Radar target identification using HRRP-based features and Extreme Learning Machines

Radar target recognition performance using extreme learning machines (ELM) is examined in this study and compared with optimal classifiers. Classification under various adverse scenarios involving additive noise, azimuth ambiguity, azimuth mismatch between library and unknown target, presence of extraneous scatterers, signature occlusion, absolute phase knowledge, etc. are examined. ELM can be trained expeditiously and are suited for radar target recognition particularly with large training database. The effectiveness of ELM (single layer or multilayered) as a target recognition tool is the focus in this study that relies on real radar data collected in a compact range environment using a stepped-frequency system.