Application Research Of Sparse Solving In X-ray Weld Image Defect Detection

With the continuous development of the petroleum industry,various problems may occur in the welding process,which will lead to defects in the weld,which will affect the quality and safety of the product.Therefore,it is particularly important to detect defects in the welded image.Due to the in-depth study of image processing technology and pattern recognition,computer intelligence detection is widely used in the detection of weld defects in oil and gas pipelines due to its high efficiency and objectivity.Aiming at the special requirements of X-ray weld image defect detection,this paper proposes a X-ray weld defect detection method based on sparse description.Different from the traditional defect detection algorithm,the proposed algorithm does not need to segment the defect and obtain the defect geometry.The texture feature value avoids the error caused by the eigenvalue finding,and directly determines whether the suspected region is a defect through the SDR image,which greatly improves the accuracy of the recognition.Since the defect detection based on the sparse description relies on a set of coefficients to determine the image to be detected,the interference of any image does not affect the overall recognition result,which greatly improves the robustness of the recognition algorithm.Based on the analysis of X-ray weld image noise and contrast,the paper gives a Gaussian filter and histogram equalization enhancement algorithm suitable for X-ray weld image denoising.The noise reduction and enhancement algorithms are verified by experiments.Experiments show that the peak signal-to-noise ratio of the image can be increased from 26.7675 to 41.9197 after Gaussian noise reduction and histogram equalization.In order to accurately identify the defects in the X-ray weld image,the paper proposes to extract the Region of Interest(ROI),which is to further process the processed image by Otsu segmentation and Sobel edge detection,and find it in the whole image.The weld boundary is extracted and the relevant parameters of the weld boundary line are extracted by Hough transform to realize the segmentation of the ROI region.Subsequently,the Otsu algorithm and the gray density clustering algorithm are used to segment the defects and image noise in the weld region.The effect,and then choose the latter to segment the Suspected Defect Region(SDR),and finally use the sparse description to identify the defect detection algorithm.According to the characteristics of RIP,this paper firstly selects the optimal dictionary matrix size based on the principle of least correlation,and constructs an X-ray weld image defect model based on sparse description.It proposes to use sensitivity and specificity to evaluate the model.A combination of coefficients obtained to determine the type of defect.In order to solve the sparse coefficient accurately,this paper further studies the L0 norm,L1 norm,L2 norm minimization algorithm.It has been verified by experiments that the accuracy of solving with L0 is the highest,and the sensitivity is 100% and the specificity is 89%.In order to further verify the accuracy of the experiment,the weld image is normalized to different sizes.In the process of solving,different penalty coefficient and regularization parameters are selected for comparison,and the comprehensive running time and accuracy are verified.When the normalization is 20*20,the accuracy of solving with the L0 norm is still the highest,so it can be considered that solving the L0 norm is the optimal algorithm for defect identification.