基于DPAM特征提取的GAN合成水管泄漏图像检测分析

打开文本图片集

中图分类号：TP391.4 文献标志码：A 文章编号：1671-5276（2025）06-0169-04

Abstract：For the dificultyinrecognizing realisticimages synthesized by generativeadversative network （GAN），a water pipe leakageimagedetectionmethodbasedonDPAMfeatureextractionisproposed，whichfusesintegratedclassifiertorealize accurate and eficient GANimage recognition.Theresultsshowthat the proposed method，withsmallsample training，has excelentdetectionperformanceandhighstability.Withsuficientnumberofsinglesamples，，thenaturalimagecanbeffectively distinguished fromGANgeneratedimage.And the methodcanefectivelyidentifythepseudo-imagesgeneratedbyStyleGAN， and show high recognition accuracy for other GAN images，with recognition accuracy higher than 99.88% ，which proves the effectivenessofthestrategyoffusingmulti-channelfeature.Theresearchishelpfultoimprovetherecognitionabilityof water pipe leakage，and canbe extended to other image recognition fields as well.

Keywords：water pipe leakage;image detection;image forensics;color channel；feature fusion;generateadversarial network; integrated classifier

0 引言

随着当前网络信息技术的快速进步，许多新型数字处理技术逐步涌现，该技术在输水管道检测装备方面得到了迅速推广，极大推动了数字图像分析方法在人们日常生活中及工业生产过程的应用拓展，由此成为网络信息传递的关键手段，同时也成为人们信息交流的核心技术[1-2]。（剩余4902字）