Extraction of urban built-up surfaces and its subclasses using existing built-up indices with separability analysis of spectrally mixed classes in AVIRIS-NG imagery

Abstract

Understanding the urban environments and their spatio-temporal behavior is necessary for local and regional planning along with environmental management. For monitoring and analyzing the urban environment, remote sensing imagery has been widely used due to its ability for repetitive coverage over large geographical areas. Compared with conventional per-pixel and sub-pixel analysis of remote sensing imagery, spectral indices have noticeable advantages because of their easy implementation and fast execution. However, most of the spectral indices are designed for multispectral imagery to extract only one land cover class, and confusion between other land cover classes still persists. This research explores the most significant spectral bands in AVIRIS-NG hyperspectral imagery for detection of built-up surfaces and its subclasses i.e. roads and roofs. Further, this study utilizes existing built-up indices for detection of urban built-up surfaces in the first level followed by its subcategories in the second level. Finally, a separability analysis between spectrally mixed urban land cover classes using various measures is also addressed. Results of the analysis indicate that BSI, NBI, and BAEI can prove to be effective for extraction of built-up surfaces with an overall accuracy (OA) of 93.89%, 90.11%, and 85.15%, respectively. Further, REI with OA of 94.40% appears to be suitable for extraction of road surfaces while NBAI with 95% OA can prove its efficacy for extraction of rooftops in AVIRIS-NG imagery. It also concludes that, for aforesaid indices, built-up surfaces (Level-1 and 2) can be effectively separated from the bare soil in hyperspectral imagery with slight confusion between road and roof surfaces.