Abstract:In view of the lack of computing power of some mine ore sorting equipment and the lack of suitable ore classification model, a ore classification and recognition model based on ShuffleNetV2 was proposed by taking the ore image of X-ray transmission imaging as the data set. Firstly, by improving the basic unit of ShuffleNetV2, the depth convolution of 3x3 is expanded into a depth convolution of 7x7 to improve the receptive field. Secondly, by introducing the convolutional block attention module, more distinguishing features are extracted, while irrelevant features are suppressed. h-switch activation function was used to reduce the occurrence of silent neurons, which avoided neuronal necrosis caused by ReLU when the input was negative The hierarchical structure of the network is adjusted to reduce the number of unit repeats to make the model more lightweight. Finally, knowledge distillation is used to improve the performance of the model. The experimental results show that the identification accuracy can reach 98.67%, which is 5.42% higher than the original model, and the number of parameters is reduced by 0.18M, and the identification speed is only 7ms. Compared with SqueezeNet, Xception, MobiletNetV2, Vgg16, DenseNet121 and the original model, the accuracy, ore identification speed and model size have been improved, and can be deployed in small and medium-sized mine ore sorting equipment with insufficient computing power