Application of Deep Learning Methods to Improve the Resolution of Small Objects Images in Aerial Photographs

Abstract

Super-resolution video sequence techniques aim to create a high spatial resolution frame from multiple low resolution frames in some local time window. The inter-frame temporal relationship is just as important as the intra-frame spatial relationship for solving this problem. However, how to use temporal information effectively remains a challenge, as complex movements are difficult to model, which can lead to adverse effects if not properly processed. The paper presents the results of a study on testing and analyzing the best methods to date for increasing the resolution of a video sequence. Research is aimed at restoring the resolution of a video sequence with minimal loss of quality of perception. The article presents the results of the proposed algorithms in several aspects. In particular, the quality of restoration of specific objects is considered. Since this study is more focused on further research in the field of compatibility of methods for increasing the detail of objects and their detection.