基于AdaBoost模型和mRMR算法的小麦白粉病遥感监测

发布时间：2019-03-19 18:09

【摘要】：除选择合适的建模方法外,选择合适的特征选择算法来优选建模特征对提高作物病害的遥感监测水平具有重要作用。选取陕西省关中平原西部小麦白粉病为对象,基于Landsat 8遥感影像共提取了18个特征变量,通过相关性分析(correlation analysis,CA)和最小冗余最大相关(minimum redundancy maximum relevance,mRMR)2种特征选择算法筛选出了2组不同的特征变量,分别将其输入Fisher线性判别分析(Fisher linear discriminant analysis,FLDA)、支持向量机(support vector machine,SVM)和AdaBoost 3种方法,构建小麦白粉病发生严重程度监测模型,并对其进行精度验证与对比分析。结果表明,2种AdaBoost模型对小麦白粉病发生严重程度的总体监测精度分别比FLDA模型和SVM模型高出27.9%、27.9%和14.0%、9.3%,mRMR算法筛选特征所建FLDA、SVM及AdaBoost监测模型的总体监测精度分别比CA筛选特征所建模型高出7.0%、11.7%和7.0%,且mRMR算法筛选特征结合AdaBoost方法所建监测模型的精度和Kappa系数分别为88.4%和0.807,为所有模型中最高。说明将AdaBoost方法用于作物病害遥感监测效果较好,在作物病害监测模型的特征变量选择中mRMR算法比常用CA算法更具优势。研究结果可为其他作物病害遥感监测提供方法参考。
[Abstract]:In addition to selecting the appropriate modeling method, selecting the appropriate feature selection algorithm to optimize the modeling features plays an important role in improving the monitoring level of crop diseases by remote sensing. Taking wheat powdery mildew in the western part of Guanzhong Plain of Shaanxi Province as an object, 18 feature variables were extracted based on Landsat 8 remote sensing image. The correlation analysis (correlation analysis,CA) and the minimum redundancy maximum correlation (minimum redundancy maximum relevance, were used to extract 18 feature variables. MRMR) two feature selection algorithms select two groups of different feature variables, and input them into Fisher linear discriminant analysis (Fisher linear discriminant analysis,FLDA), support vector machine (support vector machine,SVM) and AdaBoost, respectively, and three methods, I. E. Fisher linear discriminant analysis (Fisher linear discriminant analysis,FLDA), support vector machine (support vector machine,SVM) and AdaBoost. The wheat powdery mildew severity monitoring model was constructed, and its accuracy verification and comparative analysis were carried out. The results showed that the overall monitoring accuracy of the two AdaBoost models for the severity of wheat powdery mildew was 27.9%, 27.9% and 14.0%, 9.3% higher than that of the FLDA model and the SVM model, respectively. The FLDA, was screened by the MRMR algorithm. The overall monitoring accuracy of SVM and AdaBoost monitoring model is 7.0%, 11.7% and 7.0% higher than that of CA screening model, respectively. The precision and Kappa coefficient of the monitoring model based on mRMR algorithm and AdaBoost method are 88.4% and 0.807 respectively, which is the highest among all the models. The results show that the AdaBoost method has a good effect on crop disease remote sensing monitoring, and the mRMR algorithm has more advantages than the common CA algorithm in the selection of characteristic variables of crop disease monitoring model. The results can be used as a reference for remote sensing monitoring of other crop diseases.
【作者单位】：南京信息工程大学应用气象学院气象灾害预报预警与评估协同创新中心;中国科学院遥感与数字地球研究所数字地球重点实验室;杭州电子科技大学生命信息与仪器工程学院;安徽大学电子信息工程学院;
【基金】：中国科学院国际合作局对外合作重点项目(131211KYSB20150034) 国家重点研发计划项目(2016YFD030702) 国家自然科学基金国际合作项目(61661136004);国家自然科学基金项目(41271412、41601467) 江苏省普通高校自然科学研究资助项目(15KJA170003)
【分类号】：S435.121.46;S127

【相似文献】