Hyperspectral Image Defogging Method Based On Unmixing Theory

Hyperspectral remote sensing has high spectral resolution, map-one, multi-spectral bands and within a range of continuous spectral imaging segment. Therefore we can use the curve of full spectrum to recode the observed variety of feature information. Hyperspectral remote sensing has been widely applied in the field of geology, vegetation ecology, air, soil and water environmental. However, hyperspectral image will still be interfered by many factors during imaing. Cloud and fog are the main factors. They interfere with the propagation of visible light and infrared in atmosphere, reduce the atmospheric visibility, and seriously affect the normal operation of optical device. In addition, they make the optical remote sensing platforms to acquire blurred hyperspectral images, result in that we can’t derive accurate feature information. Therefore an effective method for the defogging development of hyperspectral images is urgent to resolve.In this paper, the defogging algorithm of hyperspectral images is proposed. Firstly, hyperspectral unmixing theory is introduced, and then we establish the model of unmixing in the fog weather. We compare the three classic endmember extraction algorithms (PPI, VCA, ATGP) and four abundance inversion algorithms based on the least squares inversion method through the test. Secondly, Non-negative matrix factorization theory is researched. Based on this theory non-negative matrix decomposition minimum volume constraints algorithm is studied. Experiments show that the method can achieve a good result. Finally, the support vector machine theory and support vector data description theory are introduced. Because the traditional hyperspectral image unmixing ignores the influence of the unknown endmember. On the basis of SVDD, the defogging method of hyperspectral image is proposed. The experimental result of the algorithm is analyzed.