Serviços Personalizados
Journal
Artigo
Indicadores
- Citado por SciELO
- Acessos
Links relacionados
- Similares em SciELO
Compartilhar
Agrociencia
versão On-line ISSN 2521-9766versão impressa ISSN 1405-3195
Resumo
MACEDO-CRUZ, Antonia; PAJARES-MARTINSANZ, Gonzalo e SANTOS-PENAS, Matilde. Unsupervised classification with ground cover color images. Agrociencia [online]. 2010, vol.44, n.6, pp.711-722. ISSN 2521-9766.
Ground cover is a geographic element that constitutes the reference base for diverse applications related with planning the use of natural resources, biodiversity and deforestation, as an indicator of climatic change and desertification. There are diverse methods for image classification, but not all of them are applicable to the classification of ground cover. Therefore, the objective of this investigation was to generate and evaluate an unsupervised classification model applicable to ground cover images in color model RGB (Red, Green and Blue), which automatically classifies the different covers and soil uses. The developed methodology corresponds to the group of unsupervised classification procedures and was derived from the thresholding criterion suggested by the Otsu method, based on the variance among and within classes. For its application, aerial photographs were used with spatial resolution of 1:19 500, spectral resolution of three visible RGB bands, and radiometric resolution of 0 to 255 levels. The results were compared with those reported by an expert with knowledge of the zone of study. The principal contribution of this investigation was to generate and evaluate a method of unsupervised image classification of high precision and low computational cost, such as the alternative use of the Otsu method, utilized in the segmentation and classification of color images. Therefore, the result was a classifier by unsupervised clustering which recognizes and classifies up to 95% of the elements identified in the field by the expert.
Palavras-chave : classification; RGB color model; Otsu method; segmentation; thresholding; artificial vision.