Assessing the Accuracy of Different Supervised Classification Methods of Satellite Image

Assessing the accuracy of the classification map is an essential area in remote sensing digital image process. This is because a poorly classified map will result in inestimable errors of spatial analysis and modeling arising from the use of such data. This study was designed to evaluate different supervised classification algorithms in terms of accuracy assessment with a view of recommending an appropriate algorithm for image processing. The analysis was carried out using Andoni L.G.A. Rivers State, Nigeria as the study area. Supervised classification of ETM+ 2014 Landsat image of the study area was carried out using ENVI 5.0 software. Seven land use/land cover categories were identified on the image data and appropriate information classes were also assigned using region of interest. The classifiers adopted for the study include SAM, SVM, and MDC and each classifier was set using appropriate thresholds and parameters. The output error matrix of the classified map produced overall accuracy and kappa coefficient for MDC as 94.00% and 0.91, SAM as 64.45% and 0.53, and SVM as 98.92% and 0.98 respectively. The overall accuracy obtained from SVM indicates that a perfect classification map will be produced from the algorithm. The advanced supervised classification should be utilized for classification of land use/ land cover for both high and medium resolution images for improved classification accuracy.