Autonomous Sensor Tasking And Object Detection For Space Situational Awareness Using Machine Learning

The space domain is becoming increasingly crowded due to the proliferation of man-made resident space objects. Our limited number of object-tracking resources is becoming overburdened by a growing catalog of space debris and active satellites. This problem necessitates the most effective use of available sensors and observatories for catalog maintenance through the optimization of sensor tasking schedules. Given the complexity of the tasking problem, autonomous scheduling is essential and creates a need for increasingly powerful computational optimization algorithms. This research explores one such optimization approach, deep reinforcement learning, as an effective solution for autonomous sensor tasking. In addition to scheduling, improved object detection algorithms are in constant demand in order to more accurately identify resident space objects in images produced by optical sensors. This research additionally explores region-based convolutional neural networks as a potential considerable improvement over traditional detection algorithms.