Research On Spatiotemporal Fusion Algorithm Of Remote Sensing Image Based On Conditional Generative Adversarial Network

Images are one of the important information carriers often used in the research of human life.Similarly,remote sensing image plays a important role in the research of agricultural production and environmental changes.With the development of deep learning,remote sensing image fusion algorithms based on convolutional neural networks have also emerged.The remote sensing image fusion algorithm based on deep learning can automatically learn effective features from the existing data,and its fusion effect far exceeds the traditional fusion algorithm.This paper mainly focuses on exploratory and innovative research on spatiotemporal fusion of remote sensing images based on convolutional neural networks,and has achieved the following results.(1)Spatiotemporal fusion algorithm of remote sensing images based on attention mechanism conditional generative adversarial network: In this paper,a spatiotemporal fusion of satellite images based on attention mechanism conditional generative adversarial network is proposed.An attention mechanism resnet block is designed in the generator network,which can improve the characterization ability of the network from the perspective of channel and space,thereby improving the network’s ability to reconstruct high-frequency and low-frequency information.Taking into account the limitations of hardware conditions,imaging and climate differences between satellite sensors,two conditional generation adversarial network models are designed to establish the relationship between the two data.Finally,through the test on the standard data set,the results of the method proposed in this paper are better than the results of the existing algorithms based on convolutional neural networks.(2)Spatiotemporal fusion algorithm of satellite images via multi-branch structure convolution conditional generative adversarial learning: Based on the conditional generation method against the spatio-temporal fusion of network remote sensing images,in order to allow the network to have a broader field of vision to learn more detailed information,multi-branch convolution is introduced into the discriminator network.Based on the particularity of the optimization training of the generative adversarial network,improving the discriminative performance of the discriminator network can also make the reconstruction result of the entire network better The difference between multi-branch structure convolution(MBC)and ordinary convolution is that the branch structure is used to combine the convolution operation,and this article also uses the dilated convolution method in the branch structure convolution,expand the receptive field of the network without increasing the network parameters,also can make full use of contextual information and improve the ability of feature learning.Experiments show that the reconstructed image has rich details and good visual effects.