基于融合特征的无人履带车辆道路类型识别方法  

Road Types Identification Method of Unmanned Tracked Vehicles Based on Fusion Features

作　　者：刘佳[1] 刘海鸥[1] 陈慧岩[1] 毛飞鸿 LIU Jia;LIU Hai ou;CHEN Huiyan;MAO Feihong(School of Mechanical Engineering,Beijing Institute of Technology,Beijing 100081,China;Science and Technology on Vehicle Transmission Laboratory,China North Vehicle Research Institute,Beijing 100072,China)

机构地区：[1]北京理工大学机械与车辆学院,北京100081 [2]中国北方车辆研究所车辆传动重点实验室,北京100072

出　　处：《兵工学报》2023年第5期1267-1276,共10页Acta Armamentarii

基　　金：武器装备预先研究项目(20195208003)。

摘　　要：无人履带车辆行驶路况复杂,将道路类型作为无人履带车辆悬架控制、自动换挡决策、路径规划等任务的先验信息,有利于提升车辆性能。针对使用单一信号识别道路类型环境适应性差或准确率低的问题,提出一种基于融合特征的道路类型识别方法,将图像的深度特征和悬置质量垂向加速度时域、频域、功率谱密度信号的统计特征相结合,利用机器学习分类算法实现道路类型识别。对单一特征和融合特征进行对比发现:融合特征实现了图像特征和悬置质量垂向加速度特征的互补,提高了道路类型识别的准确率和环境适应能力;融合特征方法与仅使用图像特征的方法实时性相差极小。对多种机器学习分类算法进行对比,试验结果表明:支持向量机和随机森林在准确性和实时性方面都表现优越,总体准确率均可以达到90%以上,识别速度可以达到14帧/s。Unmanned tracked vehicles often navigate challenging terrain,and incorporating road types as priori information for tasks such as suspension control,automatic gear decision,and path planning can enhance their performance.However,methods based on single-class features have limitations in accuracy and environmental adaptability.To overcome this,a road-type identification method based on fusion features is proposed,combining deep image features with the statistical features of vertical acceleration in time,frequency,and power spectral density domains.Machine learning classification algorithms are used to identify the road types.Compared to using single class of features,the proposed method using fusion features enriches image features and vertical acceleration features,and improves the accuracy and environmental adaptability.The response speed of the method based on fusion features is similar to that of the image-based methods.Five machine learning classification algorithms are compared.The results show that support vector machine and random forest are the most accurate and fastest classification algorithms,achieving over 90%accuracy with a speed of 14 frames per second.

关 键 词：无人履带车辆 融合特征 机器学习 卷积神经网络 道路类型识别 

分 类 号：O235[理学—运筹学与控制论]