基于DEEW-YOLO的焊缝X射线图像缺陷检测方法OA

During welding processes,various defects may occur in pressure vessels due to production techniques and environmental factors,which can compromise the mechanical properties and service life of welded structures.By employing deep learning methods for automated X-ray defect detection,convolutional neural networks(CNNs)can adaptively extract defect information,enabling efficient identification of different weld defect types while reducing human-induced subjective biases.However,most current CNN models still face challenges such as false positives,missed detections,and low recognition rates for minor defects.To address these issues,the DEEW-YOLO algorithm—an enhanced version of YOLOv8n for weld defect detection is proposed.The DEConv_C2f module integrates Detail-Enhanced Convolution(DEConv)with a two-convolution faster implementation of CSP bottleneck(C2f)to improve local detail capture and feature representation.Additionally,the ECA attention mechanism is incorporated into critical network layers to enhance cross-channel information capture.The WIoU loss function ensures balanced contributions between high-quality and low-quality anchor boxes while improving boundary prediction accuracy.The DEEW-YOLO model achieved mAP@0.5,Precision,Recall and FPS parameters of 59.7%,69.0%,58.9%and 144.9 f/s respectively,significantly outperforming the original YOLOv8n model.