HE-Yolov8n: an innovative and efficient method for detecting defects in Lithium battery shells based on Yolov8n

In the manufacturing process of lithium battery shells, defects often arise due to improper operations, highlighting the importance of effective detection. However, the diverse and complex characteristics of these defects challenge existing algorithms, leading to high missed detection rates, low accuracy, and computational inefficiency. To address these issues, this study introduces HE-Yolov8n, an optimized Yolov8n model. The AKASE module enhances feature representation and improves training accuracy, while the SCCPrecision-GAM module replaces the traditional C2f module, reducing parameters and FLOPs. Additionally, a novel loss function, QIoU, is proposed to refine bounding box accuracy by penalizing low IoU during early training and later optimizing aspect ratios and center points. Experimental results demonstrate that HE-Yolov8n significantly outperforms mainstream models in detecting surface defects. Specifically, in lithium battery shell defect detection, it achieves an mAP50 of 97.0%, representing a 4.6% improvement over Yolov8n. Its parameters and FLOPs are reduced by 18.75% and 8.05%, respectively, while maintaining a detection speed of 132.2 FPS, meeting the real-time requirements of industrial defect detection.