BMEDT:基于Transformer的双向多级边缘检测网络  

Bidirectional Multi-level Edge Detection Network Based on Transformer

作　　者：荆东东 李备备[1,2,3] 王诗宇 刘信君[1,2] JING Dongdong;LI Beibei;WANG Shiyu;LIU Xinjun(Shenyang Institute of Computing Technology,University of Chinese Academy of Sciences,Shenyang 110168,China;University of Chinese Academy of Sciences,Beijing 100049,China;Shenyang CASNC Technology Co.,Ltd.,Shenyang 110168,China)

机构地区：[1]中国科学院沈阳计算技术研究所,沈阳110168 [2]中国科学院大学,北京100049 [3]沈阳中科数控技术股份有限公司,沈阳110168

出　　处：《小型微型计算机系统》2024年第12期3042-3049,共8页Journal of Chinese Computer Systems

基　　金：辽宁省自然科学基金项目(2023-BS-214)资助;沈阳市中青年科技创新人才支持计划项目(RC201488,RC210487)资助。

摘　　要：针对现有边缘检测方法在处理复杂背景时出现的边缘丢失、线条粗糙、内容混乱等问题,本文提出一种基于Transformer的双向多级边缘检测网络.首先,网络采用Swin Transformer作为骨干网络,提取图像基本特征,并通过多尺度特征增强模块获取特征图中不同尺度的上下文信息;其次,根据特征点位置的重要性,提出边缘注意力模块,用于强化边缘与非边缘像素之间的差异性,提高重要特征的关注度;最后,为解决边缘检测任务中正负样本不均衡问题,提出一种新的损失函数,并通过分层监督的方式强化特征表示.在BSDS500和NYUDv2数据集上的实验和评估中,所提方法在边缘细节处理方面已经超过许多经典算法,并取得了最佳水平.这些结果表明,所提方法可以有效地解决边缘检测中的一些挑战,并为相关应用提供更好的支持.Aiming at the problems of the existing edge detection methods such as loss of details,rough lines,and content confusion when dealing with complex backgrounds,this paper proposes a bidirectional multi-level edge detection network based on Transformer.First,the network uses Swin Transformer as the backbone network to extract images′basic features,and obtain the context information of different scales in the feature map through the multi scale feature enhancement;Secondly,according to the importance of the feature point position,an edge attention is proposed to strengthen the difference between edge and non-edge pixels,to increase the attention of important features;Finally,in order to solve the problem of the imbalance between positive and negative samples in the edge detection task,a new loss function is proposed,and the feature representation is strengthened through hierarchical supervision.In the experiments and evaluations on the BSDS500 and NYUDv2 datasets,the proposed method has surpassed many classical algorithms in terms of edge detail processing ability,and achieved the best level.These results show that the proposed method can effectively solve some problems and challenge in edge detection,and provide better support for related applications.

关 键 词：边缘检测 TRANSFORMER 多尺度特征增强 边缘注意力 分层监督 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]