深度域自适应目标检测综述OA

Traditional supervised target detection methods based on deep learning rely on a large amount of labeled data and assume that the training and application scene data distributions are consistent.Therefore,performance often declines significantly during cross-domain transfer.Domain adaptation techniques can effectively reduce the dependence on large-scale labeling by alleviating the distribution differences between the source and target domains.Since the introduction of the first unsupervised domain adaptation target detection algorithm in 2018,researchers have successively explored unsu-pervised,few-shot,weakly supervised,unlabelled source domain,test-time,open-set,and general domain adaptation target detection methods to address practical challenges such as data scarcity,dynamic environmental changes,and unknown cate-gories.Existing reviews tend to focus on unsupervised domain adaptation or specific types of algorithms,lacking a sys-tematic summary of various types such as few-shot and weakly supervised approaches.This paper,starting from practical application needs,outlines the adaptation mechanisms and model designs of different methods and analyzes the advantages of various types of algorithms in complex scenarios like privacy protection,continual learning,and rapid deployment,pro-viding researchers with a systematic reference and insight into the development context and method selection of domain adaptation target detection.