A Restricted Embedding Transfer Model for Hyperspectral Anomaly Detection

doi:10.1109/ICBASE59196.2023.10303063

OPT OpenIR > 光谱成像技术研究室

	A Restricted Embedding Transfer Model for Hyperspectral Anomaly Detection
	Shi, Chenliang 1; Qiu, Shi2
	2023
会议名称	4th International Conference on Big Data and Artificial Intelligence and Software Engineering, ICBASE 2023
会议录名称	2023 4th International Conference on Big Data and Artificial Intelligence and Software Engineering, ICBASE 2023
页码	340-348
会议日期	2023-08-25
会议地点	Hybrid, Nanjing, China
出版者	Institute of Electrical and Electronics Engineers Inc.
产权排序	2
摘要	The purpose of hyperspectral image anomaly detection is to overcome the problem of inconsistent background distribution, suppress background information as much as possible, and highlight anomalous target information. Many existing deep learning anomaly detection methods use generative algorithms, such as those based on generative adversarial networks and those based on automatic encoders, but these algorithms are inevitably accompanied by the problem of low reconstruction accuracy or poor calibration. In order to solve these problems, this paper proposes a restricted embedding transfer model for hyperspectral image anomaly detection, which transforms the anomaly detection problem into a feature regression problem through partial knowledge transfer learning. Thus it avoiding the need for reconstruction or probability distribution evaluation. The teacher network adaptively generates descriptive embedding vectors that are used as pseudo-labels to assist the training of the student network, and only part of the normal sample is needed to complete the training. In the experimental part, the performance of the proposed method is compared with seven existing methods on twelve hyperspectral datasets. The results show that the proposed method has better detection effect, and the AUC index reaches 0.9789, which is 0.0109 higher than the second place. © 2023 IEEE.
关键词	image processing anomaly detection hyperspectral image deep learning transfer learning semi supervised learning
作者部门	光谱成像技术研究室
DOI	10.1109/ICBASE59196.2023.10303063
收录类别	EI
ISBN号	9798350329490
语种	英语
EI入藏号	20234915151193
引用统计
文献类型	会议论文
条目标识符	http://ir.opt.ac.cn/handle/181661/97046
专题	光谱成像技术研究室
作者单位	1.Shaanxi Normal University, Chinese Academy of Sciences, Xi'an Institute of Optics & Precision Mechanics, Xi'an, China; 2.Xi'an Institute of Optics & Precision Mechanics, Chinese Academy of Sciences, Xi'an, China
推荐引用方式 GB/T 7714	Shi, Chenliang,Qiu, Shi. A Restricted Embedding Transfer Model for Hyperspectral Anomaly Detection[C]:Institute of Electrical and Electronics Engineers Inc.,2023:340-348.

条目包含的文件
文件名称/大小	文献类型	版本类型	开放类型	使用许可
A Restricted Embeddi（5842KB）	会议论文		限制开放	CC BY-NC-SA	请求全文