This article in view of the different scale changes lead to the detection rate of weld defect effect is not ideal,is proposed based on a faster region-based convolutional neural network (Faster R-CNN) of weld defect detection algorithm of improved algorithm using convolution expansion characteristics under the different expansion rate,combination of convolution kernels under different receptive field more comprehensive to extract the feature information of different scales,to improve the target detection accuracy at the same time,the deep separable convolution is used to compress the model to improve the detection speed.The experiment shows that the improved network can improve the detection speed while ensuring the operation speed and the detection accuracy can reach 72%.