Hyperspectral image fusion using spectrally weighted kernels

Target detection from hyperspectral imagery requires the fusion of information from hundreds of spectral bands. In this paper, we study such fusion in the context of hyperspectral image classification. Previous research applying kernel methods such as support vector machines (SVMs) to hyperspectral image classification has achieved performance competitive with the best available algorithms. However few efforts have so far been made to extend SVMs to fit the specific requirements of this application, e.g., by building tailor-made kernels. To this effect, we propose a novel spectrally weighted kernel. Observation of real-life spectral signatures from the AVIRIS hyperspectral dataset shows that the useful information for classification is not equally distributed across bands. Hence, we propose the use of spectrally weighted kernels to assign weights to different bands according to the amount of useful information they contain. We have carried out experiments on the AVIRIS 92AV3C dataset to assess the performance of the proposed method. Results show potential for improvement in classification accuracy.