基于BP神经网络猪咳嗽声识别  

Recognition of pig cough sound based on BP neural network

作　　者：孙浩楠 仝志民 谢秋菊[1] 李嘉熙 Sun Haonan;Tong Zhimin;Xie Qiuju;Li Jiaxi(College of Engineering,Heilongjiang Bayi Agricultural University,Daqing,163000,China)

机构地区：[1]黑龙江八一农垦大学工程学院,黑龙江大庆163000

出　　处：《中国农机化学报》2022年第2期148-154,共7页Journal of Chinese Agricultural Mechanization

基　　金：国家自然科学基金项目(32072787)。

摘　　要：咳嗽是猪患呼吸道系统疾病发病早期的主要症状。为解决猪呼吸系统疾病难以被发现和人工监测准确率低的问题,提出利用BP神经网络来检测和识别猪咳嗽声音的方案。基于四麦克风阵列进行猪声音数据的采集,以猪咳嗽声、打呼噜声、尖叫声、哼哼声、咆哮声的声音为研究对象,对得到的声音数据进行滤波、端点检测等预处理,把梅尔频率倒谱系数(MFCC)作为猪声音特征参数,建立BP神经网络学习和识别的模型。经五折交叉法验证猪咳嗽声平均识别率为85.33%,猪非咳嗽声平均识别率为86.24%,识别率均在85%以上,结果表明所提出的方案是可行的。这种方法可以高效地识别猪咳嗽声,为猪呼吸道疾病发病初期的诊断提供技术支持。Cough is the main symptom in the early stage of respiratory diseases in pigs. In order to solve the problems of difficult detection of swine respiratory diseases and low accuracy of artificial monitoring, a scheme of detection and recognition of pig cough sound using a BP neural network was proposed. Based on four microphone array was the pig sound data acquisition, with pig cough, sneezing, shriek, grunts and roar sound as the research object, to get the sound data filtering, endpoint detection, such as pretreatment, MEL frequency cepstrum coefficient(MFCC) as the pig sound characteristic parameters of BP neural network learning and recognition model was established. The average recognition rate of pig cough sound was 85.33%, and that of pig non-cough sound was 86.24%. The recognition rate was above 85%. The results showed that the proposed scheme was effective and feasible. This method can effectively identify the sound of swine cough and provide technical support for the early diagnosis of swine respiratory diseases.

关 键 词：猪咳嗽识别 麦克风阵列 BP神经网络 梅尔频率倒谱系数 端点检测 五折交叉法 

分 类 号：S24[农业科学—农业电气化与自动化]