Título: Evaluation of a genetic programming – based framework for NIR estimation from RGB bands
Autores: Diego Saqui, Henrique Luis Moreira Monteiro, Rafael Rodrigues Mendes Ribeiro, Lucas Eduardo de Oliveira Aparecido, Steve T. M. Ataky & Danton D. Ferreira
Resumo: Remote sensing using the Near Infrared (NIR) band is a well-established technique for assessing plant health in precision agriculture. However, traditional approaches require costly multi- and hyperspectral sensors, limiting their accessibility for small and medium-sized farms. This study proposes a low-cost alternative by estimating NIR reflectance from standard RGB imagery using Genetic Programming (GP). GP automatically evolves mathematical models to predict NIR values, which are then used to compute the Normalized Difference Vegetation Index (NDVI)—a widely used indicator of plant vigor and health. The method was evaluated on three benchmark hyperspectral datasets (Indian Pines, Salinas, and Pavia University) across two scenarios: (1) models trained per dataset and (2) a generalized model trained on combined datasets. GP demonstrated strong performance, achieving Pearson correlation coefficients up to 0.9059 and spectral angle errors below 0.001 radians. NDVI estimations were particularly accurate, with correlations exceeding 0.97 in some cases. Due to its simplicity, interpretability, and low computational requirements, the proposed approach offers a practical and scalable solution for vegetation monitoring in resource-constrained agricultural environments, enabling small and medium-sized producers to leverage spectral analysis without expensive sensors.
Palavras-chave: Genetic programming; NIR estimation; vegetation indices; precision agriculture; remote sensing; low-cost solutions.
Páginas: 8
Código DOI: 10.21528/CBIC2025-1168264
Artigo em PDF: CBIC_2025_paper1168264.pdf
Arquivo BibTeX:
CBIC_2025_1168264.bib