| dc.description.abstract | Precise field phenotyping has always been a bottleneck in wheat breeding. Traditionally, field phenotyping has been done by physically inspecting plots one by one. A substantial amount of time, cost, and labor is required to collect data from many breeding lines. High-throughput phenotyping (HTP) is gaining interest in recent years. Advancement in Unmanned Aerial System (UAS) and sensor technology enabled the collection of high spatial and temporal resolution data which can be used in agricultural research and management. This study was conducted to develop and assess the use of UAS in wheat breeding. The major objectives of this study were to define UAS data collection and processing framework and to investigate the application of UAS data to assess disease, seasonal growth, and yield in wheat (Triticum aestivum L.).
yieldThe experiment to investigate the application of UAS to assess disease severity was conducted in 2017 and 2018 at Castroville, Texas. RGB images were acquired by flying rotary wing UAS. Images were then processed to develop orthomosaics and three vegetation indices were calculated. Visual notes on field response and leaf rust severity were taken to calculate Coefficient of Infection (CI). A significant variation in vegetation indices was found among the wheat genotypes. Normalized Difference Index (NDI), Green Index (GI), and Green Leaf Index (GLI) were linearly related to CI with coefficient of determination (R^2 ) values of 0.78 (<p0.05), 0.75 (p<0.05) and 0.72 (p<0.05), respectively. Vegetation indices used in this study showed great potential for their use in leaf rust severity assessment in wheat.
Crop growth analysis was performed to investigate the application of UAS data to assess wheat seasonal growth. This field study was conducted in 2018-2019 winter wheat growing season and RGB-based multi-temporal UAS data were collected throughout the growing season. Canopy Cover (CC) was obtained from UAS images and used to perform growth analysis by fitting several growth functions. Four-parameter logistic growth function had the best fit with R^2 value of 0.99 (p<0.05) and lowest Root Mean Square Error (3.43). Grain yield was positively associated with CC obtained during reproductive stage of wheat (R^2 =0.65, p<0.05) suggesting the importance of maintaining healthy canopy during grain filling for better yield.
The relationship between UAS obtained canopy features and vegetation indices with grain yield was analyzed to develop a wheat yield prediction model. Eight vegetation indices and two canopy features (CC, canopy height) were extracted from multispectral and RGB imagery. UAS parameters obtained during grain filling stage of wheat were significantly related to grain yield (R^2 >0.30, p<0.05). A three-layered Artificial Neural Network (ANN) model was created using multi-temporal CC, canopy height, Excess Green Index (ExG), Normalized Difference Red Edge Index (NDRE), Normalized Difference Vegetation Index (NDVI), standard deviation of NDRE and ExG measurements as input parameters to predict grain yield. R^2 values between predicted yield and observed yield were 0.78 and 0.60 (p<0.01) for training and testing data set, respectively. Satisfactory performance of ANN model shows the potential of using machine learning models to predict grain yield based on UAS parameters. | en |
