Target recognition of perimeter intrusion defense system based on wavelet packet and BP neural network

The seismic signals generated by recognizing targets in a perimeter intrusion defense system based on a seismometer are very weak and difficult to be directly identified. So the signal features of the targets need to be extracted before target identification. This paper presents a new method of target recognition based on wavelet packet analysis and BP neural networks. Firstly, target motion signals captured by a front detector were proposed by using wavelet denoising. Then, the signals were decomposed and reconstructed with wavelet packet analysis, and the feature values of reconstructed signals were extracted to construct feature vectors. Furthermore, the feature vectors were used as the inputs of the BP neural networks to carry on learning and training various types of target characteristics. Finally, the trained neural network were used to identify the targets on-line. Recognition result for 30 groups of data from the seismometer (6 kinds of distance, 5 sets) shows that the desired output vector of the network and the actual output vector of the network is consistent, and the target recognition accuracy reaches to 99%. It concludes that this method can effectively identify the target of perimeter intrusion defense systems. © 2015, Chinese Academy of Sciences. All right reserved.