笼养蛋鸡发声类型的声学特性及自适模型研究

Vocalizations could be an indicator of different physiological conditions, emotional status and behavior contexts of poultry. Non-contact, non-invasive and continuous monitoring of sound analysis may supply us with hints on their current needs and well-being in an easy way. This study focuses on performances of laying hens (Hy-line Brown Variety, U.S. standard) with confined stacked-cage. A digital monitoring system will be designed to automatically record the mixed noise from the chicken houses and laying hen houses using sound array technology. The evaluation of inner noise quality and sound power contribution of the mechanic noise to acoustic environment is analyzed using the Beamforming method and the statistically optimized near-field acoustical holography method, including mechanically ventilated system, manure collection system, feeding system and egg collected system. Methods for endpoint detection and separation will be used to acquire laying hens’ vocalizations for the mixed noise. And the acoustic features including duration of call, averaged frequency and power spectrum of the sound during different breeding periods will be analyzed using sound processing software, and then the weekly variation curve of laying hens’ vocalizations could be depicted. Multi-parameter fusion algorithms can be used to establish suitable recognition algorithms and model for laying hens’ vocalizations, which is desired as an audio parameter for sounds of laying hens to indicate the feeding procedure and growth state. The result will offer an acoustic solution where non-invasive monitoring methods for laying hens’ management are urgently needed.

畜禽发声是其丰富机体信息、情感变化和生理状态的表达，利用其无接触、无应激和连续监测的优点，可以用来表征畜禽饲养需求和饲养状态。本项目以海兰褐蛋鸡为研究对象，针对叠层笼养模式，搭建基于传声器阵列的数字化监测平台，无接触的监测与记录蛋鸡养殖场育雏育成舍和产蛋鸡舍内的混合噪声，采用波束形成法和统计最优声全息法对笼养鸡舍内的混合噪声进行分析，量化舍内生产系统运行噪声（风机、清粪、喂料和集蛋系统）对总体噪声的贡献度。利用音频信号端点检测方法提取混合噪声中的蛋鸡发声，统计分析不同饲养阶段的发声类型及其持续时间、平均频率、功率谱等声学特征参数，揭示蛋鸡发声的周龄变化规律，研究建立基于多源数据融合算法的蛋鸡发声识别方法和自适模型，挖掘能够反馈蛋鸡发声时频特征的综合性声学指标，以此作为最佳音频参量表达蛋鸡的饲养过程与生长状态，为构建适合蛋鸡饲养监测的无应激式快速识别技术奠定基础。

畜禽发声是其丰富机体信息、情感变化和生理状态的表达，利用其无接触、无应激和连续监测的优点，可以用来表征畜禽饲养需求和饲养状态。项目通过对高密度叠层笼养模式下的生产系统设备噪声与蛋鸡音频的采集、分析和挖掘，获取了高密度叠层笼养蛋鸡舍主要设备系统噪声的发生时间、运行制度和分布规律，按对舍内声环境影响的排序依次为：通风系统、清粪系统、上料系统、饲喂系统、集蛋系统，合理调整设备运行制度，有利于减少噪声危害和对蛋鸡机体的损伤；综合考虑笼养鸡舍设备噪声的动态变化、蛋鸡发声的特征规律，提出了一种基于J48决策树分类模型的子带功率比特征提取识别算法，对于低噪声环境下音频平均识别率达到93.4%，可用于高密度叠层笼养蛋鸡舍内机械噪声与蛋鸡叫声的识别；采用蛋鸡声纹及音频表达参量的数字化方法，基于频谱分析、MLCT变换与汉明距离比对，提出了一种音频指纹方法与识别模型，同类或不同类音频数据间指纹比对的误码率介于0.30和0.80之间，提高了蛋鸡发声识别的准确性和鲁棒性。这些研究结果的取得为基于发声信息的蛋鸡饲养状态实时无应激的快速识别技术应用提供参考依据。