Lightweight Image Super-resolution Network Based on CNN-Transformer

To address the challenges of high parameter count and excessive computational complexity in existing Transformer-based models for single image super-resolution(SISR), we propose a lightweight SISR model, SEST(Super-resolution Enhanced Spatial Transformer). The model incorporates a multi-head attention mechanism and a feature segmentation module, significantly reducing memory requirements, while enhancing texture detail recovery through a local feature mixer and an enhanced spatial attention mechanism. Experimental validation shows that SEST outperforms existing lightweight SISR models on multiple benchmark datasets. Specifically, it achieves a better balance in peak signal-to-noise ratio(PSNR), structural similarity(SSIM), and parameter count, providing a viable solution for image enhancement in resource-constrained environments.