Multi-Scale Fusion Lightweight Target Detection Method for Coal and Gangue Based on EMBS-YOLOv8s

The accurate detection of coal gangue is an important prerequisite for the intelligent sorting of coal gangue. Aiming at existing coal gangue detection methods, which have problems such as low detection accuracy and complex model structure, a multi-scale fusion lightweight coal gangue target detection method based on the EMBS-YOLOv8s model is proposed. Firstly, the coal gangue images collected through the visual dark box platform are preprocessed using CLAHE to improve the contrast and clarity of the images. Secondly, the PAN-FAN structure is replaced by the EMBSFPN structure in the neck network. This structure can fully utilize the features of different scales, improve the model’s detection accuracy, and reduce its complexity. Finally, the CIoU loss function is replaced by the Wise-SIoU loss function at the prediction end. This improves the model’s convergence and stability and solves the problem of the imbalance of hard and easy samples in the dataset. The experimental results show that the mean average precision of the EMBS-YOLOv8s model on the self-constructed coal gangue dataset reaches 96.0%, which is 2.1% higher than that of the original YOLOv8s model. The Params, FLOPs, and Size of the model are also reduced by 29.59%, 12.68%, and 28.44%, respectively, relative to those of the original YOLOv8s model. Meanwhile, the detection speed of the EMBS-YOLOv8s model is 93.28 f.s−1, which has certain real-time detection performance. Compared with other YOLO series models, the EMBS-YOLOv8s model can effectively avoid the occurrence of false detection and missed detection phenomena in complex scenes such as low illumination, high noise, and motion blur.