基于图像识别技术的豆科牧草分类研究  

Study on Classification for Leguminous Forage Based on Image Recognition Technology

作　　者：王敬轩[1] 冯全[1] 王宇通[2] 邵新庆[2] 

机构地区：[1]甘肃农业大学工学院,甘肃兰州730070 [2]中国农业大学动物科技学院,北京100193

出　　处：《草地学报》2010年第1期37-41,共5页Acta Agrestia Sinica

基　　金：国家科技支撑计划项目(2007BAD52B06-2);京承路都市型现代农业走廊工程科技示范项目(D08060500460803)资助

摘　　要：利用计算机图像处理技术,依据植物叶片图像的形状特征对14种豆科牧草进行分类识别。通过对叶片图像进行预处理,提取出叶片的轮廓。在此基础上提取了叶片形状的全局特征和局部特征;全局特征包括叶片的横纵轴比、矩形度、圆形度等8项几何特征和7个图像不变矩特征;局部特征为叶缘粗糙度。利用PNN(Probabilisticneural network)和BPN(Back propagation network)作为分类器进行识别分类,实现了对豆科牧草叶片图像的分类。识别结果表明,PNN网络的平均识别率为85.1%、BPN网络的平均识别率为82.4%。Traditionally,measure and species classification of plants are implemented by human experts,which is time-consuming and inefficient.In recent years,information technology including image processing and pattern recognition has been introduced into plant classification.Compared with flowers with 3D structures,leaves are easier to process by computer due to their 2D structures.This paper introduces a method of classifying plants of leguminous forage based on the leave shape features.Firstly,pre-processing method is used to extract the contour of a leaf.Then global and local features of the leaf shape are extracted.The global features include eight geometric features such as axis ratio,rectangularity,circularity,etc,and seven moment invariants.Roughness of leaf edge is selected as the local features.Finally,probabilistic neural network（PNN） and back propagation network（BPN） are applied to constructing classifiers.The experimental results show that the recognition rate of PNN and BPN is 85.1% and 82.4% respectively.

关 键 词：豆科牧草 叶片识别 图像处理 PNN BPN 

分 类 号：S126[农业科学—农业基础科学] S541