Research On Methods For Classification And Extraction Of Laying Hens’ Vocalizations In Complex Environment

Sound which could be used as an indicator to indicate animals’health and welfare is the main way to communicate between animals. Laying hens’vocalizations (Hy-Line Variety Brown) of different growth stages were recorded in different systems. In order to solve the problem that there was interference noise and human were required to select sounds in the process of sound analysis, the frequency domain characteristics of laying hens’sounds and noises were compared. Methods for rapidly detect and classify laying hens’sounds were proposed, and acoustic features of laying hens’ vocalizations in the perches system and the cage system were compared. It would be provided the basis of early warning system for commercial poultry house. The main conclusions of the paper are:(1) In order to detect laying hens’vocalizations in the perches system, a classification method based on power spectral density was developed, with average recognition rate of95%, and then an automatically detect software was developed by using Lab VIEW.(2) In order to detect laying hens’vocalizations in the commercial poultry house, a characteristics extraction algorithm based on the sub-band power ratio was proposed. A recognition model by J48decision tree on the WEKA platform was developed, with average classification rate of93.4%.(3) To get suitable de-noising method, the effect of removing fan noise by IIR filter, FIR filter and different wavelet de-noising methods including threshold, wavelet function, wavelet and wavelet decomposition level were compared. Results showed that the effect of de-noising was best by the wavelet de-noising method based on SURE, whose wavelet function was Daubechies and wavelet was Dbl2and wavelet decomposition level was5.(4) Calls, songs, vocalizations during egg laying period and screams in the perches system were used to build database. Characteristic parameters of different kinds of laying hens’sounds were different The frequency energy of vocalizations during egg laying period was the highest, the average sound level was79.5+2.9dB, while the duration time of calls and screams were short A method based on frequency energy, duration time and bandwidth was built, with average identification rate of81.8%. On this basis, a classification soft of laying hens’sounds was developed.(5) Acoustic features of laying hens’sounds in typical poultry houses were analysed, results showed that laying hens of different growth stages were almost quiet at night, while the activity of laying hens during laying period was larger than that of laying hens during incubation period at day. Duration time, main frequency and sound level of sounds of laying hens during incubation period were lower than that of sounds of laying hens during laying period, which was mainly due to more calls of laying hens during incubation period. There were more number and kinds of sounds in the perches system than that in the cage system. The number and kinds of sounds could be used to indicate animals’ welfare. Grunts of laying hens in low noise at night could be detected by laying hens’sounds classification soft, which can provide basis for respiratory diseases monitoring.