Hyperspectral imagery holds the essence of spectral and spatial attributes, making it a rich source of information for applications like target detection in the field of agriculture, urban areas, etc. The chapter provides an insight of extraction of shape-based features fused with the spectral characteristics for detection of urban targets particularly roads and roofs. Roads are considered as linear structures whereas roofs occur in clusters. Therefore, to detect them, spectral information is not sufficient. The work proposes integration of both spectral and shape aspects of a target, which has given better results using SVM. The experimentation involves results in the form of spectral analysis, shape analysis, fusion of spectral and shape-based features along with the detection rate and supporting results. Delineating the roads and roofs of urban topography is an essential fragment for city planning, urban sprawl estimation, population studies, sustainable development, etc. The work done here may help numerous governmental and non-governmental organizations for conducting related studies.
Top1. Introduction
Hyperspectral data has an advantage of high spectral resolution along with spatial information forming a three-dimensional structure often referred as hypercube, where first two dimensions corresponds to arrangement of pixels and third dimension is a continuous function of reflectance values with respect to different wavelengths. The spectral signature is unique and can be treated as fingerprint to detect, classify, identify, or map a particular target. Hyperspectral imagery is expansively used for urban forestry monitoring (Zhang, 2011), agriculture (Teng, 2009), marine research (Lou, 2008), mineralogy (Pei, 2007) and many more. Apart from the aforesaid applications, hyperspectral remote sensing extends its scope over target detection domain, where target may be confined to a single pixel (full-pixel targets) or within a pixel (sub-pixel targets). This process is based on binary hypothesis where target pixels are kept active with respect to the non-target or background pixels. The targets to be detected vary according to the application areas, such as for defense purposes (troops, bunkers), wildlife conservation (animals), similarly for urban areas, the major portion comprises of man-made infrastructure (roads and roofs).
Target detection algorithms are broadly categorized into spectral matching and anomaly detection. The former is a supervised technique working on labelled data available from standard spectral libraries, ground spectral data collection or from the image itself, whereas latter is an unsupervised approach, which extracts the abnormalities assuming them as targets. Algorithms like Spectral Angle Mapper (SAM), Spectral Correlation Angle (SCA) (De Carvalho, 2000), Matched filter (MF), Adaptive Cosine Estimator (ACE) (Truslow, 2013) belong to spectral matching and Orthogonal subspace projection (OSP), Reed-Xiaoli (RX) and uniform target detector (UTD) (Tiwari, 2011) are types of anomaly detection.
Urban advancement is happening at an exponential pace, having significant effect on national and international level. Urban expansion has far-reaching consequences on vegetative cover, water and air quality, and surface temperature, ultimately impacting the microclimate of human habitats and contributing to environmental degradation (Shafri, 2012). Fortunately, advancements in space-borne and air-borne technologies have simplified the process of mapping urban areas. This modern approach is not only more efficient but also significantly reduces the time and effort required, providing a comprehensive bird's-eye view of larger areas simultaneously. During past decade, researchers have proposed various techniques for determination of robust spectral and spatial features to detect the urban targets (Bo, 2009) (Heiden, 2007).