基于雷达传感器和改进深度学习的无人机目标检测方法  

UAV Target Detection Method Based on Radar Sensor and Improved Deep Learning

作　　者：黄鸿柳 谭果 蒋林利 谢兴祥 HUANG Hongliu;TAN Guo;JIANG Linli;XIE Xingxiang(School of Mathematics&Computer Science,Guangxi Science&Technology Normal University,Laibin 546199,China;Experimental Training Centre,Guangxi Science&Technology Normal University,Laibin 546199,China;School of Electronics and Information Engineering,Harbin Institute of Technology(Shenzhen),Shenzhen 518055,China)

机构地区：[1]广西科技师范学院数学与计算机科学学院,广西来宾546199 [2]广西科技师范学院实验实训中心,广西来宾546199 [3]哈尔滨工业大学(深圳)电子信息工程学院,广东深圳518055

出　　处：《无线电工程》2023年第9期2143-2150,共8页Radio Engineering

基　　金：国家自然科学资金(42065004);广西高校中青年教师科研基础能力提升项目(2019KY0868)。

摘　　要：针对当前无人机目标检测技术精确度低、受环境影响大的缺点,依托运动相机与激光雷达设备,提出了基于改进深度学习的无人机目标检测方法:在深度图像网络架构的基础上,引入残差网络提升算法精度,利用MobileNet加速深度学习的过程,从而利用改进的RetinaNet算法实现无人机精确目标识别与定位;针对点云数据无法通过二维投影准确计算距离的问题,提出通过直方图网络精确估计检测目标的视觉距离。实验结果表明,在不同的复杂环境条件下,与Faster R-CNN相比,所提方法检测精度更高、使用场景更广、运算速度更快,平均精度比Faster R-CNN算法高出1.5%。Considering the disadvantages of low precision and fragility to changing environment for current UAV target detecting technology,and relying on sport camera and lidar,a novel UAV target detection method based on improved deep learning is proposed:Based on the depth image network architecture,this method applies residual network to improve the precision of the algorithm,uses MobileNet to speed up the deep learning process and adopts the improved RetinaNet algorithm to achieve the accurate target recognition and localization of UAV.As the point cloud data can not accurately calculate the distance through two-dimensional projection,a histogram network is proposed to accurately estimate the visual distance of the detected target.The experimental results show compared with FasterR-CNN,the proposed method has better detection precision,wider application scenarios and faster operating speed in different complex environmental conditions.The average accuracy is 1.5%higher than that of FasterR-CNN algorithm.

关 键 词：雷达传感器 无人机 目标检测 改进RetinaNet 残差网络 MobileNet 直方图网络 

分 类 号：TN24[电子电信—物理电子学]