High-throughput phenotyping via UAS: the optimization within a breeding program and a new validation method based on simulation

• Ano de defesa: 2020
• Autor(a) principal: Galli, Giovanni
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Aerial imagery; Fenômica; Imageamento aéreo; In silico; Phenomics; Sorghum; Sorgo; Structure-from-motion
• Link de acesso: http://www.teses.usp.br/teses/disponiveis/11/11137/tde-21052020-121330/

High-throughput phenotyping (HTP), or simply phenomics, has drawn the attention of the scientific community as a field with the potential to increase phenotyping cost-effectiveness and accuracy. Nevertheless, the feasibility of this set of approaches is yet to be confirmed. In this sense, two major challenges to its application are optimizing the data-to-decision process and the validation of procedures and pipelines for specific selection scenarios. We add to this matter by reporting on two studies aimed at the optimization and validation of field HTP based on unmanned aerial systems (UAS). In the first, we presented a proof-of-concept investigation using a grain sorghum dataset with the intent of identifying when HTP data should be collected and how it should be processed for the optimization of prediction of two major traits, grain yield and plant health. Our findings suggest that there is no predictive ability increase when combining multiple vegetation indices and flight dates. Additionally, a single index and flight can be used for predicting both traits without expressive accuracy loss. In the second, we presented a new tool for validating aerial image-based HTP approaches with computer simulations. The approach was exemplified with a comprehensive study case of plant height estimation in maize. Our results show that the in silico experiments could be adequately reconstructed with structure-from-motion algorithms using UAS-like rendered images, enabling inference-making about tested factors. This study also brought new insights into the effect of experimental factors over the accuracy of plant height assessment using HTP. At last, we believe that our findings allowed the promotion of a deeper understanding of the HTP practice, enabling breeders to work towards a more reliable and cost-effective selection.