软件导刊2025，Vol.24Issue(5)：214-220,7.DOI:10.11907/rjdk.241276
基于特征复用的轻量级图像超分辨率重建
Lightweight Image Super-Resolution Reconstruction Based on Feature Reuse Method
摘要
Abstract
Increasing the depth and parameter count of the image super-resolution reconstruction network model is considered the key to achieving excellent image reconstruction performance,but it greatly increases its complexity and computational cost.Therefore,a lightweight image super-resolution reconstruction algorithm was studied,which achieves feature reuse through multi-level skip connections,effectively improving the utilization and propagation speed of feature information.Firstly,in order to improve the feature extraction efficiency of convolu-tion,a fast convolutional layer combining depthwise separable convolution was designed,greatly reducing the number of convolution kernels required for feature extraction;Secondly,the introduction of a decoupled fully connected attention module enhances the global information perception capability of the network model with minimal parameter count;Finally,the fusion mechanism of perceived features is compared to better preserve high-frequency information in the image,resulting in a reconstructed image with better visual effects.Experiments have shown that with only 636 KB parameters,the proposed model outperforms most current lightweight image super-resolution models,demonstrating the effectiveness of the research method and providing new ideas for the lightweighting of deep neural network models.关键词
图像超分辨率重建/轻量级/特征复用/解耦合全连接注意力/对比感知的特征融合Key words
image super-resolution reconstruction/lightweight/feature reuse/decoupled full-connected attention/contrast aware feature fusion分类
计算机与自动化引用本文复制引用
林子杰,郭恒,葛芳艳..基于特征复用的轻量级图像超分辨率重建[J].软件导刊,2025,24(5):214-220,7.基金项目
四川省科技计划项目(2023YFQ0072) （2023YFQ0072）
