Wireless Sensor Network(WSN) Sound Target Recognition Methods Research

Wireless sensor networks are a hot spot in recent decade’s research. In thedefense and national economy has a broad application prospect. Such as military,gas monitoring system, intelligent home furnishing, and environmentalsupervision, agriculture, and transportation, medical and other fields. Volumesmall WSN, a random distribution characteristic is very suitable for groundenemy reconnaissance; target category is one of the important applications.This paper is use wireless sensor network to realize sound recognition, thesound object are include people, cars, guns, tanks. The scope of the study oftarget recognition technology belongs to the theory of pattern recognition,recognition technology mainly lie in feature extraction and classifier. This paperadopts the JN5139node, as it does not have the ability to collect sound, so weshould buy sound module, and then use the sound module and JN5139node tobuild the sound extraction platform. To make the sound without distortion, soundsampling rate choose8000Hz. Because the resources and transmission capabilityof the sensor node is very limited,8000Hz voice data cannot be transmitted tothe computer directly, so this paper chooses sound feature extraction in the node,and then send the feature to PC for identification.This paper studies the sound collection, feature extraction algorithms andPC background recognition algorithms and PC software design, focuses onstudying the sensor nodes in the feature extraction algorithms and programming,and support vector machine principle and multi-class classification structure.This paper studied one verse one classification structure, one verse otherclassification structure, binary SVM multi-class classification structure anddecision-oriented acyclic graph classification structure respectively, andconducted experimental study, researching the influence factor of classification.Because of a large number of sensor nodes are composed the sensor network, somust carry on the multiple node decision algorithm research, and proposed avoting algorithm based on energy. In order to easy to real-time observation, thispaper also designs the PC interface in matlab, and the theory on the basis ofexperimental research was carried out, that can be observed target area categoriesin real-time.