Análise com auxilio da inteligência artificial para a plantação de eucalipto

Abstract

Eucalyptus cultivation plays a key role in the Brazilian forestry sector due to its high productivity, fast growth, and economic importance, especially in the production of paper, cellulose, and energy. Estimating productivity requires an accurate count of plants, which is currently performed manually and through sampling, a process that is costly and prone to errors. This study proposes an automated methodology based on deep learning and drone imagery to count eucalyptus trees. Plants were annotated using central points, which were then dilated with different radii ($r$) to enable the application of segmentation techniques. The methodology encompassed image preprocessing, semantic segmentation model training, and evaluation using performance metrics such as precision, recall, F1-score, and error analysis across different dilation levels. Results showed that intermediate dilations (between r=5 and r=11) achieved the best balance between precision and recall, providing more reliable estimates of plant counts. Therefore, the proposed approach demonstrated practical feasibility for forest management, offering a promising alternative to automate eucalyptus seedling counting and improve productivity monitoring.