HollowBox: An anchor?free UAV detection method

Unmanned aerial vehicles (UAV) “black flight” incidents cause serious security risks and economic losses to airports. Additionally, existing UAV detection methods are mainly radar technology at Whereas, it is unable correctly identify the number of UAVs visualise their size, which undoubtedly poses a risk. Accordingly, an anchor-free method HollowBox proposed supplement equipment for airport problem. It inspired by FoveaBox object method, feature layers reset allocation ratio positive negative samples in training phase redefined, according multi-size characteristics. Extensive experiments show that, approach achieves 90.1% AP, 6% false rate 17.2 FPS inference speed, accomplished satisfactory performance, as verified real-shot data collected Tianjin. This work great significance application detection.