Super Resolution Reconstruction Of Remote Rensing Images Based On Convolutional Neural Networks

As an important data of surface information,high-quality remote sensing images have been widely used in many fields such as meteorological observation,disaster monitoring,military reconnaissance,urban planning and so on.However,in the actual remote sensing imaging process,due to factors such as long target distance,complex scenes,and limitations of imaging equipment,the obtained remote sensing images usually have problems such as low resolution,local area blur,and geometric structure distortion.As one of the popular research directions in the field of image processing,super-resolution reconstruction technology can break through the limitations of hardware equipment,improve the resolution of remote sensing images,and restore the details of images.With the rapid development of deep learning,image super-resolution reconstruction methods based on convolutional neural networks continue to emerge,further improving the quality of remote sensing image reconstruction.Therefore,aiming at the complex problems encountered in the process of remote sensing image reconstruction,this paper proposes two remote sensing image super-resolution reconstruction algorithms based on convolutional neural network.The specific research contents are as follows:(1)Aiming at the problems of insufficient feature extraction of remote sensing images with complex structures and loss of high-frequency details by convolutional neural networks,a remote sensing image super-resolution reconstruction algorithm based on multi-scale and attention mechanism is proposed.The algorithm uses the improved Inception-Res Net module to extract features at different scales of the image,obtains global features and important structural information by expanding the receptive field of the network,and fuses the output of each module again to improve feature utilization.The convolutional block attention combining channel and spatial attention module is used to establish the relationship between different regions and spaces of the image,improve the ability to learn and represent global and local features,and then highlight the high-frequency information such as edges and textures of remote sensing images.According to the superiority of the residual structure,an adaptive residual network is added to the multi-scale feature extraction module to supplement the feature information,enhance the richness of the features,and use the global residual to accelerate the network convergence and improve the fidelity of the reconstructed remote sensing image.The experimental results show that the algorithm can extract more high-frequency features,and the reconstructed image presents a good visual effect,and has a significant improvement in peak signal noise ratio and structural similarity.(2)The traditional convolutional neural network has a simple structure and a small receptive field,which limits the acquisition of high-level semantic features of urban environmental remote sensing images with high spatial distribution.To solve this problem,a dual-channel remote sensing image super-resolution reconstruction algorithm based on self-calibration convolution is proposed.The feature mapping module of the algorithm consists of a local dense connection network and a global residual attention network.The densely connected network makes full use of the local shallow features of the image by reusing the output of each layer of the network,and uses coordinate attention to effectively obtain position information and integrate spatial coordinate information into attention mapping.The introduction of self-calibrated convolution in the global residual attention network expands the receptive field of the network,helps to capture sufficient high-level semantics,and enhances the effective features of remote sensing images.The dual-channel feature extraction structure makes full use of the complementarity of local features and global features of the image,and improves the utilization of feature information in remote sensing images.The FRe LU activation function is used in the network structure to further extract the fine spatial layout of the target in the urban remote sensing image.Sub-pixel convolution is used as the reconstruction module to reconstruct the feature map after recombination to avoid the introduction of invalid information.The experimental results show that the algorithm has better recovery effect on the geometric structure and edge texture of buildings in urban environmental remote sensing images.Compared with the comparison algorithm and the remote sensing image super-resolution reconstruction algorithm based on multi-scale and attention mechanism proposed in this paper,the objective evaluation index of the algorithm is further improved.