Foreign Object Detection of Belt Conveyor Based on Enhanced YOLOv8 Model

Object detection on conveyor belts through computer vision currently faces several challenges. To address issues such as low performance and the scarcity of training and testing data, this paper proposes an improved solution based on YOLOv8. This paper incorporates various attention mechanisms, modifies, network structure of YOLOv8, and optimizes association matching using the EIOU algorithm. As a result, it can accurately detect foreign objects on the conveyor belt, including their categories and appearance features. Furthermore, this paper constructs a dataset by onsite collection and annotation. After labeling and categorizing the data, the dataset is used for training. Experimental results indicate that the improved algorithm enhances performance metrics such as accuracy, mAP50, and mAP50-95 compared to the baseline algorithm, YOLOv8n, and these results are validated on the test dataset.