Impervious Surface Extraction from Worldview-2 Imagery Using an Object-Oriented Rule-Based Approach

Impervious surfaces are mainly man-made surfaces that are covered by asphalt, concrete, brick and rooftops. They provide detailed information for evaluating conditions of urban environment. However, it is difficult to extract detailed information about impervious surfaces due to the complex diversity of urban materials. Remote sensing data can provide a promising perspective for impervious surface extraction with the advantages of very high spatial and spectral resolution images. Object-based image analysis (OBIA) stands out as an effective method for large-scale urban structure mapping using high and very high resolution imagery. The first and crucial step of OBIA is image segmentation in that meaningful image objects are obtained based on spatial, textual and contextual information in addition to spectral information of adjacent pixels. This study examines the effectiveness and advantages of OBIA approach for the extraction of urban impervious surfaces using a pan-sharpened eight-band Worldview-2 imagery of San Clemente, California. The particular site selected for this study includes major impervious surfaces of building rooftops and asphalt roads that were extracted by applying object-oriented rule-based approach. In the rule-based approach, decision rule sets were established by determining appropriate functions according to the characteristics of the objects in the image and assigning them to the classes to which they belong. Multi-resolution segmentation algorithm provided in eCognition software was employed in this research for rule-based image segmentation. In iterative steps, the image was segmented into three-level of different scales, which were, 25, 50, and 100 respectively. All segments were classified into corresponding classes according to their spectral, spatial and geometric characteristics. An accuracy assessment was achieved for the final outcomes using contingency matrices. The maximum likelihood classifier was employed to evaluate the overall and individual class accuracies. The user's accuracies for building rooftop was 90.7% and 92.2% for the asphalt roads. As a result, an overall accuracy of 91.5% was achieved. Results show that impervious surface distribution can be derived from OBIA with promising accuracy in comparison to pixel-based approach.