Detection of rubber tree powdery mildew based on UAV multispectral image and stacked machine learning model
Rubber trees are a significant tropical cash crop in China that are vulnerable to powdery mildew.Monitoring this disease accurately and promptly is crucial in preventing widespread infestations.Recently,UAV remote sensing technology has been increasingly utilized in agriculture and forestry.This study aimed to evaluate the potential of low-altitude remote sensing technology in detecting powdery mildew in rubber trees on a large scale and enhancing detection accuracy.The process involved acquiring multispectral images of rubber tree canopies using a DJI Phantom 4 multispectral drone,calculating vegetation index(VI)and texture features(TF),conducting correlation and feature importance analyses with Pearson correlation coefficient(PCCs)and Boruta-SHAP algorithms to eliminate redundant features,and selecting Blue-MEA,WI,DVIRE,PPR,and GI as the optimal feature combinations.Subsequently,a monitoring model for rubber tree powdery mildew was developed using K-nearest neighbor(KNN),naive Bayes,support vector machine(SVM),random forest(RF),extreme gradient boosting(XGB),and Stacking integration algorithms.The results indicated that after feature selection,the Stacking integrated model achieved an accuracy(OA)and Kappa(KC)value of 96.39%and 92.78%,respectively.The integrated learning model showed a 3.15%,5.52%and 1.80%improvement in accuracy compared to the individual base models(KNN,Bayes,SVM,RF,and XGB),with Kappa values increasing by 6.32%,11.05%,3.61%,6.09%and 2.27%.Notably,the highest classification accuracy(OA=96.22%)was attained using a pixel aggregation strategy with a 17×17 window size when mapping the spatial distribution of powdery mildew in rubber trees.
powdery mildew of rubber treeunmannedaerial vehicle(UAV)multispectral imageremotesensingstacked ensemble
NETL
NSTL
万方数据
