Remote Sensing Image Scene Classification Based on Combining Multiple Features

Aiming at the problem that single feature cannot fully express remote sensing scene images, this paper proposes a classification method combining multiple features. Image depth features are extracted by convolutional neural network, and combined with PHOG features. The combined features are classified by random forest classifier. Experiments were carried out on UCM Land-Use and NWPU-RESISC45 of remote sensing image data sets, and the classification accuracy reached 99.29% and 94.63%, respectively, which improved the classification accuracy compared with other existing methods.