摘要
Abstract
During the process of drone image acquisition,external environmental factors and internal system influences often result in low image clarity,blurred detail information,and significant feature degradation,making target recognition difficult.To address these issues,this paper proposed a high-resolution reconstruction method for drone images based on deep feature embedding.The method first used style-based generative adversarial network 2(StyleGAN2)as a feature embedding space.Drone images were input into the network,where the features were extracted,controlled,and trained layer by layer,optimizing and constraining image edges and detailed features,resulting in high-resolution reconstructed images.Additionally,a loss function was introduced to preserve more of the original image's texture and detail information.Experimental results show that,under the condition of ensuring that the Q correlation coefficient is close to the optimal value,this method effectively achieves high-resolution reconstruction of drone images.关键词
深层特征嵌入/无人机图像/高分辨率重建/基于样式的生成对抗网络第二版(StyleGAN2)/纹理细节Key words
deep feature embedding/drone images/high-resolution reconstruction/style-based generative adversarial network 2(StyleGAN2)/texture detail分类
天文与地球科学
