Super-resolution of digital elevation models using deep learning methods based on detrending

The digital elevation model (DEM) is an important basic data tool applied in geoscience applications. Because of its high cost and long development cycle of enhancing hardware performance, designing the related models and algorithms to improve the resolution of DEM is of considerable significance. At present, Neural networks (NNs) have demonstrated the potential to recover finer textural details from lower-resolution images by super-resolution (SR). Given similar grid-based data structures, some researchers have transferred image SR methods to DEM. These efforts have yielded better results than traditional spatial interpolation methods. However, the deep learning(DL) models need a lot of training data, and the model is difficult to converge, resulting in high training costs, which can be challenging. Therefore, in order to reduce the difficulty and cost of DL method training, we detrend the DEM data to decompose the target DEM into a deterministic low frequency trend part and a high frequency residual part. In the process of DL training, focus on the high-frequency part. We use multiple DL models and DEM data of various landforms to verify, and the experimental results show that our proposed method can indeed reduce the difficulty and cost of DL training. At the same time, our method can also be extended to other DL models.