摘要
当前,弱监督语义分割方法主要聚焦于提高伪标签的质量,往往忽略了不同伪标签之间蕴含的语义相关性。此外,现有生成伪标签方法通常依赖类激活映射技术,未充分考量标签间的潜在语义关系,导致模型在处理对象重叠、视觉干扰和遮挡等复杂情况时,容易出现错误。为有效应对这些问题,提出一种基于标签分布建模与交叉标签约束的弱监督语义分割新方法:通过建立伪标签的标签分布模型,并借助交叉标签约束机制,有效利用标签间的语义关系,优化伪标签的生成流程,提高模型对复杂场景的适应性和泛化能力,生成更准确的伪标签,从而提升弱监督语义分割的性能。实验结果表明,该方法增强了模型在区分不同语义类别时的准确性,在PASCAL VOC 2012和MS COCO 2014数据集上,显著提升了语义分割性能。
The existing weakly supervised semantic segmentation method focuses on the quality of pseudo-labels,overlooking the semantic correlations among different pseudo-labels.In addition,the existing pseudo-label generation uses class activation mapping techniques,giving inadequate consideration to the potential semantic relationships between labels,which leads the model to err when dealing with complex situations such as object overlapping,visual interference,and occlusion.To effectively address these issues,a new method is proposed by establishing a label distribution model of pseudo-labels and using the cross-label constraint mechanism to optimize the pseudo-label generation process,and improve the adaptability and the generalization of the model.More accurate pseudo-labels are generated to enhance the performance of weakly supervised semantic segmentation.The experimental results show that the proposed method enhances the accuracy of the model in distinguishing different semantic categories,and significantly improves the semantic segmentation performance on the PASCAL VOC 2012 and MS COCO 2014 datasets.
作者
迟津生
CHI Jin-sheng(School of Electromechanical Engineering,Dalian Minzu University,Dalian 116600,China)
出处
《南通职业大学学报》
2025年第3期66-71,共6页
Journal of Nantong Vocational University
关键词
弱监督语义分割
标签分布建模
交叉标签约束
类激活映射
伪标签
语义关系
weakly supervised semantic segmentation
label distribution modeling
cross-label constraint
class activation mapping
pseudo-label
semantic relationship
