Scene classification of remote sensing image based on deep network and multi-scale features fusion

Abstract

Owing to the complexity of spatial and structural patterns of remote sensing images, scene classification of them is still an open problem and remains active in the community. The inadequacy of labeled samples leads to low accuracy of remote sensing image scene classification. To solve this problem, a classification method DCNN_MSFF is proposed based on deep convolutional neural networks (DCNN) and multi-scale features fusion (MSFF). Firstly, the remote sensing images are transformed to obtain a number of different scale ones for augmentation. Then, they are input into the DCNN for features extraction, and the different scale features of the convolutional and the fully-connected layers are encoded or pooled averagely. Finally, the processed features are fused, and the multi-kernel support vector machine (MKSVM) is used to classify the scenes. The test results in the commonly used remote sensing datasets show that, this proposed method outperforms the state-of-the-art ones in the scene classification of remote sensing images. In this paper, the multi-scale images and features of the convolutional and the fully-connected layers in the deep learning process are utilized to enhance the representation abilities of the classification features. At the same time, the MKSVM is used to improve the generalization ability of the fusion features, so the classification result is better.