EF-TSR:edge feature transformer-based DEM super-resolution network

The technology of super-resolution reconstruction has been widely used in various fields, but there are still significant challenges in reconstructing Digital Elevation Model (DEM). To overcome the problems of losing details and distorting the complex terrain features of DEM, we have developed a new model called Edge Feature Transformer-based DEM Super-Resolution Network (EF-TSR). Our model enhances the ability to capture overall information and model long-range dependencies using a series of residual feature fusion blocks. Within these blocks, the multi-scale pyramid segmentation Transformer layer integrates multi-head self-attention with pyramid segmentation attention, effectively capturing both local and global DEM features. Additionally, to improve the realism of reconstructed DEM, we include a dual-filter convolution block that extracts high- and low-frequency features through two parallel filter structures. The super-resolution reconstruction process is guided by dual constraints in the gradient and height domains. Experimental results using elevation maps of Shaanxi Qinling at two different resolutions show that the EF-TSR model outperforms other advanced models across various metrics, producing DEM with richer details and clearer textures. In summary, the EF-TSR model proposed in this study not only enhances the quality of DEM reconstruction but also provides new perspectives and methodologies for further research in related fields.