Foreign Object Recognition Method of Coal Mine Conveyor Belt Based on Lightweight Network

Foreign objects in coal mine conveyor belts are a crucial factor causing conveyor belt damage. Using intelligent methods to accurately recognize foreign objects in coal mine conveyor belts is beneficial for the safer and more stable operation of the conveyor. In this paper, a novel coal mine conveyor belt foreign object recognition method based on lightweight ShuffleNetv2 network, named ASLNet, is constructed. Firstly, the 3x3 depthwise convolution in the ShuffleNetv2 network basic module is replaced by a 5x5 size, which increases the range of the sensing field and is easy to capture the global features of the foreign object target. Second, a lightweight ECANet (Efficient Channel Attention Network) attention mechanism module is added to the ShuffleNetv2 network module to strengthen the coal mine conveyor belt foreign object features and suppress the background features. Finally, the network structure is optimized by reducing the number of stacks in each module to improve the speed of foreign object recognition and reduce the number of model parameters. The experimental results show that the ASLNet model has a parameter number of 0.33x10
⁶
, a model size of 1.26M, and a recognition accuracy of 97.62%, which is 3.91 percentage points higher than that before the improvement. Compared with the classical ResNet34, ResNet50, DenseNet, GhostNet, and MobileNetv2 recognition networks, the optimized network not only has higher recognition accuracy, but also has fewer parameters, which lays a solid technical foundation for the online real-time recognition of foreign objects in coal mine conveyor belts.