Abstract: In order to avoid that the problems of the existing hyperspectral image denoising optimization model only considers the limited intrinsic structure characteristics, and does not realize the accurate representation of image features, a denoising method based on spatial spectral depth image prior and smoothing is proposed. The model combines tight frame transform with a deep learning model with high expression and strong learning ability. Firstly, on the basis of low-rank matrix decomposition, the potential-spatial spectral features were learned by using specific depth images prior. Secondly, a tight frame of end and abundance matrix was constructed respectively to explore the local smoothing of the empty spectrum and solve the semi-fitting behavior of the depth image prior. Finally, an efficient iterative algorithm was designed to solve the model. The results show that the method based on space spectrum depth image prior has better performance under various complex noise interference, and the peak signal-to-noise ratio(PSNR) is improved by at least 1 dB, and high quality restored images are obtained. The method provides a reference for hyperspectral image denoising.