A centernet-based direct detection method for mining conveyer belt damage

As the main components of belt conveyor, the conveyor belt plays an important role in carrying materials and transferring power. Due to the harsh usage conditions, conveyor belts often suffer from surface wear, surface damage, breakdown, tears or other damage accidents, especially in mining belt conveyors. In view of the immature damage detection methods of conveyor belt and the needs of the intelligent development of coal mine equipment, a centernet-based direct damage detection method was proposed on the basis of constructing the belt damage dataset. Unlike the current commonly used anchor-based target detection mechanism, centernet target detection algorithm was adopted in this paper based on center point detection, belongs to anchor-free mechanism, which omits the process of anchor generation and regression adjustment. The prediction process is more straightforward, and the test accuracy on the conveyor belt damage dataset reaches 97%, with a test speed of 32.4 FPS. Compared with Yolov3 algorithm, the accuracy is increased by 10%, and the detection speed by 12.9%, while 7.8% in accuracy with Yolov4. Meanwhile, the performance of various target detection algorithms, including hardware usage, were also compared on the conveyor belt damage dataset by means of transfer learning, which provides an empirical reference for the intelligent development of belt conveyors and the marginalization of monitoring.