Rapid Classification of Mixed Hyperspectral Data by ROAA SVM

Capturing of high resolution hyperspectral images is one of the most expensive tasks in imaging industry. The main problem of low resolution hyperspectral data is the classification of pixels where more than one land cover type lie in one pixel, called a mixed pixel. To resolve this issue, methods composed of hard and soft classification techniques have shown good results. For rapid classification of these mixed hyperspectral images, we propose to use Reduced OAA SVM combined with spectral mixture analysis at sub-pixel level and a fast post processing step to eliminate unwanted solitary mappings. Experiments conducted over a common hyperspectral image show great improvements in terms of overall classification accuracy and computation time.