基于跨尺度特征融合的内窥镜图像增强算法

打开文本图片集

DOI：10.16652/j.issn.1004-373x.2026.01.006引用格式：，，.基于跨尺度特征融合的内窥镜图像增强算法[J].现代电子技术，2026，49（1）：34-40.

关键词：内窥镜图像；深度特征融合；CFF；曝光异常；图像增强算法；金字塔池化模块中图分类号：TN911.73-34；TP394.1；TH691.9 文献标识码：A 文章编号：1004-373X（2026）01-0034-07

Endoscopic image enhancement algorithm based on cross-scale feature fusion

LIUXuyang1²，CAI Yun1，2 ，JIANGLin1，2，

（1.KLb 2.HubeiKeyLbaoalsdfid; 3.InstituteofRoboticsand InteligentSystems，WuhanUniversityofScienceandTechnology，Wuhan430o81，China）

Abstract：Theclinicalendoscopicimageoftensufersfromlow-qualityimagingduetounevensupplementary lightsources andreflectionsfromhumantisse mucus，resultinginpoorimagequalityforinstane，alargequantityofoverexposure.However， thecurrentdeep learning based imageenhancement algorithms havelowfeature extractioncapabilitiesduetofixed-sizefeature fusion，whichleadstopoorehancementefects.herefore，anedoscopicimageeancementalgorithmbasedoncrosscale featurefusionisproposed.Inthealgorithm，aconvolutionmodule（CM）isconstructedforhighperformancefeatureextractionand aspatial pyramidpoling-fast（SPF）moduleisusedtorealizethepoolingoperationoffeature mapswithdiffrent scales. Aditionally，cross-scalefeaturefusion（CFF）moduleisitroducedintodiferentsalesofetworklayrstochevemulti-scale featurefusionandcontextinformationpropagation，soastoimproveimagedetailcaptureandimageqality.Experimentalresults showthattheproposedalgorithmoutperforms theexistingalgorithmsinPSNRand SSIM，inwhichthePSNRisimproved by （20 9.9% ，and the SSIM by 15.4% ，achieving high-quality endoscopic image enhancement.

Keywords：endoscopic image;deepfeature fusion; CFF;exposureanomaly;image enhancement algorithm;pyramid pooling module

0 引言

图像曝光的一致性是获取高质量医学影像的关键因素之一。（剩余10681字）