基于声谱图纹理特征的蛋鸡发声分类识别

Classification and Recognition of Laying Hens’ Vocalization Based on Texture Features of Spectrogram

为有效地辨别蛋鸡不同类型声音,了解蛋鸡的健康状况以及个体需求,提高生产效率的同时改善蛋鸡福利化养殖,提出一种基于声谱图纹理特征的蛋鸡发声分类识别方法。以海兰褐蛋鸡的声音为研究对象,将图像处理和声音处理技术相结合,由一维声音信号转换为二维图像信号,二维声谱图中的纹理特征呈现了蛋鸡声音的更多细节信息。最后,利用2DGabor滤波器提取蛋鸡发声声谱图中的声纹信息,并采用人工神经网络模型进行训练和分类识别。试验结果表明,本文方法平均灵敏度和平均精确度不低于92.0%,风机噪声识别灵敏度达99.3%,鸣叫声识别灵敏度最低,为76.0%。

Sound technology is an effective method to monitor animal behavior. Animal vocalization can reflect their individual health status and individual needs, and can be used as an assisted indicator for evaluating animal welfare and animal comfort level. In the process of laying hens’ breeding, it is helpful for farmers to understand their animals by effectively identifying different types of laying hens’ vocalization, so as to improve the production efficiency as well as animal welfare. A method of classification and recognition of Hy-Line Brown laying hens’ vocalization was introduced based on texture features of spectrogram. The method combined image processing with sound processing technology to analyze voiceprint information hiding in the two-dimensional spectrum of spectrograms from laying hens’ vocalization, and then the texture features were extracted from spectrogram by using2D Gabor filter. Subsequently, machine learning algorithm like backpropagation neural network was used for sound classification and recognition. Kinect for Windows V1 was selected as sound input device, and LabVIEW and Matlab software were used for developing the algorithm of sound data acquisition and sound analysis, respectively. The experimental results showed that the average precision rate and sensitivity rate were no less than 92.0%, and the sensitivity rate of fan noise was the highest one, which was 99.3%, and the sensitivity rate of normal calls was the lowest one, which was 76.0%. The research result can provide a visual and non-invasive method for farmers to identify the specific vocal behavior of laying hens, and also provide a feasible reference means for in-depth study of animal behavior and welfare.