Early detection of bacterial wilt in peanut plants through leaf-level hyperspectral and unmanned aerial vehicle data

Abstract Bacterial wilt (BW) caused by Ralstonia solanacearum is the most serious peanut diseases in South China. Its timely and accurate detection is important to opportunely implement disease management practices. This study aimed to establish and select the most appropriate leaf-level reflectance-based vegetation indices for BW detection and to determine whether these new indices can be used in UAV multispectral imaging for peanut BW detection. ANOVA, multilayer perception, and the reduced sampling method were used to analyze the spectral data. The most effective detection wavelengths, 730 nm and 790 nm, were used for developing new peanut BW detection indices. The 15 hyperspectral indices with highest correlation coefficients (R > 0.80) were obtained based on 46 hyperspectral indices and the BW severity results from Experiment 1. By testing the above vegetation indices at the leaf level and in UAV images using different methods and the results from Experiment 2, it was found that four of the developed indices (BWI1, BWI3, BWI4, and BWI6) performed appropriately (P   1.0), as they could distinguish between healthy and BW infected peanut plants, even if the plant presented minimal external symptoms. Our findings confirmed the potential of hyperspectral remote sensing including leaf-level and UAV images for peanut BW detection at early disease stages and discrimination of different BW severity levels based on vegetation indices derived from leaf-level reflectance. Timely BW severity determination based on our results could provide farmers with useful information to control peanut BW disease.