Classification of Vocalization Recordings of Laying Hens and Cattle Using Convolutional Neural Network Models

Purpose Vocalizations of livestock convey information about the health and behavior of the animals, and vocal analysis could be a useful method to monitor livestock. We propose a deep learning classification of vocal recordings of laying hens and cattle with the aim of automatically classifying laying hen and cattle sounds in South Korea using a deep learning model. Methods Audio and video recordings of laying hens and cattle were acquired. We classified laying hens’ sounds into eight classes and cattle sounds into nine classes. Classified audio files were used for the development of convolutional neural network (CNN) models. Two types of CNN structures, one based on 2D ConVnet and the other based on a 1D model with long short-term memory, were developed and tested for modeling to classify the vocalizations of laying hens and cattle. Results The classification model based on 2D ConVnet performed better with a satisfactory classification accuracy of 75.78% for laying hens and 91.02% for cattle. Conclusion Based on the results for the developed CNN models, it is expected that real-time voice monitoring could be applicable for providing animal physiological information to growers.