Pixel- vs Object-based Land Use/Land Cover Classification: a Case Study of the Mendefera Sub-zone, Eritrea

The study compared the performance of the pixel-and object-based land use/land cover (LULC) classifications for the Mendefera sub-zone, Eritrea, using Landsat 8 OLI. The supervised pixel-based image classification was conducted in ArcMap, with the Support Vector Machine (SVM) and segmentation object-based image classification in ArcGIS Pro. Post-classification smoothing and the use of high spatial resolution aerial photos, along with Google Earth images, were employed to improve the accuracy of the exercise. DEM and high spectral resolution satellite images were also used in combination with false composite colours during the creation of the training samples. Overall accuracies of 83.7% and 67% and Kappa coefficients of 76.9% and 49% were obtained for the pixel-and object-based classifications, respectively. Thus, the study concluded that pixel-based LULC classification is the best classification mechanism for the given study area.