决策树C4.5算法在新疆维吾尔草药图像分类中的应用

Application of Decision Tree C4.5 Algorithm in the Classification Xinjiang Uygur Herbal Images

目的:利用小波变换法提取新疆维吾尔草药图像的特征,对植物药图像进行分类研究。方法:此次研究选取新疆维吾尔草药图像200张,其中花类图像100张,叶类图像100张。对图像进行去噪、尺度归一化和空间转换等预处理。利用小波变换分别提取花类、叶类图像的特征向量,用类间距法获取具有较好的分类能力的特征量,使用决策树C4.5算法对特征的分类能力进行评价。结果:对得到的这些特征量,分别利用决策树C4.5算法和贝叶斯方法进行分类,决策树算法分类准确率达到了80.0%;贝叶斯方法分类准确率达到73.5%;结论:结果显示,采用小波变换提取的特征在对不同类型的维吾尔草药图像进行分类时,将最大类间距和决策树C4.5算法结合能达到一定的分类能力;因此,决策树分类算法可以在一定程度上对新疆维吾尔草药图像进行判别分类。

Objective： The extraction of the characteristics of Xinjiang Uygur Herbal images has been done by using wavelet transformation, to classfy and researches the Herbal images. Methods： This study selected 200 Xinjiang Uygur Herbal images, which consist of flower type 100 images, and the leaves type 100 images. The images were removed the noise by median filter, normalized scale and conversed type. Then the author use the four layer wavelet transformation to calculate the mean and variance of wavelet coefficients constitute the texture feature vector, and then evaluate the feature＇s classification ability by decision tree C4.5 algorithm. Results： The characteristics were classified using the method of decision tree C4.5 algorithm and Bayesian classification, respectively. The accuracy rate of Decision tree C4.5 algorithm classification reached to 80.0%, and the Bayesian method classification reached to 73.5%. Conclusions： The results show that the wavelet transformation of texture extraction method for the description of Xinjiang Uygur Herbal image features in this research, the classification of decision tree C4.5 algorithm can achieve a certain capacity. Therefore, the decision tree C4.5 algorithm can to a certain extent classify for Xinjiang Uygur Herbs image.