胶带输送机广范应用于煤炭行业,为了保证设备连续、稳定运行,必须对胶带输送机加装相应的故障检测装置,传统的检测装置检修维护工作量大,保护误动作率高。通过改进YOLOv5s的方法,对胶带输送机煤量进行检测,首先将原本YOLOv5s的骨干网络替换为MobileNetV2,然后在骨干网络之后引入注意力模块,在Neck中引入BiFPN结构,利用迁移学习的方法微调模型参数。结果表明,改进后的YOLOv5s检测速度为32帧/s,mAP50为98.9%,与常见的目标检测算法相比,具有更高的检测精度和速度,可以快速准确地实现输送机“无煤量”“小煤量”“中煤量”“大煤量”检测,进而对输送机进行实时调速,为智能视频监控系统设计提供参考。

The belt conveyor is widely used in the coal industry. In order to ensure the continuous and stable operation of theequipment, the belt conveyor must be equipped with a corresponding fault detection device. The traditional detection devicehas a large maintenance workload with a high protection malfunction rate. In this paper, the method of YOLOv5sis improvedto detect the coal quantity of the belt conveyor. First, the original backbone network of YOLOv5s is replaced by MobileNetV2, then the attention module is introduced after the backbone network, and the BiFPN structure is introduced into theneck. Finally, the model parameters are fine tuned by the method of migration learning. Through the analysis of the experimental results, it can be seen that the detection speed of the improved YOLOv5s is 32 FPS and the mAP50 is 98. 9% . Compared with the common target detection algorithms, it has higher detection accuracy and speed. The improved method canquickly and accurately realize the detection of “ no coal quantity” , “ small coal quantity” , “ medium coal quantity” , “ largecoal quantity” of the conveyor, thus providing reference for the real-time speed regulation of the conveyor and the design ofintelligent video monitoring system.