基于超分辨率图像重建的轻量化目标检测算法研究  

Research on Lightweight Super-resolution Reconstruction Algorithm Based on Progressive Feature Fusion Convolutional Network

作　　者：王超英[1] WANG Chaoying(Electronic Information School,Dongguan Polytechnic,Dongguan 523808,China)

机构地区：[1]东莞职业技术学院,电子信息学院,广东东莞523808

出　　处：《微型电脑应用》2024年第6期57-60,共4页Microcomputer Applications

基　　金：2022年度东莞市科技特派员项目(20221800500732);2022年度东莞职业技术学院国家双高计划电子信息工程技术专业群专项政校行企项目(ZXD202201);2023年度东莞职业技术学院国家双高计划电子信息工程技术专业群专项政校行企项目(ZXD202315)。

摘　　要：利用面向边缘的卷积模块、像素注意力机制和重参数化技术使超分辨率重建算法图像分辨率得到提升,使图像特征细节表现更为优越;利用YOLOv4目标检测算法并结合Focus结构、双向特征金字塔网络和轻量级子通道注意力机制,提高中、低分辨率图像目标检测精度。经实验研究,基于超分辨率重建的轻量化目标检测算法对图像目标具有较好的检测效果,有效提升了图像的检测精度,对提升图像中的细小目标检测精度具有一定的参考意义。This article utilizes edge oriented convolution modules,pixel attention mechanisms,and reparameterization techniques to improve the image resolution of the super-resolution reconstruction algorithm,resulting in superior representation of image feature details.Utilizing the YOLOv4 object detection algorithm and combining the Focus structure,bidirectional feature pyramid network,and lightweight sub channel attention mechanism,the accuracy of object detection in medium and low resolution images is improved.Through experimental research,the lightweight object detection algorithm based on super-resolution reconstruction has a good detection effect on image targets,effectively improves the detection accuracy of images,and has certain reference significance for improving the detection accuracy of small targets in images.

关 键 词：超分辨率重建 多层可分离卷积 特征金字塔网络 注意力机制 

分 类 号：TN929.5[电子电信—通信与信息系统]