Research on Gangue Target Detection Algorithm Based on SSL-YOLOv8

To further enhance the sorting performance of coal gangue, a streamlined object detection algorithm for real-time applications, based on an enhanced YOLOv8 framework. is proposed. Using YOLOv8n as the base network architecture, we first introduce the lightweight StarNet as the backbone network, leveraging its structural characteristics to decrease model parameters. The StarBlocks structure from the StarNet network is employed to replace the Bottleneck structure in the C2f module of the Neck part, forming the SC2f module. This effectively reduces the parameter count of the YOLOv8 network while maintaining lightweight characteristics and improving the model's computational speed and detection performance. Finally, a lightweight shared convolutional detection head is introduced to make the model even more portable. To assess the efficacy of the enhanced algorithm, it is experimentally compared with YOLO series algorithms on a self-built dataset. The findings suggest that the proposed algorithm demonstrates the highest overall detection efficacy. While ensuring detection accuracy, its parameter count is reduced by 54.44% compared to YOLOv8n, the processing load is decreased from 8.1 GFLOPs of the original model to 4.6 GFLOPs, the model size is reduced from 6.2 MB to 2.8 MB, and the FPS is increased from 111.11 Hz of YOLOv8n to 185.18 Hz. The research demonstrates that the algorithm is characterized by high efficiency, real-time performance, and lightweight nature, achieving a balance between model detection performance and computational resource consumption. It has significant potential and advantages in the area of coal gangue detection.