Depthwise Separable Residual Network for Remote Sensing Image Scene Classification

Abstract

In view of the large amount of parameters and slow running speed of the existing remote sensing image scene classification models, as well as the tendency to over-fit the model when the training samples are limited, proposes a scene classification model based on depthwise separable residual network. Firstly, based on the idea of residual learning, the model combines two-dimensional convolution and separable convolution to construct a residual separable feature extraction module (RSFE), which can reduce parameters of the model. Then, the module is used as the basic structure to construct a deep feature extraction network model. Finally, the extracted features are input to the softmax classifier for classification. The experimental comparisons between proposed method and other methods are carried out on the UC Merced and NWPU45 datasets. The results show that the classification accuracy of the proposed model is improved to 99.52% in the UC Merced dataset, and 92.46% in the NWPU45 dataset, respectively. This model has more advantages in the task of scene classification.