Analysis of High-Throughput Plant Phenotyping: From Images to Numbers to Statistical Analysis

High-throughput phenotyping system has become more and more popular in plant science research. The data analysis for such a system typically involves two steps: plant feature extraction through image processing and statistical analysis for the extracted features. In this dissertation, a pipeline for both of the two steps consisting of robust feature extraction and functional data analysis was constructed. First, for image processing, two RGB image processing procedures based on Double Criterion Thresholding (DCT) methods and Hidden Markov Random Field-EM Framework (HMRF-EM) were respectively included. Second, for statistical analysis, Functional Analysis of Variance (ANOVA) and the corresponding statistical inference were conducted using the extracted traits from the image processing part. An R package “implant” was developed for implementing this pipeline. In addition, a novel Spline-based procedure to solve the high-dimensionality issue of organ segmentation on hyperspectral images was conducted in this dissertation. The Spline-based procedure is designed by fitting a Penalized Spline Regression to the pixel intensities across all wavebands for each pixel point and reducing the original data dimension using the estimated coefficients. Supervised classification models are then applied to the reduced data for organ segmentation. The Spline-based procedure can reduce the dimension without sacrificing the overall classification accuracy, improving the background noise elimination effect, and recovering the original hyperspectral images effectively