Belt Blockage Detection on Low-Visibility Environment Based on YOLOv8s and MobileVit

Conveyor belt systems are essential for materials and products transportation. Blockages at discharge points, caused by equipment aging or human error, are inevitable and can lead to material combustion or production disruptions. Furthermore, the harsh working environment places high demands on the real-time performance, accuracy, and detection capabilities of blockage detection systems. To address these challenges, this paper proposes a real-time blockage detection model for conveyor belt discharge points based on an improved version of YOLOv8s. The feature extraction backbone is improved by MobileViT to enable the fusion of local and global blockage features, enhancing the model's ability to capture multi-scale blockage features under low-visibility conditions. The Shuffle-Attention mechanism is employed to rearrange and optimize the output blockage features, improving their diversity and expressiveness. In addition, the model and its runtime environment are containerized for deployment on both server and edge devices. Experimental results show that the model achieves a detection accuracy of 97.9%, with recognition speeds of approximately 123.1 FPS on the server and 20.3 FPS on edge devices, fulfilling the requirements for real-time detection.