Remote Sensing Image Super-resolution Reconstruction Method Based On Multi-scale Feature Adaptive Fusion Network

High-resolution remote sensing images have a wide range of applications in areas such as environmental detection,land cover classification,urban economic level assessment,and resource exploration.However,the remote sensing images captured from satellite optical imaging sensors are affected by factors such as relative motion,atmospheric disturbance and noise during transmission.The quality of remote sensing images is limited and it is difficult to meet the application of actual scenes.Image super-resolution reconstruction technology breaks the limitations of hardware equipment and processes,and improves image quality from the software and algorithm level.Therefore,super-resolution reconstruction of remote sensing images is a cost-effective method for obtaining high-resolution remote sensing images.With the rapid development of deep learning in various fields,a large number of image super-resolution reconstruction methods based on convolutional neural networks have emerged and achieved significant results,but most algorithms target general natural images.Because the degradation factors of remote sensing images are more complicated than ordinary digital photos,it is more difficult to infer the high-frequency details of remote sensing images.This paper focuses on reconstructing deeper detailed information in remote sensing images and learning more complex mapping relationships in degraded remote sensing images.We study convolutional neural networks from the perspective of depth and width.The main research results are as follows:(1)The classic shallow network based on the super-resolution reconstruction of convolutional neural network is reviewed.Aiming at the problem that the shallow network has a small receptive field and limited feature extraction,the efficient sub-pixel convolutional network(ESPCN)is improved.A funnel residual block is designed,and multiple non-linear funnel residual blocks are connected in series to replace the traditional super-resolution non-linear mapping part.You can learn more complex mapping relationships,dig deeper image information,and adapt to remote sensing images.Experimental results show that the improved network is better than the original network in terms of objective indicators and subjectivity,and is more suitable for reconstruction of remote sensing images.(2)On the basis of the classic super-resolution algorithm of deep network introduced,in view of the problem that the deep network has a single structure and cannot flexibly adapt to features,this paper proposes an adaptive multi-scale feature fusion network(AMFFN)for remote sensing image super-resolution.Firstly,the features are extracted from the original low-resolution image.Then several adaptive multi-scale feature extraction(AMFE)modules,the squeeze-and-excited and adaptive gating mechanisms are adopted for feature extraction and fusion.Finally,the sub-pixel convolution method is used to reconstruct the high-resolution image.Experiments are performed on three datasets,the key characteristics,such as the number of AMFEs and the gating connection way are studied,and super-resolution of remote sensing imagery of different scale factors are qualitatively and quantitatively analyzed.The results show that our method outperforms the classic methods,such as Super-Resolution Convolutional Neural Network(SRCNN),Efficient Sub-Pixel Convolutional Network(ESPCN),and multi-scale residual convolutional neural networks(MSRN).