Lightweight Image Super-resolution Reconstruction Model Based on Improved Multi-scale Feature Pyramid Network
XU Guangyu
WU Min
Abstract：To address the challenges of excessive parameter size and training difficulties caused by complex neural network architectures in image super-resolution(SR)tasks,a lightweight image SR reconstruction model based on improved multi-scale feature pyramid network(MFPNet)was proposed.First,an improved MFPNet architecture was designed.The feature representation spaces at different scales were constructed through iterative downsampling operations,which effectively enhanced the network's ability to capture multi-granularity detail features of images.Second,position aware circular convolution(ParC)was adopted as the primary feature extraction module,reducing the parameter count while expanding the network's receptive field size.Finally,a dynamic attention block(DAB)was developed.Through an attention guidance layer(AGL),the weighting of efficient channel attention(ECA)and spatial attention(SA)modules was dynamically adjusted,improving the network's capability for restoring texture details.The experimental results demonstrated that,compared with other state-of-the-art models,the structural similarity index measure(SSIM)maximum of 0.9613 and the peak signal-to-noise ratio(PSNR)maximum of 38.11 dB were achieved by improved MFPNet model.This research confirmed that the improved MFPNet model could be used to image reconstruction tasks with more natural detail textures.
Keywords：convolutional neural networksuper-resolution reconstructionattention mechanismlightweightmulti-scale feature pyramid
Publication Date：2025-09-20
Online Publishing Date：2025-09-24(First online date of this platform, not the publication date of the document)
Pages：7( 334-340 )