Defect recognition of small pipe welds based on edge enhancement super-resolution method

Aiming at the problems including low image resolution, complex weld background and difficult identification of tiny defects existing in Digital Radiography (DR) inspection for small-diameter tubes of pressure pipelines, this paper proposes a weld defect detection algorithm based on improved YOLOv11n. Firstly, a Frequency-Aware Attention Network (FAAN), a super-resolution algorithm applicable to grayscale images, is designed to improve the overall definition of images while effectively preserving edge details of defect regions. Secondly, an improved Adaptive Weber-Contrast Convolution (AWC-Conv) module is introduced into the backbone network to strengthen the feature extraction capability of the model for defects with diverse sizes. Finally, the Pixel-focused Intersection over Union (PIoU) loss function optimized for tiny defects is adopted to boost the localization accuracy of small defects under complex backgrounds via pixel-level focusing. Experimental results demonstrate that the mean Average Precision at IoU threshold 0.5 ( P_\textmAP@0.5 ) of the proposed algorithm reaches 94.2% on the self-built DR image dataset of small-diameter tube welds, which is 5.3% higher than that of the original YOLOv11n. The proposed method remarkably improves the recognition accuracy and localization reliability of weld defects on small-diameter tubes, and can satisfy the requirements of high-precision automatic inspection in practical engineering applications.