Neutrosophic Enhancement of YOLO-MD Algorithm for Automated Metal Surface Micro Defect Detection

Abstract

To achieve automation of defect detection, the metal surface micro defect detection algorithm YOLO-MD is proposed. From the perspective of object detection, YOLOv5s is selected as the backbone algorithm and the SPD-Conv module is added to reduce feature loss caused by ordinary convolutional downsampling, improve the adaptability of low-resolution images, and improve the accuracy of small object detection. Using the MPDIoU loss function to accelerate model convergence and improve detection accuracy. Considering the small size of the dataset, data augmentation methods were adopted. After model training, mAP50-95 improved by 0.02 compared to YOLOv5, which has high real-time and robustness and can more effectively detect metal surface micro defects.