Sistema de visão e inteligência computacional em ambiente de nuvem para gestão de risco da ferrugem asiática na cultura da soja

• Ano de defesa: 2024
• Autor(a) principal: Neves, Ricardo Alexandre
• Orientador(a): Cruvinel, Paulo Estevão
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de São Carlos; Câmpus São Carlos
• Programa de Pós-Graduação: Programa de Pós-Graduação em Ciência da Computação - PPGCC
• Palavras-chave em Português: Processamento de imagens; Aprendizado de máquina; Suporte à decisão; Ferrugem asiática da soja
• Palavras-chave em Inglês: Image processing; Machine learning; Decision support; Asian soybean rust
• Área do conhecimento CNPq: CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO::METODOLOGIA E TECNICAS DA COMPUTACAO
• Link de acesso: https://repositorio.ufscar.br/handle/20.500.14289/20311

Controlling Asian Soybean Rust (Phakopsora pachyrhizi) in soybeans (Glycine max (L.) Merril) often requires high fungicide use, which can lead to resistance. Thus, new control solutions are needed for mitigation. This work presents an intelligent computer vision system for assessing the presence and severity of this disease in crop areas. It involves pattern recognition and machine learning techniques, enabling diagnostic actions for prognosis and control. It considers a decision-support model using random variables related to climate, plants, and characteristics recognized in digital images of monitored soybean leaves. For feature extraction, it uses scale-invariant feature transform, histogram of oriented gradients, and Hu’s invariant moments techniques. It uses cloud-based computational infrastructure and intelligent network processing, as well as principal component analysis for dimensionality reduction of features classified by support vector machines. Additionally, a hidden Markov model is used to fuse random variables, offering robustness, effectiveness, and efficiency, as confirmed by expert cross-correlation. To evaluate data quality at various system stages, metric sets such as peak signal-to-noise ratio, mean squared error, structural similarity index, missing values, accuracy, precision, F1-score, and recall are considered. This solution prevents and reduces fungicide use, enhancing production and guiding future early spatio-temporal monitoring of the disease on an agricultural scale.