It is difficult to determine the size and number of sub-blocks in a hyperspectral image using the existing methods because of the low rank of the sub-blocks and other associated disadvantages. Therefore, we propose a hyperspectral image denoising method, which combines the features of the ground objects with the low-rank characteristics. Further, the number of sub-blocks are divided with respect to the number of categories of prior knowledge of ground object data, and optimal parameters are specified for determining the size of the blocks. Then, the low-rank characteristics of the same object space spectrum are obtained based on the correlation of the pixel space and spectrum with respect to the same feature. Finally, the spectral low-rank characteristics of the entire hyperspectral image are combined, and the noise-reduced image is obtained according to the low-rank matrix recovery model. Experiments conducted on the Washington DC Mall and Indian Pines datasets demonstrate that the proposed method not only improves the noise reduction effect with respect to each type of ground noise but also targets mixed noise containing more severe random noise and sparse noise.