Target Detection using U-Net for a DTV-based Passive Bistatic Radar System

Digital Television (DTV) has a wider bandwidth than a radio signal, and the signal power is stronger than Wi-Fi or LTE, so it is advantageous in exploiting a Passive Bistatic Radar (PBR) signal for drone detection. Most of the PBR systems detect a target using a Constant False Alarm Rate (CFAR) detector. However, CFAR detection suffers from multi-clutters, noise, and sidelobes. To overcome the limitation of CFAR, we propose a target detector exploiting semantic segmentation. The proposed detector is based on U-Net, a model for semantic segmentation. Training datasets for the proposed detector are generated by synthesized DTV signals. The performances of the CFAR and the proposed detector were compared using actual drone measurement data. The proposed detector detects drones better than the CFAR one while reducing the number of false alarms.