Research On Radiographic Image Inspection Of Welding Defects In Oil And Gas Pipelines

In machine manufacturing and petrochemical construction,welding operations are the key steps in this type of manufacturing and construction industry.The quality of completion and the efficiency of completion are directly related to the company's production level,product safety,and service life.Therefore,with automation and The continuous development of the computer field has put forward further requirements for the traditional detection of welding defects,and the application of existing digital image processing technologies and detection and identification technologies in traditional detection methods has formed a new class of topics.In this paper,the existing ray defect maps are converted into digital images,and the image processing technology is used to complete noise reduction filtering,feature detection,feature extraction and final defect classification.In the third chapter,we mainly deal with image enhancement and noise reduction,including power transformation and histogram specification to enhance the image,and Gaussian smoothing,median filtering,bilateral filtering,and morphological weight adaptive images.The denoising algorithm completes the denoising of the image.In the fourth chapter,we mainly focus on edge detection and region segmentation.The function of this part is to separate the target(defect)in the image from the background for later extraction and differentiation.In the edge detection,we pass the Sobel operator and Prewiit operator.The Roberts operator obtains the boundary of the defect part,and both the Laplacian Gaussian operator(Lo G)and the Gaussian difference operator(Do G)are the operator of the feature detection part,and the defect part is completed by calculating the response point in the image.The match paves for the next chapter feature detection match.Otsu method,iterative method,and maximum entropy threshold method were used to separate defects from background.In Chapter 5,the detection and matching algorithm based on the scale invariant feature(SIFT method)and the accelerated robust feature algorithm(SURF method)are used to detect and match the defect part and the defect map respectively.Gradient histogram algorithm(Ho G method)is used to calculate the directional gradient in the image,and the directional gradient can be trained as a training sample through corresponding processing.In the sixth chapter,we mainly use two different algorithms for classification,image recognition based on support vector machine and image recognition based on connected domain detection,and classification according to the corresponding threshold.Through the above several chapters,the preprocessing part starting from the input of the original image,namely the enhancement of the image and the noise reduction of the image,are respectively completed;and the edge detection and the area division of the initial captured image are performed to achieve the separation of the target and the background,so that the effective The defect part in the image is extracted;then the defect feature is detected and the matching of the defect part is completed;finally,the classifier is used to complete the classification and classification of several types of defects.