端元
高光谱成像
自编码
人工智能
模式识别(心理学)
计算机科学
概率逻辑
生成模型
子空间拓扑
特征(语言学)
阿卡汀
特征学习
比例(比率)
数学
编码器
光谱特征
贝叶斯概率
作者
Shu Xiang,Xiaorun Li,Shuhan Chen
出处
期刊:
[Institute of Electrical and Electronics Engineers]
日期:2025-08-03
卷期号:: 7338-7341
被引量:1
标识
DOI:10.1109/igarss55030.2025.11242593
摘要
Recently, many unmixing autoencoders aimed at addressing spectral variability have been proposed. However, most autoencoders tend to adopt probabilistic generative models and lack targeted constraints, which can lead to unstable outputs. Endmember bundles offer a reliable unmixing framework that considers endmember variability. In this paper, we introduce a novel bundle-based autoencoder that incorporates endmember bundles to address spectral variability. To construct endmember bundles, we propose a directed subspace Transformer, which efficiently explores effective information from each endmember category. Based on bundle-specific features, the network combines intra-class weights with the endmember bundle decoder to generate endmembers under the influence of spectral variability. Additionally, the multi-feature aggregation abundance encoder integrates features from multiple endmembers and estimates abundances. Extensive experiments validate the effectiveness of the proposed algorithm.
科研通智能强力驱动
Strongly Powered by AbleSci AI