Combination of spatial and spectural charactristics to improvement of sattelite images classification accuracy

Pixel-based classification techniques have low capability when different phenomena have similar spectral characteristics that result in low accuracies. But object-oriented techniques use a combination of spectral and spatial characteristics and segments (image objects) create from pixels based on shape, texture and color tone factors and with specific segmentation scale. Then, each segment belong to a class based on its fuzzy membership parameter. In this study, maximum likelihood pixel-based classification algorithm compared with nearest neighbor object-oriented classification algorithm using landsat8 OLI multispectral image. Land-use map of Rudzard basin of khoozestan was prepared using both methods. Accuracy assessment results show that, the object-oriented technique increases overall accuracy by 6% and kappa coefficient by 7% than the pixel-based technique, so has a higher accuracy for satellite images classification. Keywords: pixel-based; object-oriented; fuzzy logic; satellite image classification, khuzestan.