面向高机动目标检测的激光雷达探测图像分角域识别方法  

A Corner Domain Recognition Method of Lidar Detection Images for High Maneuvering Target Detection

作　　者：韩钰[1] 王磊[1] 郑金亮[1] 王紫玉 HAN Yu;WANG Lei;ZHENG Jinliang;WANG Ziyu(Jianghuai College of Anhui University,Hefei 230031,China)

机构地区：[1]安徽大学江淮学院,合肥230031

出　　处：《重庆科技学院学报（自然科学版）》2024年第3期93-98,110,共7页Journal of Chongqing University of Science and Technology：Natural Sciences Edition

基　　金：2022年度安徽省高等学校科学研究项目(自然科学类)重点项目“基于PP-YOLO的目标表面瑕疵检测研究”(2022AH053061)。

摘　　要：高速移动和快速变换的运动模式容易导致目标模糊,从而无法准确识别目标的轮廓、形状和位置。为此,提出了面向高机动目标检测的激光雷达探测图像分角域识别方法。首先,通过邻域范围内像素点距离异常检测,去除激光雷达探测图像中的噪声和异常信号;然后,利用L-R算法解决目标高速移动造成的图像模糊问题;最后,采用基于互信息的自适应角域划分方法将图像分割成不同的角度域,并通过卷积神经网络在各角域中进行高精度目标识别,以实现高机动目标激光雷达探测。实验结果表明,该方法能够有效去除高机动目标造成的激光雷达探测图像模糊现象;相较于其他传统方法,该方法的目标识别率和全类平均精度较高、单图识别平均耗时较低,具有良好的识别效果。The high-speed movement and rapid transformation of motion modes can easily cause target blurring,making it difficult to accurately determine the contour,shape,and position of the target.To this end,a corner domain recognition method for lidar detection images targeting high maneuverability target detection is proposed.Firstly,by detecting abnormal pixel distance within the neighborhood range,noise and abnormal signals in the LiDAR detection image are removed;Then,L-R algorithm is used to solve the image blur problem caused by high-speed moving targets;Finally,the image is segmented into different angle domains using an adaptive corner domain partitioning method based on mutual information,and high-precision target recognition is performed in each angle domain through convolutional neural networks to achieve high maneuverability target lidar detection.The experimental results show that this method can effectively remove the blurring phenomenon of lidar detection images caused by high maneuvering targets;Compared to other traditional methods,this method has higher target recognition rate and overall average accuracy,lower average time consumption for single image recognition,and has good recognition performance.

关 键 词：激光雷达探测图像 高机动目标 分角域 互信息 卷积神经网络 

分 类 号：TN957[电子电信—信号与信息处理]