基于动态集成的黄瓜叶部病害识别方法

发布时间：2018-08-15 18:14

【摘要】：对作物病害类型的准确识别是病害防治的前提。为提高病害识别的准确度,以黄瓜叶部病害识别为例,提出一种基于动态集成的作物叶部病害种类的识别方法。首先利用图像分块策略提取病害图像的75维颜色统计特征,然后采用不一致度量方法对构建的10个BP神经网络单分类器进行差异性度量,并按照差异性大小进行排序,最后根据分类器的可信度,动态选择差异性大的分类器子集对病害图像进行集成识别。在由512幅白粉病、霜霉病、灰霉病和正常叶片4类黄瓜叶片组织图像构成的测试集上,所提方法的识别错误率为3.32%,分别比BP神经网络、SVM、Bagging、Ada Boost算法降低了1.37个百分点、1.56个百分点、1.76个百分点、0.78个百分点。试验结果表明:所提方法能够实现黄瓜叶部病害种类的准确识别,可为其它作物病害的识别提供借鉴。
[Abstract]:Accurate identification of crop disease types is the premise of disease prevention and control. In order to improve the accuracy of disease identification, a method based on dynamic integration was proposed to identify cucumber leaf diseases. Firstly, the 75 dimensional color statistical feature of disease image is extracted by image partitioning strategy, and then the 10 BP neural network single classifiers are measured by using inconsistent measure method, and the difference is sorted according to the size of the difference. Finally, according to the reliability of the classifier, the subsets of the classifier with large difference are dynamically selected for the integrated recognition of the disease image. On a test set of 512 leaf tissue images of powdery mildew, downy mildew, gray mildew and normal leaves of four types of cucumber, The recognition error rate of the proposed method is 3.32, which is 1.37% or 1.56% or 1.76% or 0.78% lower than the BP neural network SVMBaggingAda Boost algorithm respectively. The experimental results showed that the proposed method could be used for accurate identification of cucumber leaf diseases and could be used as a reference for the identification of other crop diseases.
【作者单位】：北京农业信息技术研究中心;农业部农业信息技术重点实验室;北京工业大学信息学部;
【基金】：国家自然科学基金项目(61403035、71301011) 北京市自然科学基金项目(9152009)
【分类号】：S436.421;TP391.41

【相似文献】