论文部分内容阅读
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。
According to the characteristics of many kinds of stored grain pests and relatively small degree of differentiation among categories, a new method based on principal components analysis (KPCA) of kernel function was proposed. The Gaussian radial basis function is used to analyze the principal component analysis of the 10-D original digital features after feature selection. That is, the sample data is mapped from the input space to the high-dimensional feature space by nonlinear transformation and then is processed in the high-dimensional feature space Feature extraction. The KPCA method was compared with the traditional PCA method from two aspects of the inter-class separability index and the classification effect of the insects and insects. The results showed that the KPCA method was more sensitive to the nonlinear characteristics of the food insects. Using the KPCA method to extract the first two features, the nearest neighbor classifier was used to classify the nine kinds of food insects commonly found in grain silos. The recognition rate of the validation set was 86.67 %, While effectively reducing the number of features, but also maintain the separability of information between categories.