Remote sensing image classification based on dual-channel deep dense feature fusion.

In order to improve the effective utilization of features in remote sensing image scene classification and achieve the purpose of improving the accuracy of remote sensing image classification, a remote sensing image classification method based on dual-channel depth-dense feature fusion was used for theoretical analysis and experimental verification. First, the image convolution layer features and fully connected layer features was separated extracted by constructing a composite dense convolutional network model. In order to exploit the deep information of the image, the deep convolutional layer features extracted by the model were recombined and encoded by the bag of visual words to capture the deep local features of the image. Finally the linear and weighted methods were used to fuse local and global features and then classify them. The results show that using the datasets UC Merced Land-Use and NWPU-RESISC45 for experiments, the classification accuracy obtained is 93.81% and 92.62%, respectively. This method makes full use of the complementarity of local features and global features to achieve the full expression of deep image information. [ABSTRACT FROM AUTHOR]