DBFGAN: Dual Branch Feature Guided Aggregation Network for remote sensing image

Abstract

In Remote Sensing(RS) data analysis, Remote Sensing Change Detection(CD) is an important technology. The existing Remote Sensing Change Detection(RS-CD) methods do not fully consider the advantages and disadvantages of Convolution and Transformer in feature extraction, which will restrict the overall performance of the network to a certain extent. Therefore, this paper proposes a Dual Branch Feature Guided Aggregation Network composed of convolutional neural network(CNN) and Transformer. In the encoding stage, based on the respective characteristics of Convolution and Transformer, a dual-branch backbone network is constructed to extract the spatial information and semantic information of the image respectively; And through the Feature Guidance Aggregation Module, the two branches can guide each other for feature mining, so as to avoid the occurrence of false detection and missed detection of change areas due to insufficient fusion to the greatest extent. Finally, in the decoding stage, the different levels of features extracted by the two branches are fully used for fusion and decoding. And the experiment shows that compared with the existing methods, the mean intersection over union(MIoU) index on the four public datasets are improved by 1.25%, 1.55%, 1.38% and 1.71%.