Análise, classificação e detecção automática de nódulos em raízes de cultivares de soja

• Ano de defesa: 2023
• Autor(a) principal: Pacanhela, Eber Fabiano
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Tecnológica Federal do Paraná; Cornelio Procopio; Brasil; Programa de Pós-Graduação em Informática; UTFPR
• Palavras-chave em Português: Soja - Biotecnologia; Vegetação - Classificação; Agricultura - Aspectos ambientais; Soybean - Biotechnology; Vegetation - Classification; Agriculture - Environmental aspects; CNPQ::CIENCIAS EXATAS E DA TERRA; Ciência da Computação
• Link de acesso: http://repositorio.utfpr.edu.br/jspui/handle/1/32622

Soybean is the most cultivated oilseed in the world, with many varieties, high economic valeu, and higth protein content. Soybean production requires large amounts of nitrogen for the plant’s nutrition. This nutrient can be supplied through chemical fertilizers, which harm the environment. However, the biological nitrogen fixation (BNF) process has effectively increased productivity and minimized environmental impacts. BNFoccurs through the symbiotic association of diazotrophic bacteria in the system of legumes and oilseeds, where root nodules are formed. It is Known that technological advancements have driven the adoption of digital technologies inagriculture to improve efficiency and productivity, reduce environmental impacts, and promote hamn health. Due to its economic importance, soybean cultivation canbenfit from these technologies. This study proposes developing a computational intelligence methodology based on image processing and computer vision to analyze, classify, and automatically detect the root nodules resulting from BNF in soybean crops. In which, a future study, the effectiveness of nidulation can be evaluated, conseidering the quantity of nodules. For this purpose, classical models and convolutional neural networks are adopted for the classification task and computer vision models for nodule detection. The resuls show that traditional models and convolutional neural networks are promising for the classification task. In the detection task, convolutional neural networks demonstrated improved performance. However, there is still room for improvement to enhance nodule detection accurary in soybean cultivar roots.