A Method for Weld Quality Assessment and Archiving Based on Machine Vision and Neural Networks

Abstract

This invention discloses a method for weld quality assessment and archiving based on machine vision and neural networks, comprising the following steps: 1) performing digital preprocessing on digitized X-ray films; 2) using algorithms to determine the weld location and text information location; 3) using neural networks to identify damage information and text information; 4) using text information to classify the digitized X-ray films, and simultaneously using damage information to archive weld damage. This invention can effectively assist inspection personnel in achieving automatic assessment and data archiving of welding quality.

Description

Technical Field

[0001] This invention relates to the field of nondestructive testing, specifically to a method for weld quality assessment and archiving based on machine vision and neural networks. Background Technology

[0002] Welding quality is crucial for the safe use of structural components, and X-ray inspection, due to its high penetrability into the interior of materials, can effectively detect welding defects. However, after obtaining X-ray images of weld seams, professional radiographers are needed to evaluate and categorize the damage. Because the high density of X-ray films improves image quality, it requires an extremely bright cold light source for transmission. This results in irreversible visual damage to the already heavy daily workload of radiographers. Furthermore, human eye fatigue can lead to missed or incorrect assessments of weld defects, thus creating safety hazards.

[0003] Therefore, in order to solve the above problems, it is indeed necessary to provide an innovative welding quality judgment method based on machine vision and neural networks, and to design and develop a defect recognition method for radiographic films based on image characteristics, so as to assist inspectors in realizing automatic judgment of welding quality and data archiving. Summary of the Invention

[0004] To overcome the shortcomings of the prior art, the present invention provides a weld quality judgment and archiving method based on machine vision and neural networks, which can realize defect identification of X-ray inspection images and automatic archiving of X-ray inspection images.

[0005] Technical solution

[0006] A method for weld quality assessment and archiving based on machine vision and neural networks includes the following steps:

[0007] Step 1: Perform digital preprocessing on the digitized X-ray film;

[0008] Step 2: Use algorithms to determine the location of the weld and the location of the text information;

[0009] Step 3: Use neural networks to identify damage information and text information;

[0010] Step 4: Use text information to classify the digitized X-ray films, and use damage information to archive the weld damage.

[0011] Furthermore, the digital preprocessing operations in step one include median filtering and mean filtering, image contrast enhancement, grayscale conversion, and binarization thresholding.

[0012] Furthermore, the kernel sizes for the median filter and the mean filter are 7 and 11, respectively; the image contrast enhancement uses the Sin enhancement method, as shown in the following formula:

[0013]

[0014] Where f(x,y) and h(x,y) represent the pixel grayscale values ​​before and after the transformation, respectively, and m and n are the maximum and minimum values ​​of the pixels in the image before the sin enhancement, respectively; the binarization threshold segmentation adopts the OTU binarization method.

[0015] Furthermore, the method for determining the weld location and text information location using an algorithm in step two is as follows:

[0016] 1. The Canny edge detection algorithm and the contour finding algorithm are used to calculate the location of weld edges and text information on the binarized threshold segmentation image, respectively;

[0017] 2. Use the Hough line detection algorithm to determine the weld location on the edge detection map.

[0018] Furthermore, the Canny edge detection algorithm specifically uses the Prewitt operator or the Sobel operator to calculate the gradient between image pixels, thereby achieving the effect of edge detection; the Hough line detection algorithm, based on the Canny edge detection, transforms the problem of line detection of pixels in image space into the problem of point detection in parameter space, and then completes line detection by finding peaks in parameter space.

[0019] Furthermore, in step three, the method of using a neural network to identify damage information and text information involves dividing the weld into small blocks of 32×32 pixels, then using a pre-trained classification neural network to classify the small blocks, and finally synthesizing the entire probability map of weld defects. The text information includes numbers and letters, which are selected by contour selection and input into another pre-trained number and letter recognition network to obtain the weld text information.

[0020] Furthermore, the classification and archiving in step four specifically involve classifying the digitized X-ray films using textual information, archiving the weld damage using damage information, and finally importing them into a specific folder.

[0021] Beneficial effects

[0022] Compared with the prior art, the present invention has the following advantages:

[0023] 1. A novel defect identification method based on machine vision and neural networks for weld quality judgment and archiving is proposed. Starting from small to large, the weld and text positions are determined first, and then defect and text recognition is performed. This method can effectively identify weld defect information and text information in X-ray films.

[0024] 2. By identifying the numerical and textual information in X-ray films, the digitized films are automatically classified and archived, reducing the workload of workers and improving the efficiency of film inspection. Attached Figure Description

[0025] Figure 1 This is a flowchart of a weld quality judgment and archiving method based on machine vision and neural networks according to the present invention;

[0026] Figure 2 It is the original digital X-ray film;

[0027] Figure 3 It is a diagram of a binarized threshold segmentation image and a contour finding algorithm;

[0028] Figure 4 This is a hough line detection graph. Detailed Implementation

[0029] To better illustrate the content of this invention, the following description is provided in conjunction with the accompanying drawings and examples:

[0030] have Figures 1-4 As shown, this invention discloses a method for weld quality assessment and archiving based on machine vision and neural networks, comprising the following steps:

[0031] Step 1: Perform digital preprocessing on the digitized X-ray film;

[0032] Step 2: Use algorithms to determine the location of the weld and the location of the text information;

[0033] Step 3: Use neural networks to identify damage information and text information;

[0034] Step 4: Use text information to classify the digitized X-ray films, and use damage information to archive the weld damage.

[0035] Furthermore, the digital preprocessing operations in step one include median filtering and mean filtering, image contrast enhancement, grayscale conversion, and binarization thresholding.

[0036] Furthermore, the kernel sizes for the median filter and the mean filter are 7 and 11, respectively; the image contrast enhancement uses the Sin enhancement method, as shown in the following formula:

[0037]

[0038] Where f(x,y) and h(x,y) represent the pixel grayscale values ​​before and after the transformation, respectively, and m and n are the maximum and minimum values ​​of the pixels in the image before the sin enhancement, respectively; the binarization threshold segmentation adopts the OTU binarization method.

[0039] Furthermore, the method for determining the weld location and text information location using an algorithm in step two is as follows:

[0040] 1. The Canny edge detection algorithm and the contour finding algorithm are used to calculate the location of weld edges and text information on the binarized threshold segmentation image, respectively;

[0041] 2. Use the Hough line detection algorithm to determine the weld location on the edge detection map.

[0042] Furthermore, the Canny edge detection algorithm specifically uses the Prewitt operator or the Sobel operator to calculate the gradient between image pixels, thereby achieving the effect of edge detection; the Hough line detection algorithm, based on the Canny edge detection, transforms the problem of line detection of pixels in image space into the problem of point detection in parameter space, and then completes line detection by finding peaks in parameter space.

[0043] Furthermore, in step three, the method of using a neural network to identify damage information and text information involves dividing the weld into small blocks of 32×32 pixels, then using a pre-trained classification neural network to classify the small blocks, and finally synthesizing the entire probability map of weld defects. The text information includes numbers and letters, which are selected by contour selection and input into another pre-trained number and letter recognition network to obtain the weld text information.

[0044] Furthermore, the classification and archiving in step four specifically involve classifying the digitized X-ray films using textual information, archiving the weld damage using damage information, and finally importing them into a specific folder.

[0045] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the technical solutions of the present invention have been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for weld quality assessment and archiving based on machine vision and neural networks, characterized in that, The steps include: Step 1: Perform digital preprocessing on the digitized X-ray film; Step 2: Use algorithms to determine the location of the weld and the location of the text information; Step 3: Use neural networks to identify damage information and text information; Step 4: Use text information to classify the digitized X-ray films, and use damage information to archive the weld damage. The digital preprocessing operations in step one include median filtering and mean filtering, image contrast enhancement, grayscale conversion, and binarization threshold segmentation. The kernel sizes for the median filter and the mean filter are 7 and 11, respectively; the image contrast enhancement uses the Sin enhancement method, as shown in the following formula: Where f(x,y) and h(x,y) represent the pixel grayscale values ​​before and after the transformation, respectively, and m and n are the maximum and minimum values ​​of the pixels in the image before the enhancement, respectively; the binarization threshold segmentation adopts the OTU binarization method; The method for determining the weld location and text information location using an algorithm in step two is as follows:

1. The Canny edge detection algorithm and the contour finding algorithm are used to calculate the location of weld edges and text information on the binarized threshold segmentation image, respectively; 2. Use the Hough line detection algorithm to determine the weld location on the edge detection map; The Canny edge detection algorithm specifically uses the Prewitt operator or the Sobel operator to calculate the gradient between image pixels, thereby achieving the effect of edge detection; the Hough line detection algorithm, based on the Canny edge detection, transforms the problem of line detection of pixels in image space into the problem of point detection in parameter space, and then completes line detection by finding peaks in parameter space. Step three involves using a neural network to identify damage and text information. This involves dividing the weld into 32×32 pixel blocks, classifying these blocks using a pre-trained classification neural network, and finally synthesizing the entire probability map of the weld defect. The text information includes numbers and letters, which are selected by contour selection and input into another pre-trained number and letter recognition network to obtain the weld text information. The classification and archiving in step four specifically involve classifying the digitized X-ray films using textual information, archiving the weld damage using damage information, and finally importing them into a specific folder.

Images