Abstract:In response to the issues of poor timeliness and reliability in traditional coal gangue separation methods, a multi-scale coal gangue recognition method based on ST-YOLOX-S is proposed. This method utilizes machine vision and image processing technology to efficiently identify coal gangue by extracting visible light image features through deep learning models. Building upon YOLOX-S, the ST-YOLOX-S model is established by introducing Swin-Transformer to enhance global feature extraction capability. Additionally, the use of multi-scale mixed expandable convolutional technology captures feature information of different scales, effectively addressing the insufficient global feature extraction, scale diversity, and complex morphology problems associated with coal gangue targets. As a result, the detection performance of the model is significantly improved. Experiments conducted on a self-built dataset of coal gangue under actual production conditions at a coal preparation plant demonstrate that the prediction accuracy of the ST-YOLOX-S model reaches 90.11% at a confidence level of 95%, representing an improvement of 7.30% compared to the original YOLOX-S model. When compared with other mainstream object detection algorithms such as YOLOV4, YOLOV5, RCNN, EfficientDet and CenterNet algorithms, the precision rate (0.89%), parameter count (7.80 MB), and recall rate (0.92%) of the ST-YOLOX-S algorithm are significantly superior. Further ablation experiments confirm the effectiveness of each component in the ST-YOLOX-S model; after adding multi-scale mixed expandable convolutional layers and replacing Swin-Transformer as backbone network components into YOLOX-S, the precision rate, recall rate, and FPS values have increased by 5.95%, 10.84%, 25.48%, respectively. Finally, testing using ST-YOLOX-S shows that after improving this model, there is less occurrence of overlapping boxes when detecting targets, and higher probability values detected which indicates its superior performance in coal gangue detection. This has important significance for promoting clean and efficient utilization of coal.