Abstract:To solve the problems faced by traditional neural networks in copper mine image classification due to sensitivity field limitation and dimensional information blocking, an improved Swin-Transformer model based on X-ray transmission imaging technology is proposed. The model is based on Swin-Transformer. Mixing blocks are added in the second and third stages of the backbone network, and the bidirectional interaction between local window self-attention and deep convolution can significantly increase the model"s sensitivity field, thus significantly enhancing the feature representation and modeling capabilities. At the same time, the EMA module introduced reshaped some channels into batch dimensions and grouped the channel dimensions into multiple sub-features, so that the spatial semantic features were evenly distributed in each feature group, which further improved the model"s ability in the information fusion of channels and multi-scale spatial dimensions, and enhanced the characterization effect of features in regions of interest.Through experimental verification, the improved model has solved the problems of limited receptive field and dimensional information, and achieved an accuracy of 94.40% in the intelligent recognition task of copper mines.