Abstract:
Knowledge about land cover is an important input for the modeling of information that can be used for planning of proper utilization of natural resources. The derivation of such information increasingly relies on remote sensing technology due to its ability to acquire measurements of land surfaces at various spatial and temporal scales. One of the major approaches to deriving land cover information from remotely sensed images is classification. Conventionally hard classifiers are used which give the output having every pixel in a single land cover class which is far from the actual scenario on ground as well as loss of information is there. Unlike hard classifiers, sub-pixel (soft) classifiers defer making a definitive judgment about the class membership of any pixel in favor of producing a group of statements about the degree of membership of that pixel in each of the possible classes.
There is a constant endeavor for obtaining more accurate results of classification. Accuracy evaluation of such individual classification technique and mutual comparison of the performance of accuracy assessment methods are key issues of debate and research in the field of remote sensing. Accuracy is itself defined as "the closeness of results of observations, computations, or estimates to the true values or the values accepted as being true' (USGS, 1990). These methods are categorized according to their basic concept like distance, similarity, uncertainty and fuzzy data set. Some latest features like fuzzy correlation coefficient, various entropy measures, fuzzy kappa and new operators in fuzzy error matrix are also discussed. A few of these measures are applied on actual data set with Bayesian and fuzzy classifiers used. Both mixed and pure training data are used for their classification. Their comparative analysis is done through statistical results and graphs.. In both the cases mixed training data provided batter classified image than pure training data. Accuracy results obtained by fuzzy classification were found to be better than Bayesian classifier.
