摘要
Abstract
High-resolution images contain a wealth of detailed information,and there is a correlation between different image features.This correlation may lead to information overlap during the feature selection process,resulting in the extrac-tion of only partial features,which in turn affects the accuracy of subsequent surveying image reconstruction.To address this issue,this paper proposed a high-resolution adaptive reconstruction method for surveying images based on dual-domain neu-ral networks.The channel attention mechanism was integrated into the transform module to extract shallow features of survey-ing images.A spatial dual-domain fusion attention mechanism was then used to identify the deep features of surveying images,completing deep feature extraction based on the dual-domain neural network.By integrating the shallow and deep features of surveying images,the method utilized global residual learning to fuse the features of the reconstructed surveying image.The adaptive reconstruction of the surveying image was completed through a subpixel convolutional layer,resulting in high-quality surveying images.Experimental results show that the reconstructed high-resolution surveying images have a high resolution,with the similarity of the final reconstructed high-resolution images greater than 0.87,peak signal-to-noise ratio(PSNR)greater than 98.15,and color difference lower than 0.2.These results further demonstrate that the visual qual-ity of the reconstructed surveying images is significantly improved.关键词
双域神经网络/全局残差/注意力机制/高分辨率图像/变换(Transform)模块/自适应重建Key words
dual-domain neural network/global residual/attention mechanism/high-resolution image/transform module/adaptive reconstruction分类
计算机与自动化
