AUTOMATED PHENOTYPING OF REPRODUCTIVE STRUCTURES IN SOYBEAN AND THE USE OF MELATONIN: APPLICATION OF MACHINE LEARNING IN THE DETECTION OF FLORAL ABORTION UNDER ENVIRONMENTAL STRESS.

• Grant number: 25/03937-0
• Support Opportunities: Scholarships abroad - Research
• Start date: June 27, 2025
• End date: April 26, 2026
• Field of knowledge: Agronomical Sciences - Agronomy - Crop Science
• Principal Investigator: Adriana Lima Moro
• Grantee: Adriana Lima Moro
• Host Investigator: Krishna Jagadish
• Host Institution: Pró-Reitoria de Pesquisa e Pós-Graduação. Universidade do Oeste Paulista (UNOESTE). Presidente Prudente , SP, Brazil
• Institution abroad: Texas Tech University (TTU), United States

Abstract

Population growth and climate change significantly challenge agricultural production, especially in rainfed systems such as soybean cultivation. Heat and water stress are critical factors that affect crop yield, leading to reduced photosynthetic rate, hormonal imbalance, and floral abortion, which directly impact grain yield. The phenotyping of reproductive structures, such as flowers and pods, is essential to understanding the impacts of adverse environmental conditions on soybean yield. However, accurate quantification of these losses in the field is still a challenge, due to the limitation of traditional methods based on manual observation. This project proposes the use of machine learning techniques applied to image-based phenotyping to identify and quantify abortion of reproductive structures in different soybean genotypes with and without abortion tendencies. It will be developed at Texas Tech University (TTU) with the team of Dr Krishna Jagadish, renowned for the study of stress physiology and the development of a phenotyping platform. An automated platform, using an RC-car equipped with high-resolution sensors and cameras, will be employed to capture images of plants in the reproductive stage. Advanced algorithms and object detection models (YOLO and Faster R-CNN), will be trained to recognize and analyze flowers and pods, allowing the estimation of the abort rate. In addition, the research will evaluate the effect of the application of Melatonin on floral retention and its impact on physiological and biochemical metabolisms. The objective is to validate an automated phenotyping protocol and the creation of a machine-learning model for the detection of flowers and pods and the understanding of the role of Melatonin in reducing miscarriage. With this, the project will contribute to the advancement of digital agriculture in Brazil with technology transfer and the development of strategies to increase the resilience of soy in the face of climate change.