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

Abstract: 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.

Key words:  pig cough identification, microphone array, BP neural network, MFCC, end point detection, 5fold crossvalidation method