基于余弦定理和K-means的植物叶片识别方法  

Plant leaf identification method based on cosine theorem and K-means

作　　者：朱颢东[1] 申圳[1] 

机构地区：[1]郑州轻工业学院计算机与通信工程学院,郑州450002

出　　处：《华中师范大学学报（自然科学版）》2014年第5期650-655,共6页Journal of Central China Normal University：Natural Sciences

基　　金：国家自然科学基金青年科学基金项目(61201447);河南省基础与前沿技术研究计划项目(122300410287);河南省科技创新杰出人才计划项目(134200510025);河南省教育厅科学技术研究重点项目(13A520367);郑州市科技计划项目(121PPTGG362-12;131PPTGG411-8);郑州轻工业学院校级青年骨干教师培养对象计划项目(XGGJS02);郑州轻工业学院研究生科技创新基金项目

摘　　要：为了提高植物叶片识别准确率,提出一种基于余弦定理和K-means的识别方法.该方法首先通过提取叶片的Hu不变矩和形状特征得到叶片的综合特征向量,然后使用K均值聚类(Kmeans)对各类叶片训练样本的特征向量集合进行聚类以获得聚类中心特征向量,紧接着使用余弦定理计算目标叶片和训练样本的相似度并排序.仿真实验表明:在Flavia植物叶片数据库中进行测试,该文方法以96.03%的概率在前5位发现目标,优于KNN、BP神经网络方法,因此,该方法具有一定的实用价值.In order to improve the correct rate of identification and classification of plant leaves,a new method combined the cosine theorem with K-means was proposed.Firstly,the comprehensive feature vectors were obtained from Hu invariant moment and shape feature.Secondly,K-means was used to get clustering center vector from the feature set of all kinds of leaf samples.Finally,the cosine theorem was employed to get the blade leaf and the training sample similarity which were sorted.Tested results in the plants leaf set of Flavia show that：this method has a 96.03％ probability of target detection in the top five,and is better than KNN,BP neural network method,which makes this method,has some practical value.

关 键 词：叶片识别 图像处理 HU不变矩 形状特征 余弦定理 K均值聚类 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]