基于声信号时频域特征融合的路口车辆检测方法  

Intersection vehicle detection method based on time-frequency domain feature fusion of acoustic signals

作　　者：毛盼娣 廖晓文 徐道连[3] MAO Pan-di;LIAO Xiao-wen;XU Dao-lian(School of Electrical Engineering and Intelligent Manufacturing,Chongqing Metropolitan College of Science and Technology,Chongqing 402167,China;School of Arts and Media,Chongqing Metropolitan College of Science and Technology,Chongqing 402167,China;College of Optoelectronic Engineering,Chongqing University,Chongqing 400030,China)

机构地区：[1]重庆城市科技学院电气工程与智能制造学院,重庆402167 [2]重庆城市科技学院艺术传媒学院,重庆402167 [3]重庆大学光电工程学院,重庆400030

出　　处：《计算机工程与设计》2024年第12期3764-3771,共8页Computer Engineering and Design

基　　金：重庆市教委科学技术研究基金项目(KJQN202002501);重庆市教育科学“十三五”2019年度重点课题基金项目(2019-GX-151);重庆城市科技学院校级科研课题基金项目(CKKY2024010)。

摘　　要：针对路口车辆检测分类任务中特征级融合效果不理想问题,提出一种分阶段的特征融合的解决策略。将时域和频域内的特征进行融合,结合传统的长短时记忆网络和卷积神经网络的优势,构建车型分类模型。实验结果表明,所提Conv-BiLSTM模型能够获得超过98%的分类准确度,取得最高98.76%的F1分数。实验结果有效地验证了特征融合的必要性以及分类模型改进的有效性,为解决路口车辆检测分类任务中的问题提供了一种可行的解决方案。Aiming at the problem that the feature level fusion effect is not ideal in the vehicle detection and classification task at intersections,a phased feature fusion solution strategy was proposed.The principal component analysis was used to fuse the features in the time domain and frequency domain,and the advantages of the traditional long term memory network and convolutional neural network were combined to build a vehicle classification model.Experimental results show that the classification accuracy of the ConvBiLSTM model proposed can reach over 98%,and the highest F1 score of 98.76%is achieved.Experimental results effectively verify the necessity of feature fusion and the effectiveness of the improved classification model,and it provides a feasible solution for solving the problem of vehicle detection classification at intersections.

关 键 词：多特征融合 时频分析 车型分类 车辆声信号 车辆检测 长短时记忆网络 卷积神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]