Abstract: Mine remote sensing images play a crucial role in resource surveying and disaster warning. However,due to
complex imaging environments and scattering effects,they often suffer from severe noise and texture distortion,which compromises
recognition accuracy. To address this issue,a denoising algorithm is proposed for mine remote sensing images by integrating
the Nonsubsampled Contourlet Transform (NSCT) with the Hidden Markov Tree (HMT) and Fuzzy Local Information CMeans
(FLICM). Firstly,NSCT is employed to perform multi-scale decomposition. Then,the HMT is used to model cross-scale
dependencies between parent and child coefficients. Subsequently,FLICM clustering is combined with a Graph Convolutional
Network (GCN) to enhance the aggregation of both spatial neighborhood and cross-neighborhood features,thereby effectively
suppressing complex noise while preserving fine details. Experimental results on the UCM and AID public datasets show that
the proposed algorithm achieves a Peak Signal-to-Noise Ratio (PSNR) of 34. 59 dB and 33. 74 dB,a Structural Similarity Index
(SSIM) of 0. 917 and 0. 904,and a Gradient Magnitude Similarity Deviation (GMSD) of 0. 072 and 0. 082,respectively.
The per-frame processing latency is 35. 18 ms and 36. 29 ms. The performance outperforms that of Generative Adversarial Network
(GAN),Transformers,and hybrid self-supervised methods. The study demonstrates that the proposed algorithm possesses
high robustness and practical value in complex mine environments,facilitating mine remote sensing image denoising and hazard identification.

Key words: mine remote sensing image,image denoising,NSCT,HMT,FLICM