As intelligent manufacturing advances toward precision and automation, cutting tool condition critically impacts product quality, equipment safety, and production efficiency. Anomalies like wear, chipping, or fracture cause workpiece scrapping and machine failure, demanding efficient online monitoring. Traditional manual or image-based methods suffer from low accuracy in complex environments. Although deep learning excels in industrial defect detection, existing end-to-end detectors exhibit insufficient recall and localization precision for millimeter-scale cracks and blurred tool boundaries. To address these challenges, we propose MGC-YOLOv8, an enhanced framework built upon the YOLOv8 backbone. A Multi-Scale Edge-Dual Fusion (MSEDF) module is introduced to integrate feature maps across different scales, thereby strengthening the detection of minor defects. Furthermore, a Global-to-Local Spatial Aggregation (GLSA) module enriches feature representations by simultaneously capturing global context and local details. A Convolutional Block Attention (CBAM) module is embedded upstream of the prediction head to adaptively highlight critical features in both channel and spatial dimensions. Although the integration of MSEDF, GLSA, and CBAM introduces a marginal runtime overhead and a slight increase in parameter count, the optimized architecture preserves real-time inference speeds that fully satisfy the requirements of industrial inspection systems. Experimental results demonstrate that MGC-YOLOv8 substantially outperforms the baseline YOLOv8n, achieving 88.1% precision, 87.9% recall, 92.5% mAP@0.5 and 69.6% mAP@0.5:0.95 on our test set.
Guan et al. (Thu,) studied this question.