High-throughput phenotyping of plant health using machine learning and computer vision
Phenotyping is used to develop new strains of plants, understand plant-affecting diseases (phyto-pathology) and evaluate the effects of various substances on plants. A growing variety of sensors and sensor technology is used to gather data used for phenotyping, in a non-destructive manner, and this overall process of data acquisition and analysis is being automated, leading to high-throughput pheno-typing. These technological changes pose challenges both in terms of which models to apply to these heterogeneous data, as well as the scalability of the data and analytics pipeline. The proposed project aims to develop a deep learning model that would produce estimates of plant health based on imaging and other sensor data generated by Terramera, with high classification accuracy, according to threshold tolerances defined by Terramera plant physiologists. The project would also explore some of the data management challenges arising from the application of those deep learning models on automated phe-notyping data acquisition at scale.