基于多特征与多分类器的鲜茶叶分类研究  

Research on classification of fresh tea based on multiple features and multiple classifiers

作　　者：毛腾跃[1,2] 黄印 文晓国 帖军[1,2] Mao Tengyue;Huang Yin;Wen Xiaoguo;Tie Jun(College of Computer Science,South-Central University for Nationalities,Wuhan,430074,China;Hubei Provincial Engineering Research Center for Intelligent Management of Manufacturing Enterprises,Wuhan,430074,China;Agricultural Information Technology Research and Development Laboratory of Chinese Academy of Sciences and State Ethnic Affairs Commission,Wuhan,430070,China)

机构地区：[1]中南民族大学计算机科学学院,武汉市430074 [2]湖北省制造企业智能管理工程技术研究中心,武汉市430074 [3]中国科学院—国家民委农业信息技术研究与开发联合实验室,武汉市430070

出　　处：《中国农机化学报》2020年第12期75-83,共9页Journal of Chinese Agricultural Mechanization

基　　金：湖北省技术创新专项重大项目(2019ABA101);湖北省科技计划项目(2019CFC890);中国科学院—国家民委农业信息技术研究与开发联合实验室开放基金;校级质量工程项目(JYZD19045)。

摘　　要：机采茶中混有不同类型的鲜叶,传统的风选、振动筛选等分类方式准确度低,现有的基于计算机视觉的分选方式也无法满足对常见的单芽、一芽一叶、一芽二叶、一芽三叶四种类型鲜叶的准确分类。为解决茶叶机采后各类型鲜叶精确分类问题,提出了一种基于多特征与多分类器的鲜茶叶分类方法。首先,利用鲜叶的相对几何特征与纹理特征基于SVM构建鲜茶叶分类器;然后,对多边形拟合后的鲜叶图像进行特殊角点检测得到各特殊角点数量对应的各类别分类概率,并将特殊角点序列的距离矩阵相似度作为判断依据;最后利用KNN对上述两种方法的结果进行融合,得到最终分类结果。试验结果表明,该方法可以更好的利用不同类别鲜叶的特征进行分类,分类准确率达94.24%,取得了较好的分类效果。The tea leaves picked by the machine are mixed with different categories of fresh leaves,and the accuracy of traditional classification methods such as air sorting and vibration screening is low.The existing classification methods based on computer vision also cannot satisfy the precise classification among the four categories of fresh leaves which are always characterized by single bud,one bud with one leaf,one bud with two leaves and one bud with three leaves.In order to solve the problem,a new fresh tea classification method based on multiple features and multiple classifiers is presented.First,the relative geometric features and texture features of fresh leaves are used to construct a fresh tea classifier based on SVM.Then,perform special corner point detection on the polygon-fitted image to obtain the classification probability of each category corresponding to the number of special corner point and adopt the distance matrix similarity of special corner point sequences as the judgment basis.Finally,combine the results of the above two methods via KNN and get the final classification result.The experimental results show that this method can better use the features of different categories of fresh leaves for classification,and the accuracy can reach 94.24%,achieving a good classification effect.

关 键 词：鲜茶叶分类 特征 支持向量机 K最近邻算法 距离矩阵 结果融合 

分 类 号：S24[农业科学—农业电气化与自动化] TP391.4[农业科学—农业工程]