In order to improve the performance of remote sensing scene zero-shot classification by utilizing the complementarity among different image features, a zero-shot scene classification algorithm based on the fusion of image features is proposed, which combines the fusion of image features with zero-shot classification. The analysis dictionary learning is exploited to obtain the sparse coefficients of image features. The obtained sparse coefficients are concatenated as the fused image feature to reduce the redundant information and retain the characteristics of different image features. Supervised information is introduced to improve the discriminability of the model. The fused image feature is structurally aligned with the semantic word vectors to improve the transfer capability for the unseen class scenes. The fusions of image features at the same level and the different levels are carried out on UC-Merced (UCM) and Aerial Image Dataset (AID) remote sensing scenes datasets, respectively. The experimental results show that the overall classification accuracy and time-consuming of our method are superior to those of other zero-shot classification algorithms and general feature fusion algorithms, which proves the effectiveness of our method.