The remote sensing images are captured by high-altitude satellites, and the coverage of each image scene is relatively large. It tends to have large intraclass differences and small differences between classes. Using conventional Hierarchical ELM (H-ELM) and multilayer kernel ELM (ML-KELM) to classify remote sensing image scenes, the network structure of the learning model is deep and the parameters are many. This leads to long training time and large memory consumption. In order to solve this problem, based on the ML-KELM, this paper proposes a densely connected kernel ELM (Dense-KELM) learning model, which is used to classify remote sensing image scenes. Experimental results show that at the same model depth, the Dense-KELM model has higher classification accuracy in remote sensing image scenes than the H-ELM and the ML-KELM. Its training time is slightly larger than the ML-KELM but much smaller than the H-ELM. This densely connected learning model can extract high-level features of remote sensing images more effectively, represent the details between remote sensing scenes, and improve the classification accuracy of remote sensing image scenes. Moreover, the densely connected network structure can effectively reduce the number of parameters of the depth model, improve the training speed of the model, and save the storage space of the model.