Scene Classification of Remote Sensing Images Based on Deep Network Transferability and Image Complexity

In order to solve the low efficient problem of the remote sensing image scene classification, a classification method is proposed which is based on deep network transferability (DNT) and image complexity (IC). Firstly, the pre-trained deep networks are sorted by their transferability. Then, the scenes of the image datasets are sorted by their complexity. Finally, the scenes are classified by the suitable networks according to the deep network transferability and image complexity. The test results in the commonly used remote sensing dataset show that, the classification accuracy of the proposed method is similar to the GoogLeNet network, and the classification time is less than the one of the VGG16 network , thus the classification efficiency is improved obviously. As this method can choose different deep networks suitable for the image complexity, so the classification efficiency is much better.