基于ReliefF和SBS的赤潮生物图像特征选择方法研究

摘要

In order to put forward a real-time automatic classification method with high accuracy aimed at red tide algae, this paper proposes to use ReliefF-SBS for feature selection. Namely, we carry out feature analysis about red tide algae image original data set. And on this basis, we make feature selection to remove irrelevant features and redundant features from the original feature set, to get the optimal feature subset, and reduce their impact on the clas-sification accuracy. The paper also presents the results and the analysis, and meanwhile verifies the influences on two feature selection classifiers of k-Nearest Neighbor algorithm (KNN)and Support Vector Machine (SVM).%针对赤潮生物提出具有较高准确率的实时自动分类方法，本文提出采用ReliefF-SBS进行特征选择，即针对赤潮生物图像原始数据集进行特征分析，并在此基础上，对原始特征集进行特征选择以去除特征集中的无关特征和冗余特征，得到最优特征子集，减少它们对分类器分类精度的影响。文中给出了实验结果和分析，同时验证了对k-Nearest Neighbor algorithm(KNN)和Support Vector Machine（SVM）分类器分类效果的影响。