基于区域自适应阈值的无人机目标检测方法

UAV Target Detection Based on Region Adaptive Threshold

随着无人机在人们生活和工作中的普及,在给人们带来便利的同时,各种“黑飞”、“恶飞”时间层出不穷,严重威胁到低空区域空域安全,因此必须对加强无人机管控,而如何对无人机目标进行有效探测则是其中最重要的环节。针对YOLOv5小目标检测性能差、漏检严重等问题,论文提出了一种基于区域自适应阈值的无人机目标检测方法。首先,利用局部阈值分割找出目标疑似区域;其次,利用聚类方法分割疑似区域所在ROI送入YOLOv5检测网络,从而避免全局检测过程中图像压缩造成无人机目标特征丢失;最后,针对小目标置信度不高的问题,对检测结果采取基于目标尺寸大小的自适应阈值,增强对无人机小目标的检测率。

With the popularization of UAVs in people's lives and work,while bringing convenience to people,all kinds of"black flying"and"vicious flying"time emerge one after another,seriously threatening the safety of airspace in low-altitude areas,so it is necessary to strengthen UAV control,and how to effectively detect UAV targets is the most important link.Aiming at the YO-LOv5's poor detection performance of the small targets,this paper proposes a UAV target detection method based on region adap-tive threshold.Firstly,the local threshold segmentation is used to find out the suspected target area.Secondly,the ROI where the suspected region is located is fed into the YOLOv5 detection network to avoid the loss of small target pixel features such as drones caused by compression during the full-map detection process.Finally,to address the issue of low confidence of target,the adaptive threshold of target size is adopted for the results to enhance the detection rate of UAV.