Browsing by Subject "Bayesian Networks"

In this study, a classification scheme is implemented to obtain high resolution snow cover information from Sentinel-2 data using a very simple Bayesian Network (Naive-Bayes) that is trained with ground snow measurement data. Performance comparison of using Bayesian/non-Bayesian Naive-Bayes, different feature sets and different discretization methods is conducted. Results show that Bayesian NB performs the best with up to 0.88 classification accuracy for snow/no-snow classification. Use of most relevant spectral bands rather than all available bands provided improvement in some cases but also performed slighty worse in some, hence not giving a clear answer. However, effect of discretization method was clear, chimerge performed better than equal width binning but it was much slower to a point that it was not practical to discretisize a full Sentinel-2 image’s pixels.