Enhanced YOLOv11 for button cell battery defect detection: Leveraging local channel semantic guidance and multi-scale interaction

Button batteries, which serve as essential power sources in consumer electronics, are susceptible to various manufacturing defects that pose potential safety risks. This study proposes LMG-YOLO, an enhanced algorithm based on YOLOv11, specifically designed for button cell battery defect detection. The proposed method introduces a Local Channel Semantic Guidance (LCSG) module to suppress reflection-induced background interference and enhance defect-relevant semantic features. To address large scale variations of battery defects, a Multi-Scale Interaction (MSI) mechanism is designed to explicitly align and fuse adjacent-scale features, improving robustness to both small and large defects. In addition, a Grouped Iterative Attention (GIA) mechanism is incorporated to group and refine feature representations and strengthen discrimination of visually similar and hard-to-detect defects. Evaluated on a self-constructed dataset, LMG-YOLO achieves an mAP5095 of 65.4% and an mAP50 of 88.2%, surpassing state-of-the-art detection algorithms. The proposed method offers a promising solution for high-performance, automated defect detection in button batteries, addressing critical challenges in quality control and safety assessment. The code and model weights are available at https://github.com/LMGYOLO/LMG-YOLO.© 2017 Elsevier Inc. All rights reserved.