack pixel connected domains, and obtaining the minimum circumscribed rectangle of each crack pixel connected domain; Calculating to obtain the area of each crack pixel connected domain and the aspect ratio of the minimum circumscribed rectangle of each crack pixel connected domain; judging whether the area of the current crack pixel connected domain is smaller than a preset area threshold value: If yes, eliminating the current crack pixel connected domain; If not, judging whether the aspect ratio of the minimum circumscribed rectangle of the current crack pixel connected domain is smaller than a preset aspect ratio threshold, if so, eliminating the current crack pixel connected domain, and if not, reserving the current crack pixel connected domain; performing breakpoint connection processing on the region mask map of interest to repair discontinuous crack fragments, wherein the breakpoint connection processing is performed by adopting a curve fitting connection mode based on track points; the curve fitting connection mode based on the track points specifically comprises the following steps: scanning the mask map of the region of interest row by row or column by column along the main direction of the crack, extracting the centers or centroids of all the pixels of the crack, and forming an ordered track point sequence; traversing the track point sequence in sequence, and dividing the track point sequence between the current two track points when the Euclidean distance between the two adjacent track points is larger than the preset distance, so as to divide the track point sequence into a plurality of continuous track point subsequences; fitting each segmented track point subsequence by adopting a B spline curve or a polynomial curve to generate a curve; and drawing all the curves generated by fitting onto a new blank mask map to obtain an optimized crack mask map.
2. The method for post-processing optimization and quantification of a fracture splitting image according to claim 1, wherein the calculating of geometric parameters of the fracture is performed, and the calculating of the fracture length specifically includes: Skeletonizing the optimized crack mask map to obtain a single-pixel crack skeleton; and counting the number of all pixel points in the single-pixel crack skeleton, thereby obtaining the crack length.
3. The method for post-processing optimization and quantification of a fracture-segmented image according to claim 1, wherein the calculating of geometric parameters of the fracture is performed, and the calculating of the fracture width specifically includes: Skeletonizing the optimized crack mask map to obtain a single-pixel crack skeleton; performing sub-pixel level edge detection on the optimized crack mask map to obtain a crack boundary; for any skeleton point of the single-pixel crack skeleton, detecting rays are taken to two sides of a tangent line of the current skeleton point along the normal direction of the current skeleton point, so that an intersection point of the rays and crack boundaries on two sides is obtained, and the distance between the two intersection points is the width of the crack corresponding to the current skeleton point; and calculating the average width and the maximum width of the crack based on the crack widths corresponding to all the skeleton points.
4. A post-processing optimizing and quantifying apparatus for a slit segmentation image, characterized in that the post-processing optimizing and quantifying apparatus for a slit segmentation image comprises: the filtering module is used for acquiring an initial binary crack mask image corresponding to the image to be processed, and performing geometric form filtering operation to remove the non-crack region so as to obtain a mask image after preliminary filtering; The detection module is used for carrying out contour detection on the preliminarily filtered mask map, and identifying and cutting the mask map to obtain an interested region mask map containing cracks; The execution module is used for conducting breakpoint connection processing on the region-of-interest mask map to repair discontinuous crack fragments, and generating a continuous optimized crack mask map so as to conduct calculation of geometric size parameters of cracks; the geometric form filtering operation is performed to remove the non-crack area, and specifically comprises the following steps: Analyzing the initial binarization crack mask map, extracting a crack contour to obtain crack pixel connected domains, and obtaining the minimum circumscribed rectangle of each crack pixel connected domain; Calculating to obtain the area of each crack pixel connected domain and the aspect ratio of the minimum circumscribed rectangle of each crack pixel connected domain; judging whether the area of the current crack pixel connected domain is smaller than a preset area threshold value: If yes, eliminating the current crack pixel connected domain; If not, judging whether the aspect ratio of the minimum circumscribed rectangle of the current crack pixel connected domain is smaller than a preset aspect ratio threshold, if so, eliminating the current crack pixel connected domain, and if not, reserving the current crack pixel connected domain; performing breakpoint connection processing on the region mask map of interest to repair discontinuous crack fragments, wherein the breakpoint connection processing is performed by adopting a curve fitting connection mode based on track points; the curve fitting connection mode based on the track points specifically comprises the following steps: scanning the mask map of the region of interest row by row or column by column along the main direction of the crack, extracting the centers or centroids of all the pixels of the crack, and forming an ordered track point sequence; traversing the track point sequence in sequence, and dividing the track point sequence between the current two track points when the Euclidean distance between the two adjacent track points is larger than the preset distance, so as to divide the track point sequence into a plurality of continuous track point subsequences; fitting each segmented track point subsequence by adopting a B spline curve or a polynomial curve to generate a curve; and drawing all the curves generated by fitting onto a new blank mask map to obtain an optimized crack mask map.
5. A post-processing optimizing and quantizing device of a slit segmentation image, characterized in that the post-processing optimizing and quantizing device of a slit segmentation image comprises a processor, a memory, and a post-processing optimizing and quantizing program of a slit segmentation image stored on the memory and executable by the processor, wherein the post-processing optimizing and quantizing program of a slit segmentation image, when executed by the processor, implements the steps of the post-processing optimizing and quantizing method of a slit segmentation image according to any one of claims 1 to 3.

Description

Post-processing optimization and quantification method, device and equipment for fracture segmentation image Technical Field The application relates to the technical field of computer vision and digital image processing, in particular to a method, a device and equipment for optimizing and quantifying the post-processing of a crack segmentation image. Background Cracks on the surface of a structure are key indicators for assessing its health and safety. In recent years, with the development of deep learning technology, semantic segmentation is performed on images by using convolutional neural networks such as U-Net, segNet or fusion Transformer architecture to automatically identify and extract crack regions, which has become a mainstream research direction. However, the initial segmentation result simply relying on the deep learning model output often has the following problems: (1) Incomplete and intermittent, namely, the cracks identified by the model are usually broken and discontinuous fragments due to uneven illumination, surface stains, texture interference or subtle cracks, and the like, so that the real trend and the complete length of the cracks cannot be reflected; (2) Misrecognition and noise, namely, background elements similar to crack forms, such as a plate joint, scratches, water stain edges and the like, often exist in an original image, and a model can misjudge the elements as cracks to form artifacts or noise, so that a higher false detection rate is shown; (3) The quantification difficulty is that the accuracy is low when key geometric parameters such as the maximum width, the average width, the total length and the like of the cracks are directly calculated according to the initial mask diagram, and the precision requirement of grading the cracks in engineering cannot be met. Meanwhile, the existing post-processing method is mostly focused on simple morphological operation or small-area noise filtering, so that the problems of intermittent connection of cracks and complex artifact removal are difficult to systematically solve, and a set of complete automatic flow from optimization to accurate quantification cannot be formed. Therefore, there is a need for a post-processing technique that can effectively repair intermittent cracks, remove unstructured artifacts, and accurately calculate crack size parameters. Disclosure of Invention The application provides a post-processing optimization and quantization method, device and equipment for a fracture segmentation image, which can remarkably improve the accuracy and the integrity of a fracture segmentation result and realize the automation and the accurate quantization of key parameters. In a first aspect, an embodiment of the present application provides a method for optimizing and quantifying a post-process of a fracture-segmented image, the method for optimizing and quantifying a post-process of a fracture-segmented image including: obtaining an initial binary crack mask image corresponding to an image to be processed, and performing geometric form filtering operation to remove a non-crack region to obtain a mask image after preliminary filtering; Performing contour detection on the mask map after preliminary filtering, and identifying and cutting to obtain a mask map of an interested region containing cracks; And performing breakpoint connection processing on the region-of-interest mask map to repair discontinuous crack fragments, and generating a continuous optimized crack mask map to calculate geometric size parameters of the cracks. With reference to the first aspect, in an embodiment, the performing a geometric filtering operation to reject a non-fractured region specifically includes: Analyzing the initial binarization crack mask map, extracting a crack contour to obtain crack pixel connected domains, and obtaining the minimum circumscribed rectangle of each crack pixel connected domain; Calculating to obtain the area of each crack pixel connected domain and the aspect ratio of the minimum circumscribed rectangle of each crack pixel connected domain; judging whether the area of the current crack pixel connected domain is smaller than a preset area threshold value: If yes, eliminating the current crack pixel connected domain; If not, judging whether the aspect ratio of the minimum circumscribed rectangle of the current crack pixel connected domain is smaller than a preset aspect ratio threshold, if so, eliminating the current crack pixel connected domain, and if not, reserving the current crack pixel connected domain. With reference to the first aspect, in an implementation manner, the breakpoint connection processing is performed on the region of interest mask map to repair discontinuous crack segments, where the breakpoint connection processing is performed by adopting a short-distance connection manner based on end points or a curve fitting connection manner based on track points. With reference to the first aspect, in one implemen