Along with the developing remote sensing and hypespectral imaging systems, many image archives have been collected, and many systems that contain this growing hyperspectral image space have been developed. Providing fast image search and access on rapidly growing hyperspectral image archives has become a great need. In the project, based on this need, aim is to realize content based image retrieval on the hyperspectral image collection. Existing methods in the literature were examined and experiments were carried out with deep learning methods, considering that hyperspectral images have rich spectral information.
In the study, deep networks with 2D and 3D CNN architectures were developed to evaluate the bands in hyperspectral images, some deep network architectures developed for hyperspectral data in the literature were implemented and scores were obtained on the AnkaraHSI dataset. Scores are presented in comparison with reference studies.
As a result of the study, the performance scores in the reference study were increased as expected with the developed models and the experimental results were recorded.
Check the report for further documentation.