ACCURACY ASSESSMENT OF PIXEL-BASED IMAGE CLASSIFICATION OF KWALI COUNCIL AREA, ABUJA, NIGERIA

In this study, Kwali Council Area located on the western part of the Federal Capital Territory, Abuja was selected as a study area covering approximately 1,206 km² for comparing the two major pixel-based image classification algorithms (Supervised and Unsupervised classification). For this purpose, land use and land cover classification of the study area was conducted by supervised classification particularly maximum likelihood classification (MLC) and Iso-cluster unsupervised classification procedures and the results were compared with one another using 2011 Landsat-7 ETM+ satellite. However, the result of classification accuracy illustrates that light vegetation shrubs records dominance value of 27.54%, savannah grasses 23.04%, cultivated areas 20.12%, wetland flood plain 13.78%, sand open surfaces 11.01% and water body 4.52%. Overall, supervised pixel-based classification methods are found to be more reliable, accurate and outperformed unsupervised pixel-based classification methods in this study. The higher accuracy was attributed to the fact that supervised classification took advantage of spectral information of land cover, based on the spectral signature defined in the training set and digital image classification software that determines each class on what it resembles most in the training set in the remotely sensed imagery. This study is a good example of some of the limitations of unsupervised pixel-based image classification techniques, whereby the unsupervised image classification technique is commonly used when no sample sites exist. These improvements are likely to have significant benefits for land-cover mapping and change detection applications. It is recommended that, the two approach can be used together to provide a standard, accurate and finest result for specific applications by users in different parts of the world. Keywords: Accuracy, assessment, pixel-based image classification algorithms, iso-cluster unsupervised, ML