KNN分类器在新疆维吾尔药材图像分类中的应用

Classification of Xinjiang Uygur medicine image based on KNN Classifier

目的：探讨 K 近邻结点算法（k-Nearest Neighbor algorithm，KNN）分类器在新疆维吾尔药材图像分类中的应用。方法采用 KNN 分类器对新疆维吾尔药材图像的灰度-梯度共生矩阵特征和 Tamura 纹理特征进行判别分类。选取训练样本为80、100、120的3个训练集，训练并得到最优 K 值，并分别在测试样本为120、100、80的3个测试集中验证结果。结果K 值越小（3～13），KNN 分类器对叶类图像分类准确率越高；K 值越大（63～71），KNN 分类器对花类图像分类准确率越高。当 K 值取3～13时，120、100、80的3个测试集中叶类图像的平均分类准确率分别为94．72％、89．45％、82．61％；K 值取63～79时，120、100、80的3个测试集中花类图像的平均分类准确率分别为74．71％、72．79％、76．55％。结论KNN 分类器可为新疆维吾尔药材图像类型判断提供一定的依据，为新疆维吾尔药材图像检索系统的检索精度的提升奠定了基础。

Objective To investigate the classification capability dealing with Xinjiang Uygur medicine by means of k-Nearest Neighbor algorithm （KNN）classifier.Methods Matlab was used to preprocess and ex-tract features based on gray gradient co-occurrence matrix and Tamura texture features.KNN classifier was used to classify image features.We selected training samples of 80,100,120 as train sets,trained and got optimal k value,then tested results in three test samples of 120,100,80.Results The smaller the k value （3 - 13 ）was the higher accuracy KNN classifier on the leaf image classification.The average accuracy rate in three test samples reached 94.72%,89.45% and 82.61% respectively.The larger k value （63-71）was the higher accuracy KNN classifier on the flower image classification.The average accuracy rate in three test samples reached 74.71%,72.79% and 76.55% respectively.Conclusion The data show that when adopting mixed texture combined with the KNN classifier,the classification ability can be im-proved and provide a certain basis for judgment of Xinjiang Uygur medicine types.This laid the foundation for improvement of accuracy of Xinjiang Uygur medicine image retrieval system.