This research aims to efficiently create detailed bathymetric charts. Our approach is to obtain fine seafloor details from coarse depth measurements only, making full use of existing data and minimizing new observation. To this end, treating gridded bathymetric data as digital images, we propose to apply super resolution, which is a technique to enhance image resolution, to bathymetry. Specifically, we employ learning-based super resolution to automatically extract characteristic features of bathymetric images. In experiments, we prepared pairs of low and high-resolution images, and let a deep neural network learn their relationship and estimate a high-resolution image from each low-resolution one. Then, we evaluated results in terms of numerical error and visual quality, and confirmed that the proposed method can recover detailed seafloor structures more plausibly than naive interpolation.
