Detecting salient regions in a bi-temporal hyperspectral scene by iterating clustering and classification
- PDF / 1,488,823 Bytes
- 22 Pages / 595.224 x 790.955 pts Page_size
- 61 Downloads / 165 Views
Detecting salient regions in a bi-temporal hyperspectral scene by iterating clustering and classification Annalisa Appice1,2 · Pietro Guccione3 · Emilio Acciaro1 · Donato Malerba1,2
© Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract Hyperspectral (HS) images captured from Earth by satellite and aircraft have become increasingly important in several environmental and ecological contexts (e.g. agriculture and urban areas). In the present study we propose an iterative learning methodology for the change detection of HS scenes taken at different times in the same areas. It cascades clustering and classification through iterative learning, in order to separate salient regions, where a change occurs in the scene from the unchanged background. The iterative learning is evaluated in both the clustering and the classification steps. The experiments performed with the proposed methodology provide encouraging results, also compared to several recent state-of-the-art competitors. Keywords change detection · clustering · classification · hyperspectral data
1 Introduction Change Detection (CD) in multi-temporal datasets is a key task in visible (VIS) and near infrared (NIR) remote sensing. The Earth’s surface is constantly changing due to anthropogenic causes such as landslides, floods, deforestation, crop growth or harvesting and natural reasons such as the progression of desert areas, glacier movement and earthquakes [2]. Careful monitoring of these changes with time provides valuable information on the transformation of the environment and would allow a better Annalisa Appice
[email protected] Pietro Guccione [email protected] Emilio Acciaro [email protected] Donato Malerba [email protected] 1
Department of Informatics, Universit`a degli Studi di Bari Aldo Moro, via Orabona, 4, I-70125, Bari, Italy
2
Consorzio Interuniversitario Nazionale per l’Informatica CINI, Bari, Italy
3
Dipartimento di Ingegneria Elettrica ed Informazione, Politecnico di Bari, via Orabona, 4, I-70125, Bari, Italy
decision policy on the use of territory or could be used to reduce the risk of disaster [45]. Thanks to continuous technological development, multispectral and hyperspectral cameras have reached unprecedented spatial and spectral resolutions [27, 48]. Mounted on spaceborne systems (LandSAT, Sentinel-2, Deimos, Ikonos, GeoEye and others), such sensors allow a frequent revisit time with constant characteristics, such as the same sun illumination and incidence angle if platforms are put in sunsynchronous orbit. On the other hand, the light weight of a spectral camera also permits airborne campaigns, such as the NASA Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) or UAV fast acquisitions for small-sized experiments (such as the RTK X8 Hyperspectral Mapping described by [22]). The large amount of acquisitions in the same area at different time points has unleashed the potential of methods for comparing images. In this view, the goal is the analysis of how the spectral reflectanc
Data Loading...
