基于边界增强和加权大核的多尺度 RGB-D显著性目标检测

打开文本图片集

关键词：RGB-D显著性检测；多尺度特征融合；边界增强；加权大核卷积 中图分类号：TP391 文献标志码：A 文章编号：1001-3695（2025）12-037-3815-08 doi：10.19734/j.issn.1001-3695.2025.03.0100

Multi-scale RGB-D salient object detection based on boundary enhancement and weighted large kernel

YanLingyua，b，Zhou Tinga，b，Gao Ronga，b，Ye Zhiweia，bt （a.Schoolofutece，ubroalKboaoryfeetellgenCompuineekubeiUnesit logy，Wuhan 430068，China）

Abstract：Toaddressthechallenges of suboptimal foreground-backgroundseparationand insufficientbackground noisesuppressioninsalientobjectdetection，thispaper proposedaweightedlarge-kernelboundary-enhancedmulti-scaleRGB-Dsalient objectdetectionnetwork （LKMNet）.It designedaboundary-enhanced weighted large-kernel fusion module（BWLKF）to integrateboundarycueswithlarge-kernelconvolutions，enhancingforegroundfocusandboundarylocalization.Inadition，itintroduced adynamic gating multi-scalefusion module（DGMF）tobalance localandglobal featuresthrough anadaptivegating mechanism，whichhighlightedspatiallyrelevant informationandsuppressedbackground interference.Experimentalresultson four benchmark RGB-Ddatasets demonstrate thatLKMNetachieves higherdetectionacuracycompared to existing methods， confirming its superior performance in salient object detection tasks.

Keywords：RGB-Dsalientobjectdetection；multi-scalefeaturefusion；boundary enhancement；weightedlarge kernelconvolution

0 引言

显著性目标检测（salientobjectdetection，SOD）是计算机视觉领域的重要研究课题，其目的是模拟人类视觉注意机制，识别并提取图像中最显著的目标或区域。（剩余19535字）