基于集成神经网络的植物叶片识别方法  

Plant leaf recognition based on artificial neural network ensemble

作　　者：恩德[1] 忽胜强 

机构地区：[1]河南理工大学电气工程与自动化学院,河南焦作454000

出　　处：《浙江农业学报》2015年第12期2225-2233,共9页Acta Agriculturae Zhejiangensis

基　　金：国家自然科学基金(41074090);河南省科技计划重点攻关项目(092102210360)

摘　　要：针对植物叶片识别中识别精度低的问题,从叶片特征描述、分类器设计两个角度出发提出了一种基于集成神经网络的植物叶片识别方法。叶片特征由区域几何特征与纹理特征共同构成,其中区域几何特征由不变矩特征和叶片几何描述参数共同构成,叶片纹理特征利用灰度共生矩阵进行提取。在分类器设计方面,采用一种集成神经网络学习算法,用于解决多类别植物叶片分类问题,其基分类器由二类别分类器和互补分类器构成。为避免叶片特征受到旋转等因素的影响,需要对叶片图像进行预处理。在预处理后,利用集成神经网络分类器对叶片样本进行训练与识别。在Flavia叶片数据库中选取20类叶片,每类30张共计600张叶片进行试验,基于集成神经网络的植物叶片识别方法的平均识别精度为91%。与其他叶片识别方法相比,试验结果表明,此方法可以提高叶片识别的精度。To improve the accuracy of the automatic plant identification system, this paper proposed a novel methodol- ogy of characterizing and recognizing plant /eaves using shape and texture features with neural network ensemble. Shape features of the leaves were captured using invariant moments together with geometric parameter of leaves. Tex- ture of the leaf was modeled using gray level co-occurrence matrix （ GLCM ）. And an artificial neural network ensem- ble in which the base classifier was composed by the union of a binary classifier and a multielass classifier was used for the resolution of multi-class problems. Since some of these features were in general sensitive to the orientation of the leaf images, a pre-processing stage prior to feature extraction was applied to make corrections for varying rotation. After the pre-processing stage, the neural network ensemble was used to train and classify the plant leaf samples. Ex- perimentations to demonstrate the efficacy of the proposed approach were performed on a dataset of 600 images divid- ed into 20 classes with 30 images per class, collected from Flavia, and the accuracy was 91%. In contrast with the other plant leaf recognition methods, the results showed that this method could significantly improve the accuracy of the system.

关 键 词：植物叶片识别 几何特征 纹理特征 集成神经网络 

分 类 号：S126[农业科学—农业基础科学]