The Research Of Animal Behavior Recognition System Based On Sound Features

Along with economic development, increasingly serious pollution of the environment,predatory human development of natural resources, illegal hunting of wild animals, seriously damaged to the animals' habitat and population structure, so that the protection of wild animals has become more urgent. But the reality is that wild animals normally live in deserted the field, the human is usually heard and not seen his form, for research and protection of animals to be difficult. Therefore, based on animal sounds to study the animal's living condition has practical significance, while biological research and environmental monitoring is also of great significance. Communicate with each other through the language of human transmission of information, but in most cases is through animal calls and other members within a population exchange, which calls on the behavior of animal identification is based on certain facts.The purpose of these article topics of digital signal processing theory, pattern recognition theory, and related cutting-edge computer technology used in animal protection, developed a set of automatically monitoring the feedback on the state of animal survival intelligent simulation system. Through the study of existing voice recognition technology, after got the recognition principle and structure of the system, to model the animals' sound of special behavior as the standard model of the behavior, finally we make a determine by the setting decision strategy.Gaussian mixture statistical model (GMM) with many other models unparalleled advantages to more robust description of sound signals, to capture one of the important acoustic characteristics. Therefore, this paper relies on BREWMP platform using the Gaussian mixture statistical model identification based on the behavior of animal sounds do basic research. The following major elements:(1)Studied BREWMP platform of sound basic methods of signal processing, proposed for the QUALCOMM unique set of audio format (QCP format) decoding algorithm, and discussed the algorithm to decode the data as the input for model training way;(2) Studied specific behavior of the animal sounds of the law, through the call signal in time domain and frequency domain feature extraction, and make the evaluation of the extracted characteristic parameters, to find more flexibility and robustness to the characterization of specific animal behavior characteristic parameters of sounds;(3) Specific to dog attacks, warning, please barking as the object of study, used their power spectrum, MFCC, pitch frequency as characteristic parameters, to establish GMM model for each behavior, in the calculation of model characteristics when the likelihood of the time-domain characteristics and frequency domain features were weighted, and time domain characteristics to improve the weight, and then treat calls signal measurement likelihood calculation, through the decision-making significance of animal behavior to identify the purpose.On this basis, on the BREWMP simulation environment, to establish test platform of behavior recognition system this based on animals' calls. Test showed that the system can be better achieved through acts of animal sounds to identify the function of identification obtained satisfactory results.In this paper, the dog as the research object, through the characteristics of their calls to identify attacks, warning, please, three types of behavior to the purpose of further analysis of behavior with other dogs call the behavior characteristics of the biological characteristics of wild animals, provides a theoretical basis and methods.