Surface Defect Detection of Aluminum Sheet Based on CS-YOLO Algorithm

An enhanced algorithmic model CS-YOLO, based on the YOLOv5 network, is proposed to address issues with leakage detection, low detection accuracy, and efficiency in traditional methods for identifying defects on aluminum flake surfaces. Firstly, to leverage input contextual information fully and enhance visual representation capability, a portion of the CSPDarknet53 (C3) module in the backbone network is optimized for YOLOv5s by integrating contextual information mining and self-attentive learning into a unified architecture, thus forming the CoT3 module. Secondly, to mitigate target miss-detection issues arising from the traditional non-maximal suppression (NMS) algorithm, the soft non-maximal suppression (Soft-NMS) algorithm is introduced to replace NMS in the network. Experimental results demonstrate that the enhanced CS-YOLO surface defect detection algorithm effectively reduces leakage detection rates, while improving average detection accuracy.