摘要
织物领口性能评价是织物性能评价中的一个重要部分,为实现评价系统分类的自动化,提出了采用核Fisher判别分析(KFDA)方法来识别新样本的类别,将已分类的样本分为训练样本和测试样本。应用核函数将输入空间映射到特征空间,并在特征空间中求取训练样本投影矢量和构建判别函数组,然后用测试样本来验证判别函数组的识别效果。最后对KFDA、BP神经网络(BPNN)和径向基神经网络(RBFNN)3种方法的识别效果进行了对比,结果显示KFDA方法对于新样本具有较高的识别率。
Neckline performance evaluation is an important part for knitted sweater. In order to implement classification automation of testing system, kernel Fisher discriminant analysis （KFDA） is proposed for identifying the grade of new testing samples. Classified samples are divided into two parts： training samples and testing samples. Input space is mapped into feature space by kernel function, projection vector is acquired and discriminant functions is constructed in feature space, subsequently, identification effect is verified by testing samples. Finally, identification effects of KFDA, back promulgation neural network （BPNN） and radial basis function neural network （RBFNN） are compared. Experiment results show that KFDA method is more effective for identification of new samples.
出处
《纺织学报》
EI
CAS
CSCD
北大核心
2007年第3期76-78,共3页
Journal of Textile Research
关键词
核FISHER判别分析
特征提取
分类
领口
kernel Fisher discriminant analysis
feature extraction
classification
neckline
