Automatic Detection of Weld Defects in Pressure Vessel X-Ray Image Based on CNN

The visual automatic detection method based on artificial intelligence has attracted more and attention. In order to improve the performance of weld nondestructive defect detection, we propose DRepDet (Dilated RepPoints Detector). First, analyze dataset in detail summarize distribution characteristics data, that is, scale is very different aspect ratio range large. Second, according design DResBlock module, introduce dilated convolution with rates process feature extraction expand receptive field large-scale defects. Based anchor-free framework RepPoints, DRepDet. Extensive experiments show our proposed detector can detect 7 types When using combined rate network AP50 Recall50 big defects are improved by 3.1% 3.3% respectively, while small not affected, almost same or slightly improved. final whole a large margin, 6% 4.2% compared Cascade R-CNN 1.4% 2.9% RepPoints.