Aiming at the problems of fuzzy reconstructed image and loss of high-frequency information caused by low sampling rate of multi-spectral filter array, sparse Raw data and so on, a new eight-spectral filter array distribution scheme is proposed, which is based on neighborhood gradient extension method to reconstruct the spectral Raw image. Firstly, based on the binary tree generation method, an eight-band filter distribution scheme with equal spatial probability ratio is designed in the 4×4 array of repetitive arrays. Then, for the sparse Raw image directly acquired by the sensor, the gradient information of sampling points in each spectrum segment is calculated. On the basis of maintaining the image structure features and texture information, the pixel value and gradient value of sampling points in the neighborhood are used to reconstruct the unsampled points, so as to obtain the complete spectral image information. Finally, based on the reconstructed eight-band spectral images, the pseudo- inverse matrix method is used to reconstruct the 31-band spectral values of each pixel position. The results show that compared with the mainstream image reconstruction methods, the proposed algorithm improves the peak signal-to-noise ratio and composite peak signal-to-noise ratio of reconstructed eight-band spectral images, reduces the mean square error of the spectrum, and better preserves the texture and edges of the image. Reduced artifacts such as color artifacts and image blur in multi-spectral filter array imaging.