Comparação de Classificadores Multitemporais em Cascata

Author: Ligia Marcela Tarazona Alvarado

Original title: Comparação de Classificadores Multitemporais em Cascata

Language: Brazilian Portuguese

Electronic version: Portuguese | English


This dissertation made a comparison between three cascade multitem-
poral image classification methods. The methods are based on: (1) Support
Vector Machines (SVM); (2) Hidden Markov Models (HMM) and (3) Fuzzy
Markov Chains (FMC). A prior knowledge of the models regarding the
transition between classes were designed and classification data was eval-
uated. The robustness of the classification model is verified, outliers were
introduced in the data entry. Additionally, method’s performance is eval-
uated when the number of occurrence of transition classes in the training
set differs the number of occurrence in the testing set. The real gain of
a prior knowledge of the models was determined regarding the transition
between classes. A experimental analysis was conducted over two sets with
different characteristics, one a pair of IKONOS images of Rio de Janeiro
and the other a pair of LANDSAT7 images of Alcin´opolis, Mato Grosso do
Sul. This study has concluded that overall accuracy of the three approaches
had a similar behavior for different experiments. A confirmation that mul-
titemporal approach has a better overall accuracy rate than monotemporal
approach was achieved.