| Paper Details: | Downloads: 1961 |
| Serial Number: | P1151616490
|
| Title: | Spectral-Spatial Classification of Hyperspectral Image based on Oversampling and Multi-Feature Kernels
|
| Authors: | Rafika Ben Salem and Karim Saheb Ettabaa and Mohamed Ali Hamdi
|
| Abstract: | Spectral-Spatial classification of hyperspectral image suffers from two problems: the existence of various feature extraction methods that complicate the choice of those applying and the availability of limited number of labeled training samples. To overcome these difficulties, this paper presents new spectral-spatial classification approach for remotely sensed hyperspectral image which integrates different spectral and spatial features via multi-feature kernels and process accurately with limited number of training samples. In fact, the proposed method introduces different methods to extract the spectral and the spatial features and exploits the oversampling based on interpolation techniques to generate new labeled samples. First, each pixel must be characterized by two spectral vectors computed according to the application of the principal components analysis, the independent components analysis and three spatial features calculated by using three methods: the average of neighbourhood pixels, the textural features and the extended multi-attribute profiles. Then an oversampling step is introduced to create new labeled samples used to train the classifier. Finally, a support vector machine (SVM) with multi-feature kernel is efficiently trained to generate the classification map. The proposed classification approach is experimentally evaluated using the AVIRIS Indian Pines data set, exhibiting higher performance when compared with the multi-feature classification without oversampling.
|
| Keywords: | Hyperspectral images, SVM, composite kernels, interpolation techniques
|
| Journal/Conference: | International Journal of Graphics, Vision and Image Processing
|
| Volume: | 16
|
| Issue: | 3
|
| Submission Date: | 4/9/2016 12:00:00 AM
|
| Review Date: | 9/11/2016 12:00:00 AM
|
| Publishing Date: | 12/6/2016 12:00:00 AM
|
| Article Downloads: | 1961
|
| Download: |
|