Using non-parametric classifiers to assess the performance of Sentinel 2 and Landsat 8 for land cover classification
Abstract
There are several factors that have been proved in literature to affect the accuracy of remote sensing classification images, among those are the choice of the classification algorithm as well as the spatial resolution of the satellite image used.
Although a large number of new image classification algorithms have been developed and more satellite images availed for improved remote sensing image classification, these two parameters are rarely studied together to determine their combined effect tested and the focus was on the accuracy of resultant classified images. In this research, with the same classification scheme over Mbale Municipality, three non-parametric classifiers popularly used in remote sensing and two satellite images were tested, focus was mainly on how the choice of classifier and satellite image influences accuracy achieved.
Analysis was based on different accuracy measures as well as the reliability and processor speed demonstrated by the individual algorithms using the spatially different image data sets. Results show that all the three algorithms achieve confidently good results while using both images with a little more training data for when using Landsat 8. However, RF classifier proved to be more reliable and provides the greater accuracies with both images.