MÉTODOS PARA REDUÇÃO DO ESPAÇO N-DIMENSIONAL DE IMAGENS ORBITAIS
Orbital images contain redundancy information in spectral bands and, in hyperspectral sensors, with grate dimensionality (grate number of bands), difficult processing. In methods by reduction date the objective is reduce the dimensionality and redundancy at orbital dates. In this work the principal...
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Universidade Federal de Uberlândia
2006-04-01
|
| Series: | Revista Brasileira de Cartografia |
| Subjects: | |
| Online Access: | http://www.rbc.ufrj.br/_pdf_58_2006/58_01_3.pdf |
| Summary: | Orbital images contain redundancy information in spectral bands and, in hyperspectral sensors, with grate dimensionality (grate number of bands), difficult processing. In methods by reduction date the objective is reduce the dimensionality and redundancy at orbital dates. In this work the principal methods for reduction of the data dimensionality w ere compared using Landsat TM image for application case. The objective was available these methods with relation the information loss that affect the classification accuracy. The methods were: bands selections by divergence index, Information Analysis Mutual and Cramer’s V index, and Transformation of the n-dimensional space by Canonical Analysis, Principal Components analysis, and Tasseled Cap. The results indicated which Cramer’s V is the potential index by bands selection. The information analysis mutual was unstabled, selecting worst band combination and the transformation for canonical analysis was better midst transformations. |
|---|---|
| ISSN: | 0560-4613 1808-0936 |