Clustering Approach on Land Use Land Cover Classification of Landsat TM Over Ulu Kinta Catchment

Providing accurate land use/ land cover (LULC), as an important component in remote sensing, have been a point of concern which subjected to many studies. Researchers have made great efforts in developing superior classification approaches and techniques which can lead to classification accurateness. This study was sought to investiagte LULC over Ulu Kinta Catchment in multispectral Landsat TM by clustering + approaches. The proposed technique is carried out on data from training part by per-pixel analysis in corresponding image and principal component (PCA) map to yield a better understanding of DN data set representation for each LULC classes. A version of PCA has been used on the basis of an optimum order sample correlation coefficient for enhancing the contribution of the image bands and simulating the best band combination for unsupervised classification. The result showed that in spite of overlapping in some classes, first component of PCA map enhanced discrepancies between features and could be an acceptable layer in the band combination. Furthermore, by considering band combination with the maximum separability the NDVI index map, band 3 and band 4 of Landsat TM was proposed. Initial number of cluster in the study area fixed to 30 and an unsupervised classification relate to arbitrary selection of the number of classes was carried out. The application of post-processing method to labeling clusters was also discussed and 8 major classes were represented. Additionally, the performance of the proposed band combination based on 8 classes was evaluated and the higher accuracy determined in proposed technique i.e.from 72.1% in previous study incresed to 80.1%.