In this paper, a deep learning approach is proposed for target recognition based on high range resolution profiles (HRRPs) in multistatic radar systems. The proposed deep learning approach employs deep convolutional neural networks to automatically extract features from the HRRPs. In a multistatic radar system the target can be observed from multiple aspect angles simultaneously which improves the reliability and robustness of target recognition. In this paper, the spatial diversity offered by a multistatic radar system is exploited by averaging the slocal classifier output probabilities from the different monostatic and bistatic transmitter/receiver pairs. The highest global target probability is compared to a threshold to decide whether the target is classified to one of the preknown classes or as unknown. Simulation results show that the proposed deep learning approach has very reliable classification performance even at low signal-To-noise ratios.
|Title of host publication||2016 IEEE Radar Conference, RadarConf 2016|
|Publication status||Published - 3 Jun 2016|
|MoE publication type||A4 Article in a conference publication|
|Event||IEEE Radar Conference - Philadelphia, United States|
Duration: 2 May 2016 → 6 May 2016
|Conference||IEEE Radar Conference|
|Period||02/05/2016 → 06/05/2016|