عنوان مقاله [English]
Digital images which have been acquired by UltraCamD advanced aerial camera from Northern forests of Iran have potentially valuable data for obtaining useful information. In this study a pan-sharpened imagery (in 4 bands with 16 bit spectral and 7 cm spatial resolution) collected in 2008 from a forestation area near the town of Nur, was analyzed for tree identification. On-board GPS/IMU parameters along with ground control points using DGPS were applied to ortho-rectify the image precisely. Appropriate enhancement methods were accomplished and different band sets were utilized based on original and derivative images. The same training sets were utilized for both pixel-based and object-based classification methods. Reference map was produced through fieldwork for assessment of the accuracy of resulted maps. In pixel-based method supervised maximum likelihood classification was carried out. For object-based classification, segmentation was conducted stepwise at two levels in order to construct a hierarchical image object network. Initially various alternatives of segmentation with different color, shape, compactness, smoothness and scale parameters were tried. The classification hierarchy was developed and Nearest Neighbor classifier based on Fuzzy logic, using integration of different object features was performed. By examination of different features and band sets along with the revising training areas, the optimum classification framework was established based on Class separability, Best classification results, Class stability and Accuracy assessment. The comparison of the resulted maps with reference data showed that object-based approach produced significantly higher overall accuracy and Kappa index. Meanwhile the resulted maps indicated the nonexistence of “salt and pepper” effect, like pixel-based results. Furthermore object-based method separate properly the species mixed spectrally, considering spatial information. The accuracy of detailed vegetation classification with very high-resolution imagery is highly dependent upon the type and separability of the classes, segmentation quality, sample size, sampling quality, classification framework and ground vegetation distribution and mixture.