【摘要】 针对储粮害虫种类多、类别之间区分度比较小的特点,提出基于核函数主成分分析(KPCA)的粮虫特征提取方法。利用高斯径向基核函数,对特征选择后的10维原始数字特征进行核函数主成分分析,即通过非线性变换将样本数据从输入空间映射到高维特征空间,然后在高维特征空间进行特征提取。从类间可分性指数和粮虫分类效果2个方面,将KPCA法与传统的主成分分析(PCA)法进行了比较分析。结果表明,KPCA法对粮虫的非线性特征更为敏感,应用KPCA法提取的前2个特征,由最近邻分类器对粮仓中常见的9类粮虫进行分类,验证集的识别率为86.67%,在有效降低特征维数的同时,还保持了类别之间的可分性信息。更多还原

【Abstract】 According to the multi-species and less separable characteristics among various species of stored-grain insects,an approach for insect feature extraction based on kernel principle component analysis(KPCA) was proposed.Using the Gaussian RBF kernel function,ten morphological digital features of insects after feature selection were analyzed based on KPCA.The sample data were projected from the input space to high dimensional feature space through a nonlinear mapping function.By performing PCA on the high dimensional feature space,the nonlinear principal components of raw feature space were obtained.KPCA was compared with PCA from the separableness index among species and the recognition ratio of the classifier.The results indicated that KPCA was sensitive to the nonlinear features of insects.The nine categories of the stored-grain insects in grain depot were automatically recognized by the nearest neighbor classifier,making use of first two features based on KPCA,and the correct identification ratio was 86.67%.The experiment showed that the KPCA effectively reduced the feature dimensions and well kept the separability among species simultaneously.更多还原