Services on Demand
Journal
Article
Indicators
- Cited by SciELO
- Access statistics
Related links
- Cited by Google
- Similars in SciELO
- Similars in Google
Share
Revista Facultad Nacional de Agronomía Medellín
Print version ISSN 0304-2847
Abstract
ARANGO GUTIERREZ, Mauricio; BRANCH BEDOYA, John William and BOTERO FERNANDEZ, Verónica. NONSUPERVISED CLASSIFICATION OF VEGETABLE COVERS ON DIGITAL IMAGES OF REMOTE SENSORS: "LANDSAT - ETM+". Rev. Fac. Nac. Agron. Medellín [online]. 2005, vol.58, n.1, pp.2611-2634. ISSN 0304-2847.
The plant species diversity in Colombia and the lack of inventories of them suggests the need for a process that facilitates the work of investigators in these disciplines. Remote satellite sensors such as LANDSAT ETM+ and non-supervised artificial intelligence techniques, such as self-organizing maps - SOM, could provide viable alternatives for advancing in the rapid obtaining of information related to zones with different vegetative covers in the national geography. The zone proposed for the study case was classified in a supervised form by the method of maximum likelihood by another investigation in forest sciences and eight types of vegetative covers were discriminated. This information served as a base line to evaluate the performance of the non-supervised sort keys ISODATA and SOM. However, the information that the images provided had to first be purified according to the criteria of use and data quality, so that adequate information for these non-supervised methods were used. For this, several concepts were used; such as, image statistics, spectral behavior of the vegetative communities, sensor characteristics and the average divergence that allowed to define the best bands and their combinations. Principal component analysis was applied to these to reduce to the number of data while conserving a large percentage of the information. The non-supervised techniques were applied to these purified data, modifying some parameters that could yield a better convergence of the methods. The results obtained were compared with the supervised classification via confusion matrices and it was concluded that there was not a good convergence of non-supervised classification methods with this process for the case of vegetative covers.
Keywords : SOM; ISODATA; LANDSAT; principal component analysis; non-supervised classification; vegetative covers.