基于矢量量化的猪咳嗽声识别

发布时间：2018-08-06 17:43

【摘要】：针对猪的规模化养殖中频发的呼吸道疾病问题,提出通过监测咳嗽状况对猪的健康状况进行预警,以谱减法去噪和端点检测为猪咳嗽信号主要预处理方法,以矢量量化(vector quantization,VQ)匹配算法为核心算法,分别构建基于标准梅尔频率倒谱系数(Mel-frequency cepstral coefficient,MFCC)和改进的MFCC 2种猪咳嗽声识别模型。测试结果显示,以标准MFCC为特征矢量构建的识别系统的识别率、误判率和综合识别率分别达到88%、14%和87.3%,基于改进的MFCC特征矢量构建的识别系统与之相比有很大提高,其识别率、误判率和综合识别率分别达到91%、12%和90.0%。试验表明,采用改进的MFCC与矢量量化相结合构建猪咳嗽识别系统是可行的,能够应用于猪的呼吸道疾病预警。
[Abstract]:Aiming at the problem of respiratory tract diseases frequently occurring in the large-scale breeding of pigs, this paper puts forward that the main preprocessing methods of pig cough signal are spectrum subtraction de-noising and endpoint detection through monitoring cough condition. Based on the vector quantization (VQ) matching algorithm, two kinds of pig cough recognition models based on standard Mel-frequency cepstral frequency cepstrum coefficients (Mel-frequency cepstral Cepstrum) and modified MFCC (modified MFCC) were constructed. The test results show that the recognition rate, error rate and comprehensive recognition rate of the recognition system based on the standard MFCC feature vector are 8814% and 87.3%, respectively. The recognition rate of the recognition system based on the improved MFCC feature vector is much higher than that of the system. Error rate and comprehensive recognition rate reached 91% and 90%, respectively. The experimental results show that it is feasible to construct pig cough recognition system by combining improved MFCC and vector quantization, and it can be applied to early warning of respiratory diseases in pigs.
【作者单位】：华中农业大学工学院;生猪健康养殖协同创新中心;华中农业大学动物科技学院动物医学院;
【基金】：湖北省自然科学基金项目(2014CFB317) ”十三五”国家重点研发计划项目(2016YFD0500506) 现代农业产业技术体系项目(CARS-36)
【分类号】：S858.28;TN912.3

【参考文献】