To overcome the shortcomings of the classic collaborative sparse unmixing algorithm and the edge blur problem caused by the total variation regular term, considering the importance of sparsity and spatial information to improve the accuracy of unmixing, a novel algorithm called superpixel and low rank for collaborative sparse unmixing was proposed. The unmixing algorithm was theoretically analyzed and experimentally verified. The superpixel segmentation was performed on the hyperspectral images, and then collaborative sparsity constraints were imposed on each superpixel. In addition, a low-rank regular term was used instead of the traditional total variation regular term to utilize spatial information. A set of simulated data and a set of real data were selected for experiments. These results show that the signal reconstruction error obtained in the simulated experiment is 19.4 when the signal-to-noise ratio is 30dB, which is about 35% higher than that of the classic sparse unmixing via variable splitting augmented Lagrangian and total variation algorithm. Real data experiment intuitively reflects that the algorithm can effectively overcome the problem of edge blur. The proposed algorithm has better unmixing performance. This research provides a reference for how to use sparsity and spatial information comprehensively.