Comparison of land cover image classification methods (Case Study: Ejisu-Juaben District)

Thumbnail Image
Journal Title
Journal ISSN
Volume Title
The use of remote sensing techniques for land cover classification have gain more and more importance and recent direction in research works indicates that image classification of satellite images for land cover information is the preferred choice. Various methods for image classification have been developed based on different theories or models. In this study, three of these methods Maximum Likelihood classification (MLC), Subpixel classification (SP) and Backpropagation Neural Network classification (BPNN) are used to classify a landsat etm+ image of the Ejisu-Juabeng district of Ghana into seven land cover classes and the results compared. MLC and BPNN are hard classification methods but SP is a soft 4 Classification. Hardening of soft classifications for accuracy determination leads to loss of information and the accuracy may not necessary represent the strength of class membership. Therefore in the comparison of the methods, the top 20% compositions per land cover class of the SP were used instead. Results from the classification, indicated that output from SP was generally poor although it perform well with land covers such as forest that homogeneous in character. Of the two hard classifiers, BPNN gave a better output than MLC. Overall, BPNN gave the best results with an accuracy of 92.50%, whiles MLC gave an accuracy of 78.95%.
A Thesis submitted to the Department of Geomatic Engineering, Kwame Nkrumah University of Science and Technology in partial fulfilment of the requirements for the degree of Master of Science,