Convolutional Neural Network Based Defect Detection in Small Diameter Pipe Welds

Aiming at the current situation of low accuracy of X-ray weld image detection rate of small diameter pipe, the defect recognition algorithm of small diameter pipe weld image based on image segmentation is proposed by analysing the image and combining with convolutional neural network. Firstly, the ROI region is obtained by two-step image segmentation, then the SDR image is obtained by grey density clustering, and finally the SDR image is classified and identified on the basis of analysing and determining the structure of convolutional neural network. The experimental results show that the accuracy of defect detection reaches 95%, the false alarm rate is only 5%, and the recognition speed is fast, which achieves the effective recognition of small diameter pipe X-ray weld images.