Target Feature Extraction And Classification Of Wireless Sensor Network

Wireless sensor network is composed of many sensor nodes which arranged by random. Each node has the capacity of environmental sensing, data processing capability, communications capability, as well as the signal self-organizing capacity of the network. Wireless sensor network technology known as the 21 most important emerging technologies, has been widely used in many occasions, such as military applications, the industrial production of security and counter-terrorism, the environment and monitoring of residential areas.In this paper, We use periodogram and AR-Burg maximum entropy power spectrum estimation method for spectral feature extraction, and use support vector machines for accurate classification of the types of vehicles. The experimental results show that The result shows that, under a certain classification precision premise, low compute complexity less generated-data volume can be obtained by the method of this paper.In this paper, the main task is extract the vehicle's sound and vibration signal spectrum and then proceed to the vehicle's Type classification. The data is from the time domain signal sequence which detect by nodes in WSN. All in all the task we have done as follows;(1) Grooming and learning the theory of the research background of wireless sensor networks, target feature extraction and classification, wireless sensor network technology development, network characteristics and key technologies of WSN.(2) As for the spectral feature extraction methods, the AR-Burg model was applied to wireless sensor network target spectral characteristics of the vehicle extraction, obtain good experimental results in this paper.(3) As for the spectral characteristics of the resolution, we propose changing the length of time-series before estimate the characteristics of single-spectrum. Comparison with conventional methods, the method of me increase the resolution of power spectral density, decrease generated-data (4) There is noise in the experimental. We obtain the impact of Gaussian white noise of background when classification by change the signal to noise ratio SNR(5) As for feature classification method. Support vector machines is to be use for verifying the above-mentioned method (2) and (3) in this paper. The effectiveness of the experimental results show that the method proposed in this paper is fully effective.All in all, although the AR-Burg spectral feature extraction and SVM classification of this paper is applied successfully in the field of wireless sensor networks, and obtain certain effects from experiment, but there are still many omissions and deficiencies need to be improved..