激光雷达硬件故障数据的模式识别研究  

Research on pattern recognition of lidar hardware fault data

作　　者：贾权 郭计云 盛彬 JIA Quan;GUO Jiyun;SHENG Bin(Shanxi Datong University College of Mechanical and Electrical Engineering,Datong 037003,China)

机构地区：[1]山西大同大学机电工程学院,山西大同037003

出　　处：《激光杂志》2022年第4期195-199,共5页Laser Journal

基　　金：山西省重点研发计划项目(No.201803D121020);山西省重点研发计划项目(No.201903D121070);大同市科技计划项目(No.2020022)。

摘　　要：利用当前方法对激光雷达硬件中的故障数据进行识别时,没有对数据进行预处理,导致准确率低、误报率高,识别时间长,因此,提出激光雷达硬件故障数据的模式识别研究方法。通过分析噪声和野值在小波变换域上的不同特性,利用双阈值小波变换法处理数据中的噪声点、野值,缩短方法的识别时间;对处理后的数据进行高维特征融合处理,提取数据的关联规则特征量,并利用故障诊断分类器的特点构建多故障分类器,将提取的数据特征量输入到分类器中,进行聚类分析、输出聚类结果,完成识别。实验对比结果表明,所提方法的准确率保持在80%以上,误报率始终低于0.5%,识别时间最高仅为0.4 s,提高激光雷达硬件故障数据提取与应用质量。When using current method to identify fault data in the lidar hardware, there is no preprocessing of the data, which leads to low accuracy, high false alarm rate and long recognition time. Therefore, this paper puts forward the research method of pattern recognition for the lidar hardware fault data. By analyzing different characteristics of noise and outliers in wavelet transform domain, the double threshold wavelet transform method is used to process the noise points and the outliers in the data, so as to shorten recognition time of the method. The processed data is processed by high-dimensional feature fusion, and the association rule feature quantity of the data is extracted, and the multi fault classifier is constructed by using the characteristics of the fault diagnosis classifier. The quantity is input into the classifier for clustering analysis and output results to complete the recognition. The experimental results show that the accuracy of the proposed method is above 80%, the false alarm rate is always below 0.5%, and the recognition time is only 0.4 s, which improves the quality of lidar hardware fault data extraction and application.

关 键 词：激光雷达 故障数据 预处理 小波变换域 特征融合 多故障分类器 

分 类 号：TN911[电子电信—通信与信息系统]