Cognitive Radar Framework for Classification using HRRP and Waveform Diversity

We introduce a cognitive framework for non-cooperative target identification using high range resolution profiles. In a sequence of successive measurements, the transmit behaviour is changed to enhance the classification performance with a minimum number of measurements. In the current work these are linear frequency modulations with different bandwidths and centre frequencies. A partially observable Markov decision process fuses the information (type and angle) over time and dynamically selects the most promising waveform to increase the overall classification performance. The framework is tested on electromagnetic models of four different civilian cars to obtain a realistic environmental simulation. The validation shows up to 15% improvement of the correct classification compared to static approach with multiple measurements using a single waveform.