Remote sensing image registration based on global sensing convolution and feature fusion

To address the problem of poor registration performance in current remote sensing image data volume and complex scenes, we propose a DenseNet model based on global-aware convolution blocks and adaptive dual-feature fusion blocks for improving the accuracy of remote sensing image registration. The method captures global features by replacing part of the dense blocks with position aware circular convolution, using the global kernel and circular convolution. Meanwhile, an adaptive dual-feature fusion attention mechanism is introduced to integrate multi-level features to obtain higher registration accuracy. First, a modified DenseNet is used to extract feature information from the image, followed by a bidirectional matching relationship obtained by a bidirectional correlation matching network, and the final parameters are synthesized by weighting the bidirectional parameters obtained by regression, and finally, the image registration is completed by the affine transformation. The experimental results show that this method can effectively improve the accuracy of remote sensing image registration and outperforms other existing methods in terms of performance.