Single Image Super-Resolution Reconstruction Based on Multilayer Residual Networks With Attention Mechanisms

Small and medium-sized unmanned aerial vehicles (UAVs) often experience loss of texture detail and edge blurring in aerial images due to limitations in flight altitude and payload capacity. To enhance image quality, this paper proposes a multilayer residual network super-resolution reconstruction method that utilizes an attention mechanism. This method effectively extracts high-frequency feature information from images and improves detail recovery through a nested residual network structure and a channel attention mechanism. The network employs a deeply parallel architecture to fuse multilayer features, further enhancing reconstruction accuracy. Experimental results demonstrate that the proposed method outperforms existing state-of-the-art techniques in terms of peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) metrics on the Set5, Set14, and various open-source aerial imagery datasets. The reconstructed images exhibit significantly improved texture details and edge clarity. This method offers an effective solution for the high-quality reconstruction of UAV aerial images. We have named this method MARNet (Multilayer Attention Residual Network), which utilizes nested residual blocks and channel attention to achieve high-fidelity reconstruction.