Aiming at the problem of weak robustness and low accuracy of convolutional neural networks in remote sensing image registration,a registration algorithm combining residual network and Transformer and integrating quadruple attention was pro-posed.A hybrid network structure of convolution and remote sensing context Transformer was proposed,in which the residual network residual block was replaced and it was combined with other pre-trained convolutional layers for feature extraction.The quadruple attention mechanism was integrated into the feature extraction network to improve the discriminatory feature represen-tation of remote sensing images.A bidirectional matching network was designed,and the Pearson cross-correlation algorithm was used to establish the correspondence between remote sensing images.Experimental results show that the model is better than other models under multiple evaluation indexes of remote sensing image registration.
维普
万方数据