摘要
针对复杂模式识别中的特征提取与选择问题,结合时间序列的参数模型和Fisher判别准则,提出了利用AR模型来拟合模式样本的时间序列,将模型参数作为原始特征矢量,然后在Fish-er鉴别准则函数取极大值的条件下,求得一组最佳鉴别矢量,最后再将高维原始特征矢量投影到这组矢量空间上来构成低维特征矢量的有效特征提取方法.对6类战场声目标的实测样本数据进行了实际的特征提取,分析了所提取特征的统计分布特性,并采用BP网络对用该算法所提取特征的有效性进行了检验,取得了令人满意的识别效果.
In view of the feature extraction and choice problem of complex pattern recognition, based on AR model parameters and Fisher's discriminant rule, an effective method of feature extraction for acoustic signal was presented. Taking the AR model parameters as initial features, under the condition of the value of Fisher's discriminant function equaling to max, a set of optimal discriminant vectors is worked out. So the low dimension effective features can be acquired through projecting the initial high dimension features to those optimal discriminant vectors. From the testing sample data of six kinds of battelfield acoustic targets, the valid features were extracted according to this algorithm, and their statisticdistributing characteristics were analyzed also. Through BP net, the validity of the features extracted according to this algorithm was proved.
出处
《上海交通大学学报》
EI
CAS
CSCD
北大核心
2003年第11期1732-1735,共4页
Journal of Shanghai Jiaotong University
基金
九五"国防预研基金资助项目(5.3.11.2)
关键词
特征提取
AR模型参数
FISHER判别准则
最佳鉴别矢量
声信号
feature extraction
AR model parameters
Fisher's discriminant rule
optimal discriminant vectors
acoustic signals
