A method to automatically extract features from hyperspectral images is presented. The technique has been tested on two images: one from MIVIS airborne sensor and the other from AHS airborne sensors. The methodology has been organized in two parts: the first one performed a cluster extraction based on Kohonen’s neural networks (unsupervised); the second used the pixels belonging to the extracted clusters to train a supervised neural network (supervised approach). Both neural networks have been trained with spectral and textural parameters. Segmentation parameters have been also considered to help road discrimination. The obtained results showed the advantages of automatic classification in urban areas with neural networks, achieving a satisfactory accuracy in feature extraction
Lazzarini, M., DEL FRATE, F. (2010). Features extraction from hyperspectral images: an approach based on spectral, textural and spatial information applied to urban environments. In Proceedings of ESA hyperspectral workshop 2010,. ESA.
Features extraction from hyperspectral images: an approach based on spectral, textural and spatial information applied to urban environments
DEL FRATE, FABIO
2010-01-01
Abstract
A method to automatically extract features from hyperspectral images is presented. The technique has been tested on two images: one from MIVIS airborne sensor and the other from AHS airborne sensors. The methodology has been organized in two parts: the first one performed a cluster extraction based on Kohonen’s neural networks (unsupervised); the second used the pixels belonging to the extracted clusters to train a supervised neural network (supervised approach). Both neural networks have been trained with spectral and textural parameters. Segmentation parameters have been also considered to help road discrimination. The obtained results showed the advantages of automatic classification in urban areas with neural networks, achieving a satisfactory accuracy in feature extractionFile | Dimensione | Formato | |
---|---|---|---|
art_abstract.pdf
accesso aperto
Descrizione: short abstract
Licenza:
Creative commons
Dimensione
7.62 kB
Formato
Adobe PDF
|
7.62 kB | Adobe PDF | Visualizza/Apri |
Questo articolo è pubblicato sotto una Licenza Licenza Creative Commons