Reinforcement Learning-Based Earth Observation System

Abstract

This chapter gives a reinforcement learning (RL)-based earth observation device for figuring out exciting occasions in satellite imagery. The device is based totally on a Markov selection system, with the RL set of rules supplying a means to optimize the choice-making technique by maximizing the expected reward. To effectively create the MDP surroundings, numerous photo features such as contrasting colors, item boundaries, and textures are applied to detect occasions of hobby. Then, movements and zooming in or converting the standpoint are taken to inspect the event further. The RL set of rules is constantly updated, primarily based on the rewards acquired from every movement taken. The chapter also describes a simulation set-up with a natural satellite TV for PC imagery that's used to illustrate the capacity of the proposed gadget. Ultimately, the system is evaluated by comparing its overall performance with baseline algorithms. Results show that the proposed gadget performs better for detecting exciting activities in satellite imagery.