Deep Reinforcement Learning based SAR Image Pre-Processing Algorithm with Finite Buffer LEO Satellite Networks

Abstract

As space technology advances, launching low Earth orbit (LEO) satellites become easier and LEO satellites are being used in various fields. In particular, LEO synthetic aperture radar (SAR) system is in the spotlight with many advantages, e.g., regardless of weather condition, 24 hour operation. SAR system can be used in various fields such as object detection and disaster observation. However, SAR image has speckling noise, so image pre-processing must be required. There are many researches on the SAR image processing, however, few publications are considering a buffer status. Therefore, in this paper, we suggest the optimal SAR image pre-processing in LEO SAR satellites and a ground station with finite buffer based on deep reinforcement learning (DRL). As a result of DRL simulation, while changing the buffer size of the LEO SAR satellites, efficiency of buffer was improved by selecting the optimal filter size according to the state of the buffer.