This study aims to focus on the rapid and accurate localization and identification of welding defects using low-brightness, low-contrast and high-resolution X-ray films. A novel algorithm is proposed to address the limitations of manual inspection in the welding inspection of special equipment.
This study integrates an improved multi-scale RetinexNet for image enhancement with an enhanced YOLO11 incorporating a small-object detection head for defect detection.
The proposed method achieves a 92.1% mAP50, significantly outperforming the reference method’s 76% and proves particularly effective on slag inclusions, boosting detection by 28.4% due to its ability to handle rough, angular textures.
This study integrates an ensemble-based data augmentation strategy with an enhanced YOLO11 framework for defect detection, resulting in improved detection performance.
