Título: A Object Detection Method for Native Seedling Identification in Caatinga and Atlantic Forest
Autores: Arysson Silva de Oliveira, Rodrigo de Paula Monteiro, Regina Lúcia Félix de Aguiar Lima, Márcia Macedo & Carmelo J. A. Bastos-Filho
Resumo: Plants are essential for agriculture, environmental preservation, and ecosystem restoration. In biomes like the Caatinga and Atlantic Forest in Pernambuco, Brazil, seedling planting is a key strategy for recovering degraded areas. However, identifying plant species at the seedling stage is a challenging task. This work applies object detection models, specifically from the YOLOv11n family, to identify pinhão-bravo and pau-ferro seedlings, analyzing the impact of dataset size and training epochs. Results show that increasing the training dataset significantly improves performance, with data augmentation further enhancing model robustness, though without statistically significant differences. Conversely, increasing the number of epochs beyond 25 brought minimal gains. The findings demonstrate that object detection-based models are practical tools for supporting seedling monitoring in reforestation projects.
Palavras-chave: Plant Seedling Identification; Object Detection; Computer Vision; Machine Learning; Deep Learning; environment.
Páginas: 8
Código DOI: 10.21528/CBIC2025-1190562
Artigo em PDF: CBIC_2025_paper1190562.pdf
Arquivo BibTeX:
CBIC_2025_1190562.bib