基于声阵列数据融合的运动车辆目标识别理论及方法研究

Classification of moving vehicles is an important part of the intelligent transportation system (ITS), and the recognition technology based on vehicle acoustic signals has been paid more and more attention because of its unique advantages. Now the acoustic target recognition technology is restricted to the internal characteristics of acoustic signals, and the vehicle classification characteristics are so limited that it’s difficult to further improve the recognition accuracy. In this project, acoustic array is introduced into the target recognition of moving vehicles, which is used to obtain the spatial distribution characteristics of vehicle sound sources, and realize the comprehensive extraction of target vehicle features by integrating with the characteristics of acoustic signals. The research contents are as follows: (1) Study the spatial moving sound source localization algorithm, and establish acoustic array structure optimization model based on the particle swarm algorithm; (2) Investigate the de-noising method based on multi-source delay characteristics of acoustic signals, and eliminate Dopple effect of moving sound source by setting up nonlinear time transformation model of acoustic signals; (3) Construct multiple feature set with reduced dimension for target vehicle; (4) Study the best matching combination method between classifier and feature vectors, based on which a recognition model with decision level information fusion will be built for moving vehicle classification. Finally, the moving vehicle target recognition results are verified by road tests. In this project, the theory and method of moving vehicle target recognition based on acoustic array is proposed, which has important theoretical significance and application value for improving the level of vehicle classification.

运动车辆分类识别是智能交通系统的重要组成部分，基于车辆声信号的识别技术由于其独特的优势而受到人们的重视。目前声目标识别技术仅限于挖掘声信号的内部特征，获取的目标车辆分类特征有限，因而难以进一步提高识别精度。本项目将声阵列引入运动车辆目标识别之中，获取车辆声源的空间分布特征信息，并结合声信号特征实现目标车辆特征的全面提取。项目的研究内容如下：（1）研究空间运动声源定位算法并建立基于粒子群算法的声阵列结构优化模型；（2）研究基于多源声信号时延特性的去噪方法，并建立声信号的非线性时间变换模型，消除背景噪声及多普勒效应的影响；（3）构建目标车辆的降维多重特征集；（4）研究分类器与特征向量的最佳匹配组合方法，建立运动车辆决策级信息融合识别模型。最后通过道路试验对目标分类识别结果进行验证。本项目系统地搭建基于声阵列的运动车辆目标识别理论及方法，对于提升车辆分类识别水平具有重要理论意义及应用价值。

运动车辆分类识别是智能交通系统的重要组成部分，基于车辆声信号的识别技术由于其独特的优势而受到人们的重视。目前声目标识别技术仅限于挖掘声信号的内部特征，获取的目标车辆分类特征有限，因而难以进一步提高识别精度。本项目将声阵列引入运动车辆目标识别之中，获取车辆声源的空间分布特征信息，并结合声信号特征及人工神经网络模型实现目标车辆的分类识别。本项目从目标声源定位、阵列结构优化、动态声目标的追踪、声信号特征提取及降维、多普勒效应消除以及声目标的分类识别几个问题出发，主要提出了以下创新方法：（1）基于阵列结构重构以及时频正交信号分离的关联模糊消除方法；（2）基于修正极坐标定位模型以及多孔阵列的远场声源定位精度提升方法；（3）基于粒子群算法的阵列结构优化方法；（4）基于改进智能粒子滤波的动态声源追踪方法；（5）基于小波变换的非平稳噪声信号阶次分析方法；（6）以非线性多项式拟合数值求解为基础的移动声信号多普勒消除方法；（7）基于互信息的二维MFCC特征矩阵降维方法；（8）融合互信息降维及注意力模型的声目标识别方法。所提出的理论模型及方法综合解决了基于声阵列的运动车辆目标识别的声源空间分布、特征信息获取及分类识别模型构建中的关键科学问题，提升了声源定位、运动轨迹以及声目标识别的精度和效率，并通过搭建的实验平台验证了所提出的方法的有效性。该项目的研究成果对于提升声阵列技术在智能交通、智慧城市乃至国防领域的应用水平具有重要的理论价值和指导意义。