摘要
根据锈病、灰斑病、小斑病、褐斑病和弯孢菌叶斑病等5种主要玉米叶部病害的特点,提出了一种基于图像处理技术和概率神经网络技术的玉米叶部病害识别方法。首先,对田间采集的玉米叶部病害图像样本进行去噪处理、图像分割和特征提取;然后,利用遗传算法优化选择出4个独立、稳定性好、分类能力强的分类特征;最后,提取目标对象的特征向量作为输入向量,由概率神经网络(PNN)分类器识别病害类别,平均正确识别率为9 0.4%,高于BP神经网络。试验结果表明了该方法的有效性,可为田间作物病虫害的快速智能诊断提供借鉴。
In view of five kinds of actual maize disease image characteristics,a method of disease identification which used the technique of image processing and probabilistic neural network was proposed.First the maize leaf disease pictures of different varieties were taken in fields;the images were deniosed and segmented;the color and the shape features were extracted.Then genetic algorithm was used to get four approximate features.Finally the probabilistic neural network classifier was created for recognition of maize leaf disease according to the features,and the average precision of five kinds of maize disease identifying was 90.4%.The efficiency was shown according to the experimental results,and it can provide a technical support for the automatic recognition of crop diseases and insets with disease image obtained in fields.
出处
《农机化研究》
北大核心
2011年第6期145-148,共4页
Journal of Agricultural Mechanization Research
基金
国家"十一五"粮食丰产科技工程项目(2006BAD01A08)
关键词
玉米
叶部病害
特征提取
遗传算法
概率神经网络
maize
leaf disease
feature extraction
genetic algorithm
probabilistic neural network
