利用复合特征进行模式识别的探雷研究

根据对探地雷达回波信号的分析,提出了一种利用复合特征训练神经网络并用其进行地雷探测的新方法.整个过程包括有预处理、特征提取及神经网络分类三个步骤.在特征提取的过程中,选取时域、频域、小波域能量及其统计量,以及Welch功率谱密度估计作为地下埋设目标的特征,在此基础上使用WILKS准则采用逐步判别的方法抽取关键特征,从而降低特征维数并将特征送入神经网络训练.使用地雷目标与其相近物体的数据进行对比试验和神经网络测试,结果表明,使用复合特征训练的神经网络可有效地将地雷与其他干扰物分开,提高了地雷的探测率,同时降低了虚警率.

A Study on Mine Detection and Identification Through Pattern Recognition with the Help of Multi-Features

Based on the analysis of the echo from buried objects, a novel method of locating buried mines using pattern recognition is recommended. The process comprises pre-processing stage, feature-extraction stage, feature reduction and a neural network classification stage. A PNN neural network is employed to identify the objects by training it to recognize the features extracted from time domain and frequency domain, wavelet domain as well as Welch power spectral density estimation of signal segments reflected from various types of buried targets. The data concerning mines and some other objects which are like mines in shape and size are compared and tested with the network, and the results indicate that the neural network using multi-feature can improve mine detection greatly.