Abstract: In order to improve the accuracy and efficiency of mine image reconstruction,a mining image reconstruction algorithm based on multi-scale feature multiplexing residual network is proposed to solve the problem of low reconstruction quality caused by detail loss in mine image reconstruction. Firstly,a multi-scale feature extraction module is designed. By stacking multiple parallel convolution layers and pooling layers,the image feature extraction module is constructed in combination with the local residual network,and the multi-scale detail feature representation of the image is fully extracted from the input image through the multi-channel combination network of different scales. These features represent different semantic information and spatial resolution,which can capture different details and texture structures in the image. Then,a feature reuse module is introduced to fuse and reuse features of different scales to enhance the accuracy of image reconstruction. Through multi-scale feature interaction and information transfer,global and local context information can be effectively used to improve image reconstruction performance. Experiments on the self-built mine image reconstruction dataset show that the proposed algorithm has achieved significant improvement in reconstruction accuracy and efficiency. Compared with other deep learning models,the proposed algorithm has better performance in detail retention and structural accuracy of reconstructed images. In addition,the algorithm has a fast training and inference speed,which is suitable for practical application scenarios.

Key words: mine image reconstruction,multi-scale feature reuse,residual network,image quality