第42卷第5期Vol.42,No.52023年9月JournalofAppliedAcousticsSeptember,2023⋄研究报告⋄基于子频带能量特征提取的汽车鸣笛声识别∗侯晓飞1穆瑞林1,2†周晋3贾自杰1(1天津科技大学机械工程学院天津300222)(2天津市轻工与食品工程机械装备集成设计与在线监控重点实验室天津300222)(3天津市房地产市场服务中心天津300222)摘要：为了快速准确地识别城市中汽车违法鸣笛声并将不同种鸣笛声进行分类，该文应用子频带能量提取鸣笛声的特征，利用BP神经网络对提取的子频带能量特征值矩阵进行学习训练，且在神经网络学习过程中利用可变学习速度的方法，减小了神经网络的迭代次数。实验表明，利用此种子频带能量特征提取法使鸣笛声与非鸣笛声的平均识别率达到了94.889%；使不同鸣笛声之间的分类正确率最大达到了93.75%，实现了不同鸣笛声之间的分类。利用子频带能量法，能够很好地满足不同种鸣笛声识别与分类的需求。关键词：鸣笛声识别分类；子频带能量；特征提取；神经网络中图法分类号:TN912.34文献标识码:A文章编号:1000-310X(2023)05-1106-09DOI:10.11684/j.issn.1000-310X.2023.05.025Recognitionofautomobilewhistlesoundbasedonsub-frequencybandenergyfeatureextractionHOUXiaofei1MURuilin1,2ZHOUJin3JIAZijie1(1CollegeofMechanicalEngineering,TianjinUniversityofScience&Technology,Tianjin300222,China)(2TianjinKeyLaboratoryofIntegratedDesignandOnlineMonitoringforLightIndustry&FoodMachineryandEquipment,TianjinUniversityofScience&Technology,Tianjin300222,China)(3TianjinRealEstateMarketServiceCenter,Tianjin300222,China)Abstract:Inordertoidentifydifferentkindsofillegalcarwhistlingincitiesquicklyandaccurately,themethodofsub-frequencybandenergyfeatureextractionwasappliedintheclassificationandrecognitionofwhistles.Andtheextractedsub-frequencybandenergyeigenvaluematrixwastrainedbyBPneuralnetwork,andthenumberofiterationsoftheneuralnetworkintheprocessoflearningwasreducedbyusedthemethodofvariablelearningspeed.Theexperimentshowsthattheaveragerecognitionrateofwhistleandothersoundsis94.889%;andtheclassificationaccuracyrateofdifferentcarwhistlesoundis93.75%,theclassificationamongdifferentwhistlescanrealizedbythemethodofsub-frequencybandenergy.Therequirementsofdifferentwhistl...
