论文部分内容阅读
针对粮仓害虫种类多、类别之间相似度比较高的特点,提出基于核Fisher判别分析的粮虫特征压缩方法.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Fisher判别分析,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.从粮虫分类效果方面,将KFDA法与FDA法、PCA法和KPCA法3种方法进行了比较分析.应用KFDA法提取的前4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93.33%.结果表明:KFDA法对粮虫特征的非线性比较敏感,在有效降低特征维数的同时,还提高了类别之间的可分性.
According to the characteristics of many kinds of grain farms and high similarities between categories, a new kernel compression method based on kernel Fisher discriminant analysis was proposed. The Gaussian radial kernel function was used to perform kernel Fisher Discriminant analysis means that the sample data is mapped from input space to high-dimensional feature space by nonlinear transformation and then extracted in high-dimensional feature space. In terms of the classification effect of grain insects, KFDA and FDA, PCA and KPCA Three kinds of methods were compared and analyzed.Firstly, the KFDA method was used to classify the first four features and the nearest neighbor classifiers were used to classify the nine kinds of food insects in the granary, the recognition rate of the validation set was 93.33% .The results showed that KFDA method Insect characteristics of the nonlinear insensitive, effectively reducing the number of features at the same time, but also improve the separability between categories.