The solution developed by Dário Oliveira to automatically classify agricultural crops using satellite images was recognized during the VIII FGV Research and Innovation Symposium
Dário received the recognition on November 25th | Photo: Reproduction
Some of the most promising applications of artificial intelligence are those that reduce human labor while increasing the accuracy of results. Excelling in these areas, the research of Dário Oliveira, a professor at the School of Applied Mathematics at Fundação Getulio Vargas (FGV EMAp), was honored at the VIII FGV Research and Innovation Symposium, held from November 25 to 27 at the FGV Cultural Center.
His postdoctoral work, titled “Learning Crop Type Mapping from Regional Label Proportions in Large-Scale SAR and Optical Imagery”, was conducted between 2021 and 2022, published in 2023, and developed a machine learning system that uses satellite images to automate agricultural crop classification. The study, carried out at the University of Wisconsin-Madison in the United States, is based on the growth cycle of crops and historical data, such as the number of seeds planted and sold by region, crop types, and cultivation data from previous years.
LLP-Co process (AI tool) for mapping crop types through clusters in satellite images | Photo: Reproduction
By gathering this information, estimated by the Brazilian Institute of Geography and Statistics (IBGE), the system builds a database on the context in which a given crop is located, providing cultivation proportions. Based on this data, it can categorize satellite images, automatically clustering information, reducing the need for human labor, and achieving high accuracy in findings.
The initial experimental results are promising, showing 90% accuracy for major crop types such as corn, soybeans, and cotton. In countries with continental dimensions like Brazil, this classification is essential for monitoring cultivated areas, estimating harvests, planning agricultural policies, providing financing, calculating rural insurance, mitigating the effects of climate change and natural disasters, and managing resources efficiently. “For Brazil, one of the world’s largest agricultural exporters, this technology can enhance agricultural management, helping to ensure economic competitiveness and environmental sustainability,” says the professor.
An Electrical Engineer from the Rio de Janeiro State University (UERJ), with a specialization in Computing and Systems, Dário Oliveira believes that scaling up data analysis can greatly benefit agricultural properties in Brazil that rely on continuous monitoring. “The study focuses on significant agricultural regions in Brazil, such as Campo Verde (Mato Grosso) and Luís Eduardo Magalhães (Bahia), which feature dominant crops like soybeans, corn, and cotton. These crops are economically vital and depend on continuous monitoring to predict yields, identify seasonal changes, and optimize the use of inputs,” explains Dário.
The system developed by Dário Oliveira recognizes crop patterns through satellite images, taking into account various characteristics such as colors | Photo: Tom Fisk - Pexels
Multidisciplinary research
One of the key elements of this idea is the integration of knowledge from various fields of study. Here, deep learning knowledge is combined with biological expertise, aligning with the mission of the FGV Symposium on Research and Innovation to promote the development of multidisciplinary research projects with a high impact on the country's socioeconomic development. “By combining contrastive learning with data proportions, the method reduces costs and delivers high accuracy in complex tropical regions,” says the professor from FGV EMAp.
Regarding the award at the Symposium on Research and Innovation, the goal is to structure networks of researchers from Brazil and abroad to expand the potential intersection of this multidisciplinary knowledge. Dário Oliveira acknowledges that “the social, economic, and environmental impact of the methodology in agricultural monitoring and risk management was a distinguishing factor recognized by the jury.” For him, efficient land use could even reduce the expansion of agriculture in preservation areas, and image monitoring could be used for actions against illegal deforestation.
Among the most promising benefits for research is ensuring food security through more efficient agriculture | Photo: Tom Fisk - Pexels
In this year's edition of the event, topics related to the 5th National Conference on Science, Technology, and Innovation and the G20 Brazil 2024 meetings were discussed. Thus, considering the benefits that support everyone from producers to the country's economy and politics, the positive impact on the population is undeniable. According to the professor, the results are essential to ensuring global food security.