Gangue Detection Method of Coal Belt Conveyor Based on Shuffle Attention and Mobilenetv2

To address the challenge of accurately detecting large gangue on coal conveyor belts, an algorithm of coal gangue detection on coal conveyor belts based on improved shuffle attention (SA) and mobilenetv2 is proposed. Firstly, the improved SA mechanism is incorporated into the inverted residual module of MobileNetV2. For the improved SA, the channel attention output and spatial attention output of the SA module adopt the Hard-Sigmoid activation function. In addition, the feature extraction network of MobilenetV2 is improved, and the bottleneck of the last three layers is replaced by
$\mathbf{1}\times \mathbf{1}$
conventional convolution. This adjustment not only enhances the network's feature extraction capabilities but also reduces the model parameters. Experimental results show that the accuracy, recall and F1-score of the proposed method are 94.22%, 99.38% and 94.50% respectively, which are 0.94%, 1.88% and 0.95% higher than that of mobilenetv2. The results indicate that the accuracy of the algorithm on the belt has been significantly improved.