Abstract: With the rapid development of remote sensing technology,obtaining multi-source remote sensing image of min ing areas has become an important means for resource monitoring and assessment.However,remote sensing image from different sources have significant differences in spatial resolution,spectral characteristics,and acquisition time.How to effectively in tegrate these data to improve the accuracy of feature extraction in mining areas is currently a research difficulty.To this end,a multi-source remote sensing image fusion method based on multi-scale feature semantic association is proposed.This method first extracts multi-scale spatial-spectral features from remote sensing images through multi-scale feature extraction technology, using image pyramids and convolutional neural networks.Then,an improved graph convolutional neural network (GCNN) is used to process these multi-scale features.By introducing a feature propagation mechanism and an adaptive weight adjustment strategy,the network's ability to learn local structures and global information is enhanced.Finally,semantic association analysis is used to fuse geological background knowledge with multi-scale physical features.The relationship between features is meas ured by cosine similarity and the weights are dynamically adjusted,achieving effective integration of physical features and se mantic information.Experimental results show that the proposed method achieves a peak signal-to-noise ratio (PSNR) of 35.2dB,reduces the spectral angle error (SAE) to 0.015,has a root mean square error (RMSE) of 0.012,and a structural similar ity index (SSIM) of 0.89.It effectively improves the spatial resolution of hyperspectral data while maintaining good spectral characteristics.This method provides a new idea for remote sensing monitoring of mining areas and has certain practical appli cation value.

Key words: multi-source remote sensing image,feature extraction,semantic association,image fusion,graph convolutional neural network