Hyperspectral image destriping method based on time-frequency joint processing method

Abstract

A novel hyperspectral image destriping method based on time-frequency joint processing(TFJP) method is proposed and demonstrated experimentally. A tripe noise may degrade the image quality and bring difficulties to data classification and information restoration. In this paper, through the analysis of SPARK microsatellite hyperspectral imager, the stripe noise is caused by the hardware circuit synchronization error. A stripe noise model of the SPARK microsatellite hyperspectral imager is constructed. On the basis of the stripe noise model and its characteristics, the TFJP method is proposed. The TFJP method removes stripe noise by replacing the wavelet components of different wave bands using two dimensional discrete wavelet transform (2D-DWT) and histogram matching. Several experiments on image quality evaluation are conducted to obtain qualitative and quantitative assessment results. Compared with histogram matching, Fourier transform, and wavelet transform, the TFJP method has the highest PSNR, the closest mean value and standard deviation to the reference image, and the optimum the fidelity of the spectral curves. Therefore, the TFJP method can reduce stripe noise effectively and preserve the image features simultaneously as much as possible.