高光谱成像
人工智能
异常检测
模式识别(心理学)
计算机科学
分解
计算机视觉
迭代重建
矩阵分解
压缩传感
遥感
光谱特征
目标检测
信号重构
算法
特征提取
异常(物理)
信号处理
全光谱成像
光谱分析
奇异值分解
作者
Dan Li,Yanyan Lv,Kong Fanqiang,Qiang Wang
标识
DOI:10.1109/tim.2026.3697018
摘要
Hyperspectral imaging provides critical information for physical target characterization and anomaly sensing through high-dimensional spectral measurements. However, in practical imaging systems, constrained by sensor bandwidth, storage capacity, and energy consumption, hyperspectral data are often acquired in a compressed measurements. This makes it difficult to obtain accurate and reliable anomaly detection results from such incomplete measurements. Existing methods typically follow a sequential framework, in which compressed sensing reconstruction (CSR) is performed first, followed by hyperspectral anomaly detection (HAD). The two processes hinder their potential for synergistic optimization. To this end, we propose an integrated model called Structured Spectral Dual-Tensor Coupled Decomposition (SSDTCD), which achieves synchronized optimization of CSR and HAD tasks from compressed measurements. The model enhances the joint representational capability among spectral bands through a Structured Coupled Spectral Low-Rank Decomposition (SCSLRD), thereby improving the robustness of sparse reconstruction and the accuracy of HAD. Furthermore, a Dual-Tensor Sparse-Enhanced Decomposition (DTSED) is incorporated to achieve coupled optimization of low-rank structures and sparse components. Experiments on different datasets show that SSDTCD outperforms several state-of-the-art methods in both reconstruction accuracy and HAD precision. Moreover, it maintains stable and reliable sensing capability even under extremely low sampling rates, validating its effectiveness and practical value under constrained measurement conditions.
科研通智能强力驱动
Strongly Powered by AbleSci AI