Target detection on hyperspectral images with composite kernel covariance descriptor

Hyperspectral imaging has become popular because of its advantages over the other imaging techniques and rapid improvements on hyperspectral technologies. In spite of its high banwidth and time-consuming process, because of its high spectral resolution and sensitivity to material on the image, hyperspectral imaging is often used in image processing. In this thesis, target detection is performed with Kernel Covariance Descriptor and Composite Kernel Covariance Descriptor on hyperspectral images. Target detection successes of proposed methods are compared to classical covariance descriptor (CD) by applying these methods on AVIRIS, KSC (Kennedy Space Center), PAVIA and ÇATALCA01 datasets. During the tests, all classes of KSC and AVIRIS and specific target classes of ÇATALCA01 and PAVIA datasets are included. The test results show that KCD and CKCD methods return better results than classical CD. Also, during the tests, different image features are applied to CD method and the results are evaluated. Within the scope of this work, spatial data is also included with spectral data for CKCD method and better results are found.