Remote Sensing Based Monitoring of Winter Wheat Powdery Mildew at a Regional Scale Using Random Forest Model

6th Annual 2018 International Conference on Geo-Spatial Knowledge and Intelligence

Huang, Linsheng^1 ; Ding, Wenjuan^1 ; Jiang, Jing^1 ; Wu, Zhaochuan^1 ; Zhao, Jinling^1

National Engineering Research Center for Analysis and Application of Agro-Ecological Big Data, Anhui University, Hefei

230601, China^1

关键词: Inter-band correlation;    Operational land imager;    Overall accuracies;    Random forest modeling;    Real time performance;    Remote sensing data;    Satellite remote sensing;    Spatial resolution;   

Others  :  https://iopscience.iop.org/article/10.1088/1755-1315/234/1/012015/pdf
DOI  :  10.1088/1755-1315/234/1/012015

来源: IOP

【 摘 要 】

Wheat powdery mildew is one of the main diseases of wheat. The emergence and spread have seriously affected the yield and quality of wheat. It is of great significance to accurately monitor the occurrence of such a plant disease at a regional scale. Satellite remote sensing imagery has been extensively utilized in monitoring and assessing various plant diseases, due to its own particular advantages of real-time performance, wide coverage, high tempo-spatial resolution, etc. It is therefore suitable as a data source for monitoring the occurrence of wheat powdery mildew. Gaocheng District, Zhao County and Jinzhou City of Hebei Province, China were used as the study area. A total of eleven features were extracted from Landsat 8 OLI (Operational Land Imager) remote sensing data. Due to the existence of inter-band correlation, the Relief-F algorithm was used to carry out the feature selection. According to the weight value of each feature, three features were determined as the input into the created monitoring model. Consequently, the monitoring model was respectively established using the random forest (RF) and support vector machine (SVM). The analysis results show that the overall accuracy (OA) of RF model reaches 84% with the Kappa coefficient of 0.67, which are better than the SVM model of 73% and 0.47.