Optimizing multiresolution segmentation for extracting plastic greenhouses from worldview-3 imagery

Multiresolution segmentation (MRS) has been pointed out as one of the most successful image segmentation algorithms within the object-based image analysis (OBIA) framework. The performance of this algorithm depends on the selection of three tuning parameters (scale, shape and compactness) and the bands combination and weighting considered. In this work, we tested MRS on a World‐ View-3 bundle imagery in order to extract plastic greenhouse polygons. A recently published command line tool created to assess the quality of segmented digital images (AssesSeg), which implements a modified version of the supervised discrepancy measure named Euclidean Distance 2 (ED2), was used to select both the best aforementioned MRS parameters and the optimum image data source derived from WorldView-3 (i.e., panchromatic, multispectral and atmospherically corrected multispectral orthoimages). The best segmentation results were always attained from the atmospherically corrected multispectral World‐ View-3 orthoimage.