Machine Learning Approaches to High Throughput Phenotyping

Abstract

Image processing, extraction of appropriate data classifiers, and machine learning algorithms are key steps in plant phenotyping that connects genomics with plant ecophysiology and agronomy. Based on a dataset of labeled images from Populus Trichocarpa genotypes cultivated under both drought and control conditions, we are able to extract potential data classifiers such as leaf color and edge morphology, and to develop a predictive model by using PlantCV. The use of Tesseract and OpenCV has not reached the needed successes that are required for a proper workflow, such as data preparation, training the module, tuning parameters, and others. Despite many existing challenges, progresses reported here gives possible future directions can mitigate these challenges.