机器视觉和AdaBoost的柑桔溃疡病自动检测  

Citrus Canker Automatic Detection Based on Computer Vision and AdaBoost Algorithm

作　　者：朱庆生[1] 张敏[1] 杨方云[2] 柳锋[1] 

机构地区：[1]重庆大学计算机学院,重庆400030 [2]中国农业科学院柑桔研究所,重庆400700

出　　处：《计算机仿真》2008年第7期237-239,285,共4页Computer Simulation

基　　金：高等学校博士学科点专项科研基金(20050611027);重庆市自然科学基金资助(CSTC2006BB1347)

摘　　要：将计算机视觉技术应用于柑桔病害识别问题,实现了快速准确的识别柑桔溃疡病。从特征构造,特征选择和识别系统设计三方面进行了研究。在特征构造上采用了Gabor变换,边缘识别等方法得到了包括颜色、纹理及形状的综合特征;在特征选择上采用了AdaBoost算法实现;最后通过AdaBoost学习方法构造分类器并利用滑动窗口技术进行病害区域检测。实验结果证明该方法对柑桔溃疡病其识别准确率高于95%,在训练轮数较多的情况下能够接近99%的识别率,且该识别率较稳定。实验结果显示计算机自动识别效果与专家目测相当,在生产中具有一定的实用价值。Computer vision technology is introduced into fast and accurate automatic detection of citrus canker.Three key issues are discussed in this paper,which are feature construction,feature selection and system design.To construct features,Gabor transformation and edge detection algorithms are used and a feature set of color,texture and shape is obtained.Then AdaBoost algorithm is applied to select the most efficient features from the feature set.AdaBoost algorithm is also used for training classifier.To fast detect citrus disease area on image,a moving window technology is used.The experiment results show that the detection accurate rate can reach over 95%,and when the training round reachs a certain amount the detection rate can reach 99%,which is about equal to the detection accurate rate by experts eyeballing,and this approach is valuable for plant production.

关 键 词：特征构造 特征选择 分类器 识别率 

分 类 号：S126[农业科学—农业基础科学]