摘要
在被动声纳目标的分类识别中,不同的特征提取方法提取的特征反映了噪声信号不同的特性,一般情况下,很难做出某种方法优劣的判断。如何把这些不同的特征提取方法提取的特征矢量融合起来,对被动声纳目标分类识别非常有意义。在应用数据融合的方法对基于倒谱的特征提取方法、基于局域判别基的特征提取方法和基于听觉响度特征提取方法提取的特征矢量进行融合。提出了基于正态分布的概率密度函数的确定基本概率赋值的方法,利用三种特征提取方法对水声目标噪声信号进行特征提取,对提取的特征矢量进行融合,并进行分类实验,结果表明,特征融合使分类过程中的不确定性样本数减少,从而相应地提高目标分类的正确概率。
The feature vectors, which are extracted by different methods, show the different characteristics of passive sonar targets. Generally, it is hard to tell which feature vector is good. So how to combine these feature vectors becomes important in passive sonar target recognition. The feature vectors are combined by using data fusion in passive sonar target recognition. The method, which is used to computer basic probability assignment, is proposed based on the probability density function of normal distribution. Three feature vectors of noise signals are combined. The fusion experiment is performed. The results show that the number of undetermined samples is decreased and the correct recognition probability is increased.
出处
《计算机仿真》
CSCD
北大核心
2009年第8期326-329,共4页
Computer Simulation
关键词
特征融合
目标识别
基本概率赋值
Feature fusion
Target recognition
Basic probability assignment