AEB-YOLO: An efficient multi-scale defect detection algorithm for copper strips

To address the challenges of low efficiency in detecting surface defects on copper strips in industrial settings, inadequate recognition of small targets, and excessive model parameters, this paper proposes a lightweight algorithm, AEB-YOLOv8n. First, an efficient multi-scale bottleneck module (C2f-EB) is designed to more effectively integrate global and local semantic information via a dual-branch mixed pooling architecture, enhancing multi-scale defect feature extraction in copper strips, significantly improving detection accuracy, and reducing model parameters. Second, an interactive collaborative attention module is proposed to enhance the feature representation of small-target defects by focusing on their spatial locations, suppressing complex background interference, and substantially improving the ability to capture micro-defects. Finally, the bidirectional feature pyramid network structure is incorporated into the neck network to construct a pruned bidirectional feature pyramid with learnable fusion weights, which reduces redundant computations and enhances feature fusion efficiency. Experimental results demonstrate that the improved AEB-YOLOv8n model achieves a 4.6% improvement in mAP@0.5 compared with the original YOLOv8 model while reducing the parameter count by 37.8% and lowering GFLOPs from 8.1 to 7.2, thus meeting the requirements for high accuracy and efficiency in copper strip defect detection.