Lightweight coal mine conveyor belt foreign object detection based on improved Yolov8n.

To resolve the drawbacks of slow speed, excessive parameters, and high computational demands associated with deep learning-based conveyor belt foreign object detection methods, a lightweight algorithm for detecting foreign objects on conveyors based on an improved Yolov8n model is proposed. Firstly, a lightweight StarNet is employed as the backbone network to enhance the speed of target detection and reduce the complexity of the model. Secondly, a C2f.-EIEM module is proposed and embedded into the Backbone section to enhance the feature learning capability for image recognition tasks. Thirdly, to enhance the algorithm's focus on key features, a Large Separable Kernel Attention mechanism (LSKA) is utilized to improve the original SPPF, thereby boosting the overall performance of the algorithm. Fourthly, the original channel attention mechanism in the Head part is replaced with C2f_MLCA, which not only speeds up the processing speed but also successfully avoids the problems of accuracy degradation caused by channel dimensionality reduction. Fifthly, the lightweight detection head Detect-LSDECD is added, which uses Detail Enhancement Convolution (DEConv) for improvement and optimization, enhancing the stability of the algorithm's recognition under various environmental factors. Lastly, the original CIoU loss function in Yolov8n is replaced with MPDIoU, which allows the model to more accurately predict the position and shape of bounding boxes in the object detection task, thereby further reducing accuracy loss. The experimental results show that compared to the original model, the improved model has reduced the number of parameters by 80%, decreased the computational load by 60.49%, shrinked the model storage size by 69.35%, increased the accuracy by 1.9%, and maintained the recall rate, which is conducive to promoting the lightweight process of real-time foreign object detection on coal conveyor belts in coal mines.