Research On Feature Extraction Of Weld Defect Image Based On X-Ray

In modern steel pipe weld defect detection, a common method is artificial marking. Artificial subjective has so many disadvantage, so it greatly influences the efficiency and accuracy of the detection. The weld defect image automatic recognition system can overcome the disadvantage. The research of domestic and foreign researchers in the defect extraction and recognition has made great achievements, but the accurate detection of small defects in nonuniform background accurate detection, many problems still exist in the effective features classification and automatic recognition.In this paper, the weld image based on X ray is treated as the object, feature extraction and automatic recognition of defects was studied. Firstly, we use average filtering and median filtering pre-processing on the weld image. And we study and comparison two kinds of image enhancement algorithm.We choose the histogram equalization method for image enhancement. Then we study some image segmentation algorithm and use the iterative threshold on weld region segmentation. And the feature extraction and feature selection of weld defect has been studied. Finally, we use the S VM classifier which based on the binary tree for weld defect classification.The issues will be discussed in following manner:1. Introducing fundamental theories related to current image denosing technology and image enhancement. And we choose median filter and mean filter to treatment the weld defect image.2. Introducing fundamental theories related to current image segmentation technology and mainly focusing on iteration threshold segmentation.3. The geometric feature parameters and the texture feature parameters are chosen to distinguish the image. A valid function based inter-class variance and correlation is used to choose the feature parameters. The results can satisfy us.4. Contour tracking method is used to extract the geometric feature, the improved LTP(CLTP) is used to extract the texture feature.5. Researching on common SVM classification method, the classifier which is based on binary tree SVM is used to implement the function.