Reinforcement Learning Approach for a Cognitive Framework for Classification

The article presents an online learning approach for a cognitive framework for classification. The classification process is realised via sequential illuminations with different waveforms and modelled by partially observable Markov decision processes. Since the operational environment is not accessible, the agent is trained on a similar one. The difference between the trained and operational environment is learned without external knowledge and an existing model. In a first step, the capability of this new approach is shown on generic data. In a second step, it is used on high fidelity simulated data and different waveforms that create high resolution range profiles. The tests on the scenario with electromagnetic models, show noticeable improvement in comparison to the framework without learning capability.