Performance Maximization of Satellite SAR image Processing using Reinforcement Learning

Synthetic aperture radar (SAR) observes a wide area during the mission in space and synthesizes the acquired data into the image of the specific area in a ground station. One scene of SAR is composed of several hundreds of kilometers for one minute observation. In a ground station, the image processing time takes few hours for one scene. Therefore, an efficient method, considering the link time of satellite SAR and ground station, is of necessity to reduce the idle computing time. In this paper, we propose a method that achieves performance maximization of SAR image processing. The proposed method considers the active resource using reinforcement learning at the separated ground stations. We analyze the predefined satellite route and select processing level according to the link time. For the performance maximization, we set a reward at the available area which can process the data, and a penalty at the idle area in our reinforcement learning model. The simulation result shows the optimal list of processing levels for avoiding idle computing. In addition, the proposed method guarantees 18% of performance improvements.