There are hundreds of bands in hyperspectral remote sensing image and the bands are highly correlative, difficulty of analysis can be reduced by using fusion technology. A novel structure of second generation wavelet weighting fusion algorithm is proposed. Firstly, image is decomposed into two serials, which are predicted and updated by two-order Neville filter on rectangle and quincunx grid by turns. Secondly, feature level second generation wavelet fusion is carried on the updated serial by using the variance of each band as the feature of fusion. Finally the image is reconstructed by reverse second generation wavelet. In order to test the effect of the new method, hyperspectral remote sensing image of airborne visible and infrared imaging spectrometer is simulated on Pentium IV computer. Compared with typical fusion method such as principal component analysis and discrete wavelet transform, the result of the experiment shows that the proposed method can retain spatial and spectral feature, the entropy is bigger than principal component analysis 0.1949, and bigger than discrete wavelet transform 0.7998.