Multi-temporal remote sensing image registration method based on double-view full convolutional neural network

The invention relates to a multi-temporal remote sensing image registration method based on a double-view full convolutional neural network, which is characterized in that transfer learning is combined, a network branch AlexNet network is added on the basis of an original pre-trained VGG16 network, the VGG16 network pays attention to the local features of an image, the AlexNet network pays attention to the global features of the image, and the AlexNet network is improved by adopting a dense connection structure. Feature reuse is improved; a double-branch network is adopted to improve a grid loss function, an original single-branch network grid loss function is improved according to four parameters obtained through regression of the two networks, errors of the single-network grid loss function are avoided, low-level semantic information and high-level semantic information are transmitted mutually, and the accuracy of the network loss is improved. Therefore, feature fusion of the low-level semantic information