基于声信号多重分形和支持向量机的目标识别研究

摘要

In order to improve the recognition rate of smart landmines for armored target, as the acoustic signals radiated from armored vehicles have been proved to be nonlinear, an identification model based on multifractal analysis and support vector machine (SVM) was established. 40 sample signals for each armored target( a certain type of wheeled armored vehicle and a tank) running in different speeds(2 working conditions) were collected by outdoor experiment. The generalized fractal dimension spectrums ( GFDS) for both target signals were calculated based on multifractal analysis, and the characters of GFDS under 2 working conditions were analyzed. The GFDS values were input into SVM classification model, and the optimal identification results were obtained by training the model. After an identification effect comparison between GFDS and wavelet packet energy ( WPE) method, the results show that the model based on GFDS and SVM has a recognition rate of 92. 5% , which is higher than the 85% by WPE method.%为提高智能地雷对地面装甲目标的识别率,针对地面装甲目标辐射的噪声信号具有非线性的特性,建立了一种基于多重分形和支持向量机(SVM)相结合的分类识别模型.通过野外场地实验,采集到两种装甲目标在不同工况(运行速度)下的各40组样本信号；利用多重分形分析计算得到两种目标信号的广义分形维数谱(GFDS),分析了两种目标信号在不同工况下多重分形谱的特征；将GFDS值作为目标特征向量输入SVM分类模型,经训练得到最优分类结果,并与小波包能量(WPE)法提取样本特征后输入SVM的识别效果进行了对比,结果表明前者的识别率达到92.5％,高于后者的85％的识别率.