Remote Sensing Image Registration based on Attention and Residual Network
Ying Chen, Jineng Li, Wang Dongzhen · 2020
Multi-view remote sensing registration has important applications in ground target recognition, image aided navigation, missile image guidance and so on. In order to improve the registration accuracy of remote sensing image with change of view angle, a registration method combining attention mechanism and residual network is proposed. The residual network is used as the backbone structure of feature extraction to improve the abstract ability of the model for complex features. At the same time, the attention mechanism based on channel and spatial is introduced into the feature extraction network to improve the distinguish and location ability of the model to image features. Finally, in the feature matching stage, a mutual correlation operation that improve the performance of feature matching is proposed. Compared with the comparison method, the registration accuracy is improved by 10% on average, Experiments show that this method improves the accuracy of multi view remote sensing image registration.