一种基于三支决策的花卉图像分类

Classification Method of Flower Images Based on Three-way Decisions

传统的基于二分类模式的花卉图像分类方法因图像信息不完整和不确定,分类精度不高.三支决策因其获取事物更多信息后再进行延迟决策,成为一种有效的分类模型.为提高花卉图像分类精度,本文给出一种基于三支决策的花卉图像分类方法.首先利用单一特征对花卉图像进行分类,并依据决策状态值选取适当的阈值将图像集三分为POS1域、BND1域和NEG1域;其次定义域之间的转移规则,并分别对BND1域和NEG1域所包含的图像提取新特征;然后融合新特征后分别对BND1域和NEG1域包含的图像进行分类,并依据决策状态值选取适当的阈值对相应的图像集继续三分,依此进行多层"三分-治略",进而提高图像分类精度;最后采用牛津大学VGG小组的花卉图像集,实验验证了该方法的有效性.

The traditional classification methods of flower images based on the binary classification model are not good because the image information is incomplete and uncertain.The three-way decisions become an effective classification model because it can delay decision after obtaining more information.In order to improve the classification precision of flower images,a classification method of flower images based on the three-way decisions is proposed.Firstly,the images set is classified by single feature and trisected POS1 regions,BND1 regions and NEG1 regions by choosing appropriate threshold according to decision state value.Then,the rules which achieve transfer among three regions are defined,and the new feature of flower images is extracted respectively in BND1 and in NEG1.Next,the new feature is combined to classify the flower images in BND1 and NEG1,and the images set of BND1 and NEG1 continue to trisect by choosing appropriate threshold according to decision state value.According to these,the multi-level"trisecting-and-acting"is carried out and the precision is improved.Finally,the experiments are conducted on the flower images sets in VGG group of Oxford University and the validity of the method is verified.