Comparison of object-oriented and pixel-based classification methods for land use mapping (Case study: Isfahan-Borkhar, Najafabad and Chadegan plains)

Change detection algorithms of remote sensing image can be divided into two categories: pixel-based and object-oriented, according to the minimum processing unit. This paper deals with the comparison between application of pixel-based and object-oriented approaches in land use classification in Isfahan-Borkhar, Najafabad and Chadegan plains and evaluation of land use changes with Landsat TM (1985) and OLI (2015) data during the study period. The object-oriented approach involved the segmentation of image data into objects with multi-resolution segmentation algorithm by eCognition software. Then objects were assigned and classified with the nearest neighbour algorithm in object-oriented classification The supervised pixel-based classification involved the selection of training areas and a classification using a maximum likelihood algorithm. Accuracy assessments of both classifications were undertaken. The results show better overall accuracy (higher 90%) of the object-oriented classification over the pixel-based classification. The land use maps indicate that residential area is increased 2.09, 9.66 and 3.74% and rangeland area is decreased 7.48, 10.94 and 17.73% in Isfahan-Borkhar, Najafabad and Chadegan plains in the study period, respectively. In Chadegan plain the increase in agriculture and fallow land use has been equal to 8.31 and 5.64%, respectively.