Research On Sound Source Target Recognition Method Based On Deep Learning

With the changes in the battlefield of modern warfare and the development of war concepts,target detection technology has received increasing attention in more complex combat scenarios such as close urban lane stations and anti-terrorism operations.In these complex environments,detection technologies such as optical and electromagnetic detection have heavily degrades the precision of target detection due to occlusion and interference.Based on sound wave signals with strong penetrating ability,sound source detection technology can be used as a supplementary detection method for optical and electromagnetic detection equipment to effectively Ascertain the source of enemy fire and the position of hostile personnel.This paper does the research on the sound field signal acquisition method and the sound source target recognition method.This paper puts more stresses on hardware system acquired from sound field and sound source target recognition method under the guidance of deep learning.At first,the paper makes a research about principles of signal propagation of far-field sound source and near-field sound source and provides a more specific understanding on acoustic source sign characteristic messages.It is determined that the sound source studied in this paper is based on the far-field sound source,and its propagation principle is studied.,Studied the role and extraction methods of various characteristic information of acoustic signals;secondly,this paper devises an overall hardware system program in the basis of sound field message.In the light of the features of sound signals and sound source recognition,the paper completes a microphone sound probe selection.The accomplishments of signal modulation circuit and sample data storage system help to archive the building of hardware system.;Then then the MFL-Res Net model of sound source target identification for the application of deep learning is introduced,and the multi-feature extraction method,convolutional network layer structure,and feature fusion layer of the model are layout.And feature classifiers,the method is verified on the DCASE2019 data set,and finally the sound source target recognition system composed of the sound signal acquisition hardware system and the deep learning sound source recognition method is experimentally verified.The source and target signals are collected and stored,and then input into the MFL-Res Net model for training after preprocessing and feature extraction,which verifies the feasibility of the sound source recognition system.The sound source target recognition system proposed in this paper has passed method verification and experimental test verification.In the DCASE2019 data set,under 8 acoustic type recognition conditions,it can achieve 87.6% classification and recognition accuracy;in the experimental verification stage,6 military target classification and recognition conditions Its recognition accuracy can reach 91.6%;at the same time,it can still achieve a better recognition effect even with a small sample size.The results suggest that the sound source target recognition system programmed in this paper can help to achieve the acquisition and accurate classification of actual sound source signals,and has certain anti-reverberation and noise capabilities.