Satellite image classification is a meaningful task, such as helping recognize the damaged houses. Traditional methods like ANN and BoW are not good at satellite image classification. CNN(Convolutional Neural Network) has been generally used to do image classification, and it can be applied to many daily situations like damage assessment. In this paper, we build a CNN model and discuss its advantage over other present models. Besides, to increase the robustness of our model, we add image enhancement to our training data. The preliminary model can reach an accuracy of 0.9706 in our experiment.