object-based classifiers

Due to increase of demand for industrial and agricultural products, many tropical regions of Iran have experienced landscape changes. The present study aims to detect the land use/land cover (LULC) using some pixel/object-based approaches.  

This research was conducted in Jiroft area using some pixel-based and object-based image analysing methods (PBIA and OBIA respectively). To this end, at the first phase, the LULC maps were extracted using PBIA for September, 2020. The PBIA are including as Mahalanobis distance (MD), maximum likelihood (ML), neural network (NN), support vector machine (SVM). At the second phase, the LULC was produced using OBIA approach, encompassing the multi-resolution method and decision tree (DT) technique, for segmentation and classification respectively. Using a hybrid methodology, the high-resolution images of Worldview-2 were segmented. The segmented objects were later combined with the 7-month time series of NDVI, to find the necessary thresholds for DT.  

Results of the LULC maps demonstrated that the kappa coefficient and overall accuracy for ISODATA, MD, ML, NN, and SVM methods were calculated to be (51%, 66%), (81%, 86%), (88%, 91%), (90%, 93%) and (88% and 92%), respectively. The outcomes of the second phase for mapping the LULC showed the OBIA achieved a high overall accuracy of about 96%.

among the PBIA techniques and regarding both accuracy and execution time, the ML was the best. Although both PBIA and OBIA approaches are applicable in mapping LULC, the OBIA significantly outperformed the PBIA classifiers by higher overall accuracy and Kappa statistics