A dilated CNN model for wide-band remote sensing image classification

In this paper, a model of dilated convolutional neural network (CNN) is proposed. Dilated convolution is used to instead the convolution layer in the traditional CNN and to realize a faster and more accurate image classification model. In the experiment, the remote sensing image of the earth's terrain, including ocean, desert and city et al. taken by the wideband imaging spectrometer onboard Tiangong-2 was used as the training and testing set to record the accuracy and training time, and compared with the traditional CNN. The results show that the dilated CNN model performs well in the wide-band remote sensing image data set and achieves better accuracy with less computational resources consumed.