基于KH-KELM的鸟类声音分类识别  

Bird sound classification and recognition based on KH-KELM

作　　者：沈希忠 陈菱 SHEN Xizhong;CHEN Ling(School of Electrical and Electronic Engineering,Shanghai Institute of Technology,Shanghai 201418,China)

机构地区：[1]上海应用技术大学电气与电子工程学院,上海201418

出　　处：《应用技术学报》2023年第3期279-285,共7页Journal of Technology

摘　　要：鸟鸣是鸟类生物学最重要的特征之一,鸟鸣特征参数的选取和鸟鸣分类提高精度是学者们一直研究的方向。基于鸟鸣识别技术提出基于磷虾群优化的核极限学习机(KH-KELM)分类模型:采用Mel频率倒谱系数(MFCC)对上海周边具有代表性的30种鸟类声音信号进行特征提取,提取出的特征参数用极限学习机(ELM)作为基础分类模型进行识别和分类,结合核函数思想优化基础模型并使用磷虾群算法(KHA)对训练参数优选,实现对鸟鸣信号的识别分类。为验证磷虾群算法优化的核极限学习机分类模型的分类效果和分类稳定性,对5、10、20和30种鸟类声音信号进行分类,测试结果表明,与极限学习机(ELM)、反向传播神经网络(BP)、支持向量机(SVM)和核极限学习机(KELM)分类模型对比,并与基于遗传算法(GA)、粒子群算法(PSO)和蚁群算法(ACO)优化的核极限学习机(KELM)模型对比,磷虾群算法优化的核极限学习机分类模型的分类识别率分别为99.65%、97.79%、94.48%和89.21%,具有最好的分类精度、分类稳定性和更强的泛化能力。Birdsong is one of the most important features of bird biology.The selection of bird song characteristic parameters and the improvement of birdsong classification accuracy have been the research directions of scholars.Based on birdsong recognition technology,a kernel extreme learning machine classification model based on krill herd optimization was proposed.The mel frequency cepstral coefficient(MFCC)was used to extract the features of the representative 30 kinds of bird sound signals around Shanghai.The extracted feature parameters were identified and classified by extreme learning machine(ELM)as the basic classification model.The basic model was optimized with the combination of the kernel function idea.The krill herd algorithm(KHA)algorithm was used to optimize the training parameters for the realization of the recognition and classification of bird song signals.In order to verify the classification performance and stability of the krill herd-optimized kernel extreme learning machine(KEML)classification model,5,10,20 and 30 bird sound signals were classified and compared with the ELM,BP,SVM and KELM classification models,as well as the KELM model based on genetic algorithm(GA),particle swarm optimization(PSO)and ant colony optimization(ACO).The results showed that the classification recognition rates of the krill herd optimization kernel extreme learning machine classification model were 99.65%,97.79%,94.48%and 89.21%,respectively,with higher classification accuracy,stability and stronger generalization ability.

关 键 词：鸟鸣信号 MEL频率倒谱系数 核极限学习机 磷虾群算法 分类识别 

分 类 号：TN911.73[电子电信—通信与信息系统]