In this study, we applies the non-negative matrix factorization (NMF) algorithm to space object image recognition. First, we obtain the sparse NMF algorithm by improving the iterative rules of two traditional NMF algorithms and separately apply the three algorithms to the two-dimension (2D) and (2D) 2 dimensions. Then, we simulate the space optical environment and acquire multiple sets of space-object-scaling model images in the laboratory. After image preprocessing, we establish the training and the testing sample databases, and extract the features of the training samples using different NMF algorithms. Finally, the minimum distance classifier is used to classify the testing samples. The results show that the recognition rates of various NMF algorithms are all above 78%, and the maximum is up to 90%. The experimental results confirm the effectiveness of the proposed algorithm. Compared with the existing methods for space object image recognition, the NMF algorithm is advantageous owing to its high accuracy, fast speed and low resource cost.