| Sumario: | Many studies have been carried out to find the best method for classification of remote sensing data. Traditional methods depend only on pixel value without using the spatial and geometric information of object which is consider an important source to classify the satellite imagery. This study compared between pixel-based and object-based classification algorithms using Sentinel-2 imagery. As for object-based image analysis, the image was segmented in to homogeneous groups using appropriate variables such as scale, shape, size and probability. Classification based on segments was done by a nearest neighbor classifier, while the classification based on pixel value was done by manual digitizing by using the principles of visual interpretation. For accuracy assessment, an error matrix was used to assess the classification between two types of classification. The producer's accuracy for gully class using object-based image analysis was 90.7% while the user's accuracy for gully using pixel classification was 84.8%. These classification techniques were subjected to accuracy assessment and the overall accuracy was 89.2% and 0.78 for Kappa coefficient. The results of classification and its accuracy assessment show that the object-based image analysis approach gave more accurate results.
|
|---|