Research on a YOLOv8-Based Metal Residue Detection Algorithm

Abstract

To address the recognition challenges caused by the tiny size of metal residues, high reflectivity, and severe background texture interference in industrial production environments, this paper proposes a YOLOv8-based metal residue recognition algorithm. First, to overcome the scarcity of high-quality samples in industrial sites, a metal residue dataset containing multiple complex backgrounds such as welds and oil stains is constructed by simulating actual working conditions. Targeted data augmentation techniques are employed to enhance the model's generalization ability in extreme environments. Second, to tackle the problems of extremely low pixel occupancy of metal targets and the difficulty of feature extraction, a lightweight coordinate attention mechanism is introduced into the backbone network. By capturing cross-channel directional and positional information, the model's perception and extraction capabilities of subtle metal edge features are strengthened. Experimental results show that the improved algorithm achieves a mean average precision (mAP) of 0.724 on the self-built dataset. While maintaining real-time inference speed, it can accurately identify metallic foreign objects with extremely small diameters, effectively reducing the miss rate and false alarm rate under complex working conditions, thus providing reliable technical support for the deployment of industrial precision inspection and automated rejection systems.