基于矢量量化的猪咳嗽声识别

Recognition of pig cough sound based on vector quantization

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。

Respiratory tract diseases,frequently outbreaking in pig barns,bring economy lost in modern pig production.In this paper,a method was proposed to predict and monitor respiratory tract diseases by recognition of the cough sound of pigs.The sounds of pig were processed by the spectral subtraction denoising and the endpoint detection before being made into the improved Mel-frequency cepstral coefficients(MFCC),which are feature vectors input into the vector quantization matching algorithm(VQ algorithm).Thus a pig cough recognition model based on the VQ algorithm was developed to recognize the sound of cough.The results showed that the recognition rate,the false positive rate and the comprehensive recognition rate of the model based on the improved MFCC were 91%,12% and 90.0%,respectively,which were better than 88%,14% and 87.3% of the standard MFCC,suggesting the feasibility of the early warning of the respiratory tract disease of pigs in the future.