Super-resolution reconstruction method of mine image based on multi-path adaptive information enhancement
QI Ailing
FU Yuanyuan
ZHANG Guangming
Abstract：The complex underground coal mine environment suffers from poor illumination,high humidity,and suspended dust-condi-tions that easily form water mist and glare.These factors lead to the loss of high-frequency information and blurring of edge details in cap-tured images,while also superimposing noise interference.To improve mine image quality and address the challenge of synergistically suppressing noise and restoring details in mine scene super-resolution reconstruction,a mine image super-resolution reconstruction meth-od based on multi-path adaptive information enhancement is proposed.Methodologically,a Residual Multi-path Feature Aggregation Block(RMFAB)is designed first,leveraging residual learning and a Multi-path Adaptive Convolution Network(MACN)to fully utilize features from different paths,significantly enhancing the modeling capability for both global and local high-frequency information.Second,a Multi-attention Fusion Module is introduced to focus on high-frequency information across channel and spatial dimensions,im-proving feature representation.Finally,a Large Kernel Perception Block(LKPA)is constructed,employing multi-scale convolution to ex-pand the receptive field and fuse hierarchical features,optimizing texture and structural details.Experimental results on the public CMUID mine dataset demonstrate that the proposed method outperforms existing state-of-the-art algorithms in both Peak Signal-to-Noise Ratio(PSNR)and Structural Similarity(SSIM).Particularly at a scaling factor of 4,the algorithm achieves PSNR improvements of 2.88,2.04,1.94,1.52,0.53,0.36 dB over Bicubic,CRAFT-SR,PAN,ESRGCNN,DiVANet,and SMAFNet,respectively.Corresponding SSIM im-provements are 4.32%,3.37%,3.20%,2.74%,3.19%,1.08%.The method achieves refined extraction and fusion of multi-level features in mine images,effectively suppressing noise interference while restoring complex texture features.This enhances the super-resolution recon-struction quality of mine images,thus contributing to intelligent perception in coal mine environments.
Keywords：mining imagessuper-resolution reconstructionmulti-path adaptive convolutionlarge kernel perception blockparallel multi-attention mechanism
Publication Date：2025-11-30
Online Publishing Date：2025-12-15(First online date of this platform, not the publication date of the document)
Pages：13( 172-184 )