Deep Reinforcement Learning Based Adaptive Synthetic Aperture Radar Image Filtering Algorithm

Abstract

Low Earth orbit (LEO) satellite synthetic aperture radar (SAR) is an efficient technology for Earth observation, and a lot of research is in progress. However, due to the time-limited property of LEO satellites, research on time-efficient SAR image processing is essential. In this paper, we propose an adaptive speckle noise filtering algorithm based on deep reinforcement learning for efficient processing of LEO SAR images. The proposed algorithm adaptively selects the filter size according to the buffer state to derive the maximum image resolution in a limited time. As a result of the simulation, the proposed algorithm selects the filter size more efficiently according to the buffer state than the method of conventional algorithm.