摘要
针对显著性目标检测中前景与背景分离效果不佳及背景噪声抑制不足的问题,提出了一种加权大核边界增强多尺度RGB-D显著性目标检测网络(LKMNet)。该方法通过引入边界增强加权大核融合模块(BWLKF),结合边界信息与加权大核卷积结构,提升前景聚焦能力并增强边界检测精度。同时,设计了动态门控多尺度融合模块(DGMF),通过自适应门控机制平衡局部与全局信息,从而突出空间相关特征并有效抑制背景干扰。实验结果表明,该方法在四个RGB-D数据集上的检测精度优于现有方法,验证了其在显著性目标检测任务中的优越性能。
To address the challenges of suboptimal foreground-background separation and insufficient background noise suppression in salient object detection,this paper proposed a weighted large-kernel boundary-enhanced multi-scale RGB-D salient object detection network(LKMNet).It designed a boundary-enhanced weighted large-kernel fusion module(BWLKF)to integrate boundary cues with large-kernel convolutions,enhancing foreground focus and boundary localization.In addition,it introduced a dynamic gating multi-scale fusion module(DGMF)to balance local and global features through an adaptive gating mechanism,which highlighted spatially relevant information and suppressed background interference.Experimental results on four benchmark RGB-D datasets demonstrate that LKMNet achieves higher detection accuracy compared to existing methods,confirming its superior performance in salient object detection tasks.
作者
严灵毓
周婷
高榕
叶志伟
Yan Lingyu;Zhou Ting;Gao Rong;Ye Zhiwei(School of Computer Science,Hubei University of Techno-logy,Wuhan 430068,China;Hubei Provincial Key Laboratory of Green Intelligent Computing Power Network,Hubei University of Techno-logy,Wuhan 430068,China)
出处
《计算机应用研究》
北大核心
2025年第12期3815-3822,共8页
Application Research of Computers
基金
国家自然科学基金资助项目(62472149)
湖北省高等学校优秀中青年科技创新团队计划资助项目(T2023006)
湖北省科技计划立项项目(2023BEB024)。
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
RGB-D salient object detection
multi-scale feature fusion
boundary enhancement
weighted large kernel convolution
