噪声环境下的仔猪求食发声信号识别  

The recognition of vocal signals from piglets begging for food in noisy environments

作　　者：李博宇[1] 赵翠莲[1] 吴佳伟[1] 徐昱琳[1] 

机构地区：[1]上海大学机电工程与自动化学院上海市智能制造及机器人重点实验室,上海200072

出　　处：《电声技术》2016年第7期44-49,共6页Audio Engineering

摘　　要：动物声音识别技术发展迅速,将其应用在养殖产业是目前国内外研究热点。以养猪业为例,为了在农场集约化养殖中,能够根据仔猪意愿及时让哺乳器自动送料进行饲喂,节省人力,提升禽畜福利。提出了通过识别仔猪饥饿哼叫声音进而控制自动哺乳器送料的方法。对仔猪饥饿求食哼叫,以及两种背景噪声:抢食尖叫噪声、机械送料噪声进行采集。通过时域,功率谱密度(PSD)等分析了这三种声音特征。通过提取Mel倒谱系数(MFCC)并用矢量量化(VQ)进行分类识别。针对混有背景噪声的声音识别,提出了一种基于VQ-PSD的识别方法。通过实验表明,基于此方法能够有效地提高对混有背景噪声的仔猪饥饿求食声音的识别率,平均识别率提高了9.1%。此方法也可拓展到于其他动物声音识别应用中去。The recognition technology of animal＇ s voice has developed rapidly, its＇ application in aquaculture industry is the research focus. The acoustic features of sounds from piglets begging for food and recognizing them in noisy environments are analyzed, which methods will be used to control a piglet feeding device in order to save manpower and improve livestock welfare in the intensive farming.There are many existing research based on Vector Quantization （VQ） algorithm in sound recognition. A combined VQ-PSD algorithm for recognition in noisy environments is presented in this paper. These vocaliza- tions are analyzed with the time domain and power spectral density （PSD） , and features of Mel-frequency cepstrnm coeffi- cient （MFCC） are extracted, and then recognized with the improved VQ-PSD algorithm.The results indicate that the aver- age recngnititm rate of this improved algorithm is 9. 1% higher than that of the VQ algorithm.This method can also be extend- ed to the sound recognition application of other animal.

关 键 词：仔猪 声音识别 禽畜福利 功率谱密度 

分 类 号：TN391.4[电子电信—物理电子学] S831.4[农业科学—畜牧学]