Publication details

Využitelnost algoritmů strojového učení pro klasifikaci multispektrálního družicového snímku

Title in English Usefulness of machine learning algorithms for multispectral image classification


Year of publication 2001
Type Article in Proceedings
Conference GIS Ostrava 2001. Sborník konference
MU Faculty or unit

Faculty of Science

Field Earth magnetism, geography
Keywords machine learning; multispectral imagery; classification; decision tree
Description The paper deals with the testing of the machine learning algorithms - especially decision trees - for multispectral image classification. At first common approaches to image classification were used. Six channels of LANDSAT TM data with pixel size 30 m were trained for 6 land cover types and classified with per-pixel classifiers. There were used the following methods: maximum likelihood method, minimum distance method, and paralellepiped methods. For each method land cover map and error matrix were prepared. Training data sets were then used for learning and construction of the decision tree classifier. In this project C4.5 and C5.0 algorithms were used and decision trees of various depths have been tested. Moreover, experiments with various rate of learning and testing data sets have also been done. From the first results we can conclude, that there is a great potential of decision tree classifiers for supervised image classification.

You are running an old browser version. We recommend updating your browser to its latest version.

More info