Research On Far Field Sound Source Localization And Recognition Technology Of Distributed Pickup Device Array

Acoustic warning has the advantages of low power consumption and low production cost compared to radar warning and optical warning.The current acoustic early warning technology is mainly realized by arranging a ruled pickup array or an irregularly distributed pickup network,wherein the regularly arranged pickup array has problems of weak anti-interference ability,poor arrangement flexibility,and low detection fault tolerance.Although the irregular distribution pickup network has the advantages of large monitoring area,high detection precision and high fault tolerance,its processing algorithm has high complexity and the computational complexity in the fusion center is large.This paper studies the sound source localization and recognition technology based on distributed pickup arrays to combine the advantages of regular pickup arrays and irregular pickup networks to overcome the limitations of the two early warning methods.It has extremely important theoretical guiding significance for the establishment of an acoustic early warning network in the rugged environment of outdoor terrain,and has great application prospects in the fields of military investigation and urban security.Firstly,the research status of sound source direction finding,sound source identification and sound source localization based on data fusion of distributed pickup arrays is generalized and summarized.Secondly,the array structure of nodes is designed for the outdoor rugged scene of the early warning network,and the optimal deployment methods of node rule layout and random throwing are analyzed.The node signal preprocessing based on improved variable step size LMS adaptive noise cancellation is proposed.The convergence characteristics and the interference cancellation performance of the algorithm are analyzed.Next,the sound signal receiving model of L-shaped array is analyzed.The MUSIC two-dimensional direction finding algorithm for L-shaped array and sound signal is introduced.Based on this,the fourth-order cumulant MUSIC sound source two-dimensional direction finding algorithm which can suppress color noise is studied.And the cramer-rao bound of sound source two-dimensional direction finding for L-shaped array is analyzed.Then,the node direction finding data correlation system based on sound source identification is constructed.Aiming at the problem of sound source identification in the system,the MFCC sound source feature extraction method for suppressing high frequency components is proposed.Based on this,the missing alarm and false alarm performance of two sound source identification methods,GMM-UBM and i-vector-GPLDA,is compared and studied.Finally,a sound source localization model based on distributed multi-node direction finding crossover is constructed for the data fusion problem after node association.Based on this model,two data fusion localization algorithms based on least squares estimation and maximum likelihood estimation are introduced.An improved maximum likelihood estimation algorithm based on angle estimation weighting is proposed.The positioning accuracy of the improved algorithm and least squares estimation algorithm is compared and analyzed.