Research On Target Recognition Technology Based On Sound Sensor

At present,biometric recognition technologies such as fingerprint recognition,iris recognition,and voice recognition are developing rapidly and have been widely used in military,financial,and security fields.Because sound target recognition technology has the advantages of easy sample collection,low algorithm complexity,and strong concealment,it is more significant for more scenarios and has become a research focus in recent years.However,the related researches on the existing sound target recognition technology are mostly carried out in the laboratory,and they are not suitable for use in noisy environments such as the wild.This article mainly studies the recognition of sound targets in a noisy environment.Using a sound sensor placed in the wild environment to collect sound data,after a series of processing,it can finally recognize whether the sound is emitted by a person,a car,or a small aircraft.To this end,this article will discuss and study from the following aspects:(1)The sound signals collected in a noisy environment are pre-processed to extract relatively "pure" sound signals.This paper uses endpoint detection based on Teager energy operator to achieve the purpose of preprocessing.Most of the traditional endpoint detection methods are based on the short-term energy and short-term zero-crossing rate characteristics of sound signals.These two features are detected in the time domain and are simple and easy to implement,but they are not suitable for situations where the signal-to-noise ratio is relatively low,which increases the possibility of misjudgment.But in the wavelet domain,the wavelet coefficients of the useful sound signal segment are significantly larger than the silent and noise segments and are more stable.Therefore,endpoint detection based on the Teager energy operator in the wavelet domain is more suitable for sound target recognition in the wild environment.(2)The existing sound characteristic parameters are divided into two categories: time domain parameters and frequency domain parameters.The principles,advantages and disadvantages of commonly used sound characteristic parameters are introduced one by one.The sound data used in this paper is collected by sound sensors placed in the wild environment,which will cause problems such as low signal-to-noise ratio and instability.A single feature parameter can not achieve a good recognition effect,so the fusion feature of Mel frequency cepstrum coefficient and linear prediction cepstrum coefficient is used as the characteristic parameter of this study.(3)This experiment needs to identify whether the sound is emitted by a person,a car,or a small aircraft in a short time,so there is a certain requirement for the recognition speed under the condition of ensuring the recognition rate.Therefore,the final choice was to use a vector quantization model(VQ)for classification.After analysis and research,it is found that the current vector quantization models mostly use Euclidean distance as the distortion measure.Because the fusion features of Mel frequency cepstrum coefficient and linear prediction cepstrum coefficient are used as feature parameters in this paper,the log-likelihood ratio is used as the distortion measure of the vector quantization model.The model was designed and implemented with MATLAB.Experiments show that the recognition system can improve the recognition speed while guaranteeing the recognition rate,and can be applied to a wider range of scenarios.