Apple production in Washington is a leading agricultural industry with 70% of total U.S. production. Development of new cultivars adapted to local climatic conditions, and producing cultivars resistant to diseases with improved fruit quality, texture and outstanding storability are the major goals of breeding programs. Given the importance of plant phenotyping in apple breeding program, the overall goal of this study was to investigate high-throughput phenotyping techniques to evaluate field (stomatal conductance, fire blight disease resistance), and post-harvest (bitter pit) traits. In regard to field traits, hyperspectral sensing (visible-near-infrared spectral reflectance) and imaging technique (RGB, multispectral, and thermal) were evaluated to estimate the stomatal conductance and fire blight resistance in young apple trees. The results demonstrate a reliable means of profiling tree responses to abscisic acid (ABA) effects and disease. In ABA study, ABA was applied to young trees and responses were captured using sensing systems at 1 and 3 days after treatment. The maximum average accuracy ranged between 80-85% at 3 days after treatment, when control and ABA-treated trees were classified. In fire blight disease rating evaluation study, normalized difference spectral indices (NDSIs) were created from visible-near-infrared reflectance spectra with high correlations. The new NDSIs were combinations of 1275-889, 1286-879, and 1320-562 nm.

In regard to bitter pit detection, hyperspectral reflectance spectra (350-2500 nm) and imaging (550-1700 nm) were used. An algorithm was developed to automatically detect and discriminate bitter pit-affected apples. As a result, the partial least square regression (PLSR) and stepwise discriminant analysis-based selected features classified the dataset with accuracies of about 80-93%. Regression analysis indicated a strong relationship between the PLSR-based spectral features and magnesium-to-calcium ratio in fruit peel. In hyperspectral images, the spatial data were used to extract the pit area on each apple incorporating the selected wavelength from reflectance data. The best discriminating wavelengths of hyperspectral images were added to the analysis that could classify apples at an accuracy of 85%. The final reflectance wavelengths (665, 731, 797, 1217, 1283, 1349, and 1410 nm) selected can be used towards developing a custom sensing system for storage facilities.