Research Of Vehicle Detection And Classification In Wireless Sensor Network

In military reconnaissance, ground reconnaissance has continuously automatic surveillance and reconnaissance function in the complex topography with the concealment and camouflage, expanding the temporal and spatial scope of information detection. Developed countries have attached great importance to its research and application. The recent development of wireless sensor networks have the characteristics of small size and random distribution. These make wireless sensor networks suitable for reconnaissance of the enemy, monitoring force, equipment and supplies.Vehicles detection, classification and monitoring are one of its important applications. Vehicles make strong acoustic and seismic signal while running. So vehicles are monitored using acoustic and seismic sensors. At the present time, wireless sensor network monitoring system at home and abroad have the two kind of sensor commonly. In order to monitor military vehicles, wireless sensor network are ordinarily disposed in jumping-off with bad environment. Therefore, there would be amounts of disturbing noises in the signals detected by the sensor nodes. It is a challenging task how to extract useful vehicle signals out of noises and make an accurate detection identification.In this paper, the vehicles are detected by using the acoustic and seismic signals produced while vehicles running under the background of wireless sensor network monitoring vehicles. There are several parts in the process of wireless sensor network detecting and identification: detecting the appearing of the target signals, pretreating the target signals, extract the features of signals, classifying and decision fusion by classifier. Fusion decision is made by synthetizing the single decision of the multi-nodes that detect the same target.In order to make special needs of detecting acoustic and seismic signal of vehicles in the wireless sensor network, a double threshold detecting algorithm is proposed based on constant false alarm rate and distribution of spectrum. It combines methods of the energy detecting in time domain and choosing maximum power detecting in frequency domain. Acoustic and seismic signals of vehicles can be extracted accurately form signals polluted severely by noises by the algorithm. After the de-noise processing by wavelet packet method and desample, the pretreatment of vehicle signals is finished.Whereafter, using FFT and wavelet packet analysis methods, features of target signals are extracted, which are sent to nearest neighbor classifier and support vector machine to be classified afterwards. The identification result of single node is obtained. Large amount of sensor nodes are deployed to detect target in wireless sensor network in monitoring area. Therefore, when target appears, several nodes will detect it and make their own identification results at the same time. Based on this character, a multi-nodes global decision fusion algorithm based on energy is proposed. The final identification result is obtained by global decision fusion.For evaluation purposes we use real data from DARPA's SensIT project, which contains lots of acoustic signatures of two types of vehicle type, a tracked vehicle and a heavy truck. By comparison with the relevant reference, our experimental results show that the algorithm is valid for battlefield vehicle classification in wireless sensor network.