Wheat Fusarium head blight monitoring based on Bayesian optimization machine learning with the fusion of RGB and multispectral sensors from UAV

Fusarium Head Blight (FHB) is a major disease caused by Fusarium graminearum, poses a significant threat to yield and quality of wheat production. Timely and accurately monitoring of wheat FHB is essential for effective management and loss assessment. In this study, unmanned aerial vehicle (UAV) equipped with RGB digital and multispectral cameras were employed to capture remote sensing images of wheat fields, and the color space of L*a*b* and HSV from RGB images, as well as vegetation indices (VIs) and texture features (TFs) from multispectral images were integrated. First, the Lasso-MGRS method was utilized to screen and reduce the dimensionality of different types of features. Then, support vector machine (SVM) were applied to estimate the severity of wheat FHB using both single and integrated features. Finally, Bayesian optimization (BO) was used to fine-tune the parameters of the SVM, random forest (RF) and extreme gradient boosting (XGBoost) to improve the models’ accuracy. The results indicated that the combination with different types of features showed a better estimation performance with the overall accuracy (OA) and kappa coefficient were 0.7703 and 0.7123 respectively. The models with BO achieved the highest accuracy with the OA and kappa coefficient were 0.8649 and 0.8309 respectively.