A New Strategy of Satellite Autonomy with Machine Learning for Efficient Resource Utilization of a Standard Performance CubeSat

A mega constellation of Nano/microsatellites is the contemporary solution for global-level Earth observation demands. However, as most images taken by Earth-observing satellites are covered clouds, storing and downlinking these results in inefficient utilization scarce onboard resources bandwidth. In addition, trend making satellite task execution plans ground operators demands efforts experts or simulators to predict real-time situation decide which tasks should be executed next. Granting controlled autonomy perform will boost mission effectiveness. We experimented with granting performing image classification scheduling. designed a convolutional neural network-based binary model more than 99% accuracy classifying clear cloudy images. The configured inference low-performance computers ordinary Cubesats. Moreover, we an autonomous scheduling mechanism based on reinforcement learning. It performs better custom heuristic-based method tasks. As result, proposed techniques machine learning ensured efficient memory, power, bandwidth highly resource-constrained CubeSat platforms accomplishment Nano/microsatellite constellations.