Research on leather defect detection algorithm based on GND-YOLOv8

In the leather manufacturing industry, implementing online leather defect detection can significantly reduce waste during the production process. This study addresses three main issues encountered in the team's online leather defect detection project: the difficulty in collecting leather defect data, slow detection speed, challenges in achieving real-time detection, and insufficient accuracy in detecting small target leather defects. To tackle these problems, a lightweight leather defect detection algorithm based on GDN (Ghost-DFC Attention-NWD-YOLOv8) is proposed, which improves upon YOLOv8. The research employs the Albumentations data augmentation technique to expand the dataset size and combines the Mosaic data augmentation method to further enhance sample diversity during training. By introducing the Ghost module to replace the C2f module in the Backbone and Neck, the model's parameter count and computational load are significantly reduced, thereby improving detection speed. Additionally, the DFC Attention mechanism is utilized to strengthen the model's ability to learn leather defect features, while the NWD metric is adopted to replace the traditional IOU measurement method, further enhancing the detection accuracy of small target defects. Experimental results demonstrate that the proposed algorithm, compared to YOLOv8, reduces computational load by 18.3% and parameter count by 16.4%, achieves an AP of 94.5% for small target defects, and attains an overall mAP of 94.1%.