Foreign Object in Coal Belt Conveyor Detection and Tracking Based on Improved YOLOv9 and Bytetrack

Conveyor belt transporting coal is an important scene in coal mining and transport operations. Once foreign objects are mixed in the conveyor belt, it will lead to damage of the conveyor belt and cause safety accidents in serious cases. However, the current mainstream algorithms have problems such as poor real-time performance, low recognition rate, frequent ID switching, and poor occlusion resistance, etc. To address the above problems, this paper proposes a tracking method based on the improved YOLOv9 algorithm and Bytetrack algorithm. The CEDNet module for acquiring multi-scale features is added, and the information representation of the features is enhanced by changing the loss of the main branch during the training process, which accelerates the convergence speed of the model. For long sequential information in the foreign object tracking process, the Vim module was introduced in the data association section, which uses global context modelling and positional embedding to better track and predict missing trajectory segments. ID switching due to occlusions was reduced. On the Belt Coal Foreign Dody Detection and Tracking dataset, MOTA and IDF1 achieve recognition rates of 91.2% and 92.5%, respectively, both of which are improved by 2.9%. The algorithm proposed in this paper not only identifies foreign objects in the conveyor belt coal transport scenario, but also tracks foreign objects accurately and stably in the dense occlusion scenario. The algorithm can also be applied to vehicle and pedestrian detection and tracking.