Building detection using hyperspectral Images by Support Vector Machines

Building detection is one of the important applications in processing hyperspectral images. In order to detect complete and precise building information from hyperspectral data, advanceddata analysis methods are required. Algorithms based on spectral-identification are sensitive to spectral variability and noise in acquisition. In most cases, the spatial distributions and spectral signature are unknown, so each pixel is separately examined and if it significantly differs fromthe background, it is regarded as an object. On the other hand, there are many classic (e.g. Maximum Likelihood (ML)) and non-classic (e.g. Modified Spectral Angle Similarity (MSAS) as a Deterministic and Adaptive Coherence Estimator (ACE), Covariance-based Matched Filter Measure (CMFM) as sub-pixel approach) algorithms for building detection. In this study, first we propose a theoretical discussion aimed at understanding and assessing the potentialities of MLC, MSAS, ACE, CMFM algorithms. These algorithms work only based on spectral image data. In order to evaluate the detection algorithm based on hyper-dimensional feature spaces, Support Vector Machines (SVM) has been implemented which in the case of building detection, it may be regarded as a new application. The study includes accuracy assessment of effectiveness of SVM with respect to mentioned conventional algorithms regarding the performance indicators. The experiments on the building detection application, using three CASI hyper-spectral images taken from an urban area allow concluding that, SVM is a suitable and effective alternative to conventional detection algorithms.

Keywords:

CASI hyperspectral image , Building Detection , SVM , Kappa Coefficient