%0 Journal Article
%T 待产梅山母猪咳嗽声识别算法的研究
%A 徐亚妮
%A 沈明霞
%A 闫丽
%A 刘龙申
%A 陈彩蓉
%A 许佩全
%J 南京农业大学学报
%D 2016
%R 10.7685/jnau.201510035
%X [目的] 目前对于母猪是否患有呼吸系统疾病的诊断主要依靠饲养员观察,存在由于疏忽未能及时发现并处理患病母猪而造成大量母猪死亡的情况。为解决这一问题,笔者以待产梅山母猪咳嗽声为对象,对其识别方法进行了研究,旨在将母猪咳嗽情况作为诊断早期呼吸系统疾病的依据,以达到自动监控的目的。[方法] 基于无线多媒体传感器网络进行母猪声音数据的采集与传输,对采集到的声音信号进行滤波、分帧等预处理操作后,由于不同声音的功率谱密度曲线的波动性不同,依托曲线目标优化的思想提取声音功率谱密度特征,并以此特征作为聚类中心,运用改进的模糊C均值聚类算法对咳嗽声和尖叫声进行识别分类。[结果] 训练出了母猪咳嗽声和尖叫声的功率谱密度特征,差异明显;忽略个体差异,咳嗽声和尖叫声的总体识别准确率分别约为83.4%和83.1%,识别算法是有效的。[结论] 针对待产梅山母猪咳嗽声,创新性提出了一种声音识别算法,该方法简单,高效,识别率高,为母猪呼吸系统疾病的早期自动诊断提供了技术支持。</br>[Objectives] Currently,it depends mainly on the breeder’s observation to diagnose if sows suffer from respiratory infections. But due to negligence,breeders didn’t timely detect and treat sick sows,which would result in a large sow death. To solve this problem,an identification method of predelivery Meishan sow cough as a study object was researched initially in this paper to make the sow cough condition as a basis of diagnosing early respiratory disease,and to achieve the purpose of automatic monitoring.[Methods] Sow sounds data could be collected and transmitted based on wireless multimedia sensor networks. According to the volatility of different sounds’ power spectral density line,the power spectral density characteristics of sow sounds were extracted relying on curve target optimization after the collected sound signals were preprocessed including smoothing,framing and so on. Making the power spectral density characteristics as cluster centers,an improved fuzzy C-means clustering algorithm was used to identify and classify sow cough and scream sounds.[Results] The power spectral density characteristics of sow cough and scream sound were trained out,and they were different significantly. Moreover,with ignoring the individual differences,the overall recognition accuracy of cough sounds reached about 83.4%,and that of scream sounds reached about 83.1%. Thus,the recognition algorithm worked efficiently.[Conclusions] To predelivery Meishan sow cough sounds,a simple and efficient sound identification algorithm with high accuracy was innovatively proposed,and provides technical support for diagnosing sow respiratory disease early and automatically
%K 母猪咳嗽声
%K 无线多媒体传感器网络
%K 功率谱密度
%K 目标优化
%K 模糊C均值聚类
%K 识别分类&lt
%K /br&gt
%K sow cough
%K wireless multimedia sensor networks
%K power spectral density
%K target optimization
%K fuzzy C-means clustering
%K identification and classification
%U http://nauxb.njau.edu.cn/oa/darticle.aspx?type=view&id=201604022