文献类型：期刊文献

中文题名：三维机载雷达主动成像的PointRCNN目标识别研究

英文题名：Research on PointRCNN target recognition for active imaging of 3D airborne radar

作者：王祥仲[1];李玉玲[2];陈福祥[2];王丰周[2];马永华[2]

第一作者：王祥仲

机构：[1]华北科技学院计算机学院,北京101601;[2]北京联合大学生物化学工程学院,北京100023

第一机构：华北科技学院计算机学院,北京101601

年份：2026

卷号：47

期号：1

起止页码：77-82

中文期刊名：激光杂志

外文期刊名：Laser Journal

收录：;北大核心:【北大核心2023】；

基金：国家自然科学基金项目(No.42377200);河北省自然科学基金项目(No.D2022508002)。

语种：中文

中文关键词：三维机载雷达主动成像;PointRCNN;目标识别

外文关键词：three dimensional airborne radar active imaging;PointRCNN;target recognition

摘要：为了有效提升目标识别的准确性,提出一种三维机载雷达主动成像的PointRCNN目标识别方法。利用三维机载雷达主动成像技术获取激光雷达图像,随后应用基于图的分割(GBS)算法对图像像素点进行初步分割。提取各类区域的像素值,并进行层次聚类分析,以确定各区域像素的类别标签。根据层次聚类结果和预设的类别范围,调整各像素点的初步分割结果。同时,依据区域合并规划,生成新的分割图像,从而确定激光雷达图像中目标的位置区域。引入流形自注意力机制对PointRCNN的点云编码网络进行改进,应用改进后的PointRCNN进行目标识别。实验结果表明,所提方法能够实现高准确率的激光雷达图像目标识别。
In order to effectively improve the accuracy of target recognition,a PointRCNN target recognition method for active imaging of three-dimensional airborne radar is proposed.Using 3D airborne radar active imaging technology to obtain LiDAR images,and then applying graph based segmentation(GBS)algorithm to perform preliminary segmentation of image pixels.Extract pixel values from various regions and perform hierarchical clustering analysis to determine the category labels of pixels in each region.Adjust the preliminary segmentation results of each pixel based on the hierarchical clustering results and the preset category range.At the same time,based on the regional merging plan,new segmentation images are generated to determine the location area of the target in the LiDAR image.Introduce manifold self attention mechanism to improve the point cloud encoding network of PointRCNN,and apply the improved PointRCNN for object recognition.The experimental results show that the proposed method can achieve high accuracy in laser radar image target recognition.