Study of Sub-Pixel Classification Algorithms for High Dimensionality Data Set

In this work fuzzy set theory based as well as statistical learning algorithm have been studied at sub-pixel classification level. Here two Fuzzy set theory based classifiers, namely, Fuzzy c-Means (FCM) and Possibilistic c- Means (PCM) have been used in supervised modes. Support Vector Machines (SVMs) have been used in this study for density estimation as a statistical learning based sub-pixel classifier while using Mean Field (MF) method for learning. An in-house package SMIC (Sub-Pixel Multi-Spectral Image Classifier) was used and sensitivity of all the three algorithms (FCM, PCM and SVMs) has been checked for dimensionality data sets at 3 to 14 bands from ASTER data. The accuracy of sub-pixel classification outputs has been evaluated using Fuzzy Error Matrix (FERM). In contrast to FCM and PCM, SVM approach showed a clear increase in the accuracy with higher dimensionality data and clearly out performed other two approaches for sub-pixel classification.