A new classification scheme for urban impervious surface extraction from UAV data

Urban planning and comprehensive understanding of human-environmental interactions within cities require precise and up-to-date information of urban impervious surfaces. In this regards, the emergence of accurate data, particularly short-range photogrammetric data with high spatial and spectral resolutions has proven to be essential. Urban impervious surfaces are diverse and comopsed of disperate materials; this introduces challenges in the extraction of these surface from remote sensing images. Despite this diversity, only limited types of impervious surfaces (up to six classes) can be diffierenciated in remote sensing approaches, that is insufficient to meet the requirements of researchers, managers, and urban planners. Therefore, this study proposes a novel classification scheme to extract as much information as possible from UAVs visible and elevation data. The substantial influence of classification scheme on segmentation accuracy was assessed. To achieve this objective, a classification scheme consisting of 17 classes was developed to identify the various types of impervious surfaces present in Ahvaz (with diverse impervious surfaces), Iran. Then, UAV RGB and digital elevation model (DEM) images with 5 and 10 cm spatial resolutions, respectively, were segmented through 45 different parameter combinations in the study area. Then, the optimal parameter for segmention the urban impervious surface classes individually and collectively were selected. Finally the images were classified by two methods, namely simple and stratified, using the SVM algorithm. Results demonstrated that impervious surfaces could be accurately extracted in 16 distinct classes with error less than 20%. Image segmentation and classification based on the stratified method in comparison to simple ones improved the urban impervious surface extraction overall accuracy from 72 to 81%, while also improving the minimum user and producer accuracies from 53 and 14 percent to 55 and 62 percent, respectively. The proposed detailed classification scheme can improve urban impervious surface classification in diverse and heterogenous urban environment.