可见光条件下伪装目标智能化分割算法综述

Review of Intelligent Camouflaged Object Segmentation Algorithms under Visible Light

对可见光下的伪装目标分割(Camouflage Objects Segmentation,COS)算法的国内外研究现状进行了梳理。首先,将COS的26个模型分为多阶段学习、联合辅助任务和引入额外先验信息3类,阐述了这3类方法的核心思想及各自的优势和不足;其次,详细介绍了COS常用的4个评价指标和8个公共数据集;然后,在3个公共数据集上对这些模型进行定量、定性的分析评估;最后,介绍了COS的主要应用领域,指出了现有算法的局限性,并对COS未来的研究方向进行了探讨。

The research status of camouflaged object segmentation(COS)methods under visible light at home and abroad were summarized.Firstly,the 26 models of COS were divided into three categories:multi-stage learning,joint auxiliary tasks and additional prior information in-troduction.The core ideas,advantages and disadvantages of the three types of methods were elaborated.Secondly,four commonly used evaluation indicators and eight common datasets in COS were introduced in detail.Then,quantitative and qualitative analysis of these models was conducted on three public datasets.Finally,the main application fields of COS were introduced.Accordingly,limitations of existing COS algorithms were pointed out and the future research di-rections of COS were explored.