Multispectral Data Classification Based on Spectral Indices and Fuzzy C-mean

Numerous applications make use of data on land use and land cover (LULC). Given their importance and use, land cover data is assumed to be readily available or trivially acquired for a given landscape. Unfortunately, this is often not the case. LULC data at hand are often out-of-date, inappropriate for a particular application [1], or contain other difficulties. Thematic mapping of remotely sensed data is typically some form of image classification. Supervised image classification is generally achieved by either visual or computer-aided analysis, including classical statistical algorithms such as maximum likelihood, evidential reasoning, and artificial neural networks. One major limitation to the use of conventional supervised classification techniques for mapping land cover is that they were developed for the classification of classes that may be considered to be discrete and mutually exclusive, and assume each pixel to be pure, that is comprised of a single class. As a pixel is an arbitrary spatial unit, it may represent an area on the ground which comprises more than one discrete LULC class. This problem will be more apparent with coarse spatial resolution data. Fuzzy classification techniques can, however, accommodate the partial and multiple class membership of mixed pixels, and be used to derive an appropriate land cover representation. In this work, we propose the use of cascaded fuzzy C-mean classifiers for LULC classification. The main contribution to LULC classification is the use of spectral indices, and cascaded classifiers for the classification of remotely sensed data.