Kısa ve Uzun Dalga Boyu Kızılötesi Hiperspektral Görüntülerde Hedef Tespiti

Abstract

Main purpose of hyperspectral target detection methods is using a materials’ previously known spectral signature to find its position on a hyperspectral image which is taken later on. Nevertheless, even if signature of the material is previously known, atmospheric effects, reflections, subpixel targets and even temperature on long wavelength infrared hyperspectral images makes this application challenging. In this thesis, on two different spectral bands, namely the short and long wavelength infrared, methods are proposed to find a previously known target on a newly acquired image. On the conducted experiments, firstly comparisons on pixel, group of pixels and superpixel based matchings are made, secondly basic target detection algorithms from the literature are evaluted, and then depending on the band which is used, comparisions on the usage of radiance, emisivity and reflectance domains are made. And as the result of those, SWIR and LWIR bands are compared based on their target detection results. Then, the images belonging to those two bands are registered using their GPS location informations, and target detection performance is evaluated on those fused images consisting of the two different bands. For the last part, as an alternative to spectral matching algorithms, usage of temperature profiles on target detection with varying parameters like image sampling period and time interval are evaluated. SEABASS, HySpex and SPICE datasets are used on the conducted experiments.