Field-Based Classification of Different Resolution Images and the Filtering Effects on the Accuracies

Field-based classifications of multispectral SPOT 4, SPOT 5, IKONOS, QuickBird and the pansharpaned (PS) QuickBird images were performed for an agricultural land located near the 'Karacabey Plain' of Turkey. The main objectives were to evaluate the classification accuracies and to find out the effects of filtering on the accuracies of different resolution images for agricultural crop mapping. To fulfill these objectives, first, a Maximum Likelihood Classification (MLC) was performed on each satellite image in a per-pixel manner. Second, for each agricultural field, the frequencies of the classified pixels were computed and the majority class was assigned to each parcel as the class label. In order to see the differences between all bands (blue, green, red, near-infrared, and shortwave near-infrared), common bands (green, red, near-infrared), and filtered bands, the images were analysed using different number of bands. Then, the confusion matrices were computed to get the accuracies of the results. Of the images used, the IKONOS XS, QuickBird XS, and QuickBird PS images provided relatively higher overall accuracies (about 85%) than the other images. On the other hand, for the SPOT4 XS image, the lowest accuracies of around 75% were computed.