栖架养殖模式下蛋鸡发声分类识别

Classification Methods of Vocalization for Laying Hens in Perch System

摘要:

针对栖架养殖模式下蛋鸡的发声，采用频谱分析技术，运用音频分析软件Sound Analysis Pro提取不同行为状态下的发声图谱，采集其声学参数作为特征向量，应用J48决策树算法、朴素贝叶斯理论和支持向量机模型分别构建蛋鸡发声分类识别器，利用开源的数据挖掘平台Weka 3.6进行实验。结果表明，栖架养殖模式下，7:00~8:00的蛋鸡发声中，产蛋叫声、愉悦叫声分别占全部发声的42.2%、21.6%，相比于传统的笼养模式，有效地表达了蛋鸡生长过程中的自然行为和生理活动；基于J48决策树算法的蛋鸡发声分类模型识别率最高，达到88.3%，具有较好的识别效果，可运用于蛋鸡发声的实时监测和不同情感的分类识别。

Abstract:

Multi-taper spectral analysis was used to perform vocalization classification for laying hens in perch system. Sound Analysis Pro software was applied to compute spectral derivatives and acoustic features. Three methods including J48 decision tree algorithm, NaiveBayes theory and support vector machine were used to classify sounds of laying hens by using the open source data mining tool of Weka 3.6. Experimental results showed that vocalization of egg laying process and pleasure notes accounted for 42.2% and 21.6% between 7:00~8:00, while the natural behaviors and physiological activities were strongly performed with a comparison to traditional cage system. It was found that J48 decision tree algorithm had the highest classification rate (88.3%) for vocalization of laying hens, which could be used for different animal vocalization.