Fig 1.
A block diagram showing the high-level workflow used in this study.
Fig 2.
The temperature and precipitation of the study area during the experiment period (15/12/2018- 31/05/2019).
Note that Temp. and Precip. denote the monthly average maximum temperature and total precipitation respectively.
Fig 3.
Study area map a) Field location in Australian Map. Note that the shapefile for this map is obtained from (https://www.diva-gis.org/gdata), b) RGB Orthomosaic of whole UAV trial c) Peanut field with 72 peanut breeding plots used in this study. Note that the black boxes layered over the images represent the Region of interest (ROI).
Fig 4.
UAV images acquisition date of peanut trial.
Table 1.
The list of vegetation indices used for LLS monitoring in this study.
Note that R, G, B, NIR and RE represent the spectral bands: Red (630-690nm), Blue (460-510nm), Green (545-575), Near-Infrared (820-860nm), and Red-edge (712-722nm) respectively.
Fig 5.
The RGB orthomosaic of the peanut field acquired at DAP-96 used in this study a), and its corresponding shape-file overlayed on the VI images b) DVI, c) SAVI, d) NDRE e) LAI.
Table 2.
The correlation coefficient between the LLS at DAP-112 and VI images taken in various peanut growth stages.
Note that a,b represents the correlation at 0.01 and 0.05 level of significance respectively.
Fig 6.
The sample plot distribution of three kinds of plots: Healthy, average, and highly diseased plots.
Note that the density plots are drawn by taking an average of five randomly chosen plots for each category of plots from the image taken at DAP-96. Note that the plots with disease scores (1-2), (3-5) and (6-9) are considered healthy, moderately diseased and severely diseased plots respectively for this illustration.
Fig 7.
The change in a coefficient of determination (R2) with the selected threshold value for three vegetation indices DVI, NDRE, and SAVI a) and LAI b) with UAV image taken at DAP-96.
Fig 8.
The normal distribution plot with a mean (μ) and standard deviation (σ).
Fig 9.
The convergence curve for the optimal value of the parameter (α) in MI-formula for six selected Vegetation Indices (VI).
Table 3.
The performance comparison of four methods (threshold-based, mean-based, CV-based and MI-based) on six selected vegetation indices for UAV-derived images taken before (DAP-96) manual LLS rating (y).
Note that the best thresholds chosen for NDRE (0.50), SAVI (0.7) and DVI (0.45), LAI (3.0), SR (25.0), REVI (0.22) for the threshold-based method (Ref. to Fig 7).
Table 4.
The performance comparison of four methods (Threshold-based, Mean-based, MI-based and CV-based) on selected six vegetation indices for UAV-derived image taken after (DAP-129) manual LLS rating.
Note that the best thresholds chosen for NDRE (0.45), SAVI (0.67) and DVI (0.45), LAI (2.6), SR (27.0), RDVI (0.55), REVI (0.24) for threshold-based method.
Fig 10.
The linear association between LLS rating and Average MI derived for vegetation index a) NDRE, b) SAVI c) DVI d) LAI e) SR and f) REVI. Note the MI is averaged based on the LLS manual rating measured on DAP-112.
Fig 11.
The linear association between LLS rating and Average CV derived for vegetation index a) NDRE, b) SAVI c) DVI d) LAI e) SR and f) REVI. Note the CV is averaged based on the LLS manual rating measured on DAP-112.
Fig 12.
The correlation plots of the relationship between vegetation index a) NDRE, b) SAVI c) DVI d) LAI e) SR and f) REVI derived LLS rating based on MI at DAP-96 and Manual rating measured on DAP-112.
Fig 13.
A high-level flowchart of proposed cooperative scheme for automatic disease estimation using UAV multispectral images.
Fig 14.
The visual mapping of late leaf spot (LLS) in a peanut field of 72 plots a) actual LLS score, and LLS score estimated with b) MI-based method (using REVI) c) CV-based method (using SR), e) mean-based method (using DVI).