<p>Infrared images often suffer from low resolution, blurred textures, and missing details due to inherent imaging principles and sensor limitations, which greatly restrict their applicability in complex environments. To overcome these challenges, this paper introduces a novel infrared image Super-Resolution method - RMEAGAN (Residual Multi-Scale Edge Attention GAN). The proposed approach enhances detail representation and perceptual quality by integrating multi-scale convolution, an edge-guided channel attention mechanism (ECA), and a low-frequency compensation strategy. Specifically, RMEAGAN incorporates a Multi-Scale Edge Attention Enhanced Residual Dense Block (MEAERDB) in the generator, which leverages receptive fields of varying scales to extract rich features while reinforcing high-frequency information through edge attention. Furthermore, residual fusion between neural network outputs and bicubic interpolation results improves structural integrity and thermal feature consistency. To reduce artifacts and noise while improving naturalness and visual coherence, the method jointly optimizes perceptual loss, charbonnier loss, weighted total variation loss and edge loss. Additionally, a network parameter interpolation strategy is introduced to balance perceptual quality and PSNR, harmonizing reconstruction performance between perception-driven GANs and PSNR-oriented networks. Extensive experiments on multiple benchmark infrared datasets show that RMEAGAN outperforms existing state-of-the-art methods, achieving notable improvements in Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index (SSIM) and Learned Perceptual Image Patch Similarity (LPIPS). These results confirm the effectiveness and strong generalization capability of RMEAGAN for infrared image super-resolution tasks.</p>