Study On Dimension Reduction And Classification Detection Algorithms Of X-ray Weld Defect Image

With the development of image processing technology and pattern recognition technology,computer intelligence reviews are widely used in the field of weld defects detection in oil and gas pipelines due to its advantages of high efficiency,objectivity and economy.Traditional defect detection methods is to first extract the region of interest and then segment the region of interest.For the segmented suspected partial image,whether or not it is a true defect is judged by various feature values.However,there are many forms of actual weld defects.It is not only the amount of calculation that determines the correctness of the segmentation by the feature values such as the shape of the segmentation,but also the accuracy is influenced by the image quality.The convolutional neural network is a new type of artificial neural network method that combines artificial neural network and deep learning technology.It has the characteristics of local sensing area,hierarchical structure,feature extraction and classification process combined with global training,and it avoids the complex feature extraction and data reconstruction process in the traditional classification algorithm has been widely used in the field of image recognition.This article adopts the weld defect detection platform to identify the welding defects,such as: circle defects and linear defects,which may exist in the welding seam,a process includes: image processing,feature description and classification recognize.First of all,through image filtering the results of the analysis type of weld image noise and image after enhancement experiment,select appropriate image processing algorithms to remove noise,then the Sin enhancement,OTSU segmentation and sobel method is used to find the welding seam boundary in the entire image,and then the Hough transform is adopted to calculate the related parameters of boundary line,in order to achieve the ROI region segmentation.And then,by comparing the result of Ostu algorithm and density clustering for weld defects and segmentation of image noise,and then choose the latter way to split the ROI region.Secondly,the method of extracting geometric characteristic parameters and the method of Laplacian eigenprojection are used to reduce the dimension.Finally,BP neural network and support vector machine are respectively used to model and weld defects identification based on thecharacteristic parameters.Under the premise of the high recognition rate of the defect recognition system based on SVM,the characteristic value can not be guaranteed accurately due to the fact that the weld defect image itself is difficult to be accurately segmented during the image preprocessing,Can not avoid the final recognition results have an impact.In view of this problem,this article further uses the identification system based on convolution neural network to realize the recognition of weld defects,effectively avoids the calculation error and achieves a higher recognition rate.