Xu Chenguang
Jiangxi Province Key Laboratory of Water Information Cooperative Sensing and Intelligent Processing, School of Information Engineering, Nanchang Institute of Technology, Nanchang, 330099, China.
National and Local Joint Engineering Research Center of Fruit Intelligent Photoelectric Detection Technology and Equipment, East China Jiaotong University, Nanchang, 330013, China.
Sci Rep. 2024 Oct 10;14(1):23705. doi: 10.1038/s41598-024-70395-6.
Hyperspectral sparse unmixing, an image processing technique, leverages a spectral library enriched with endmember spectral information as a prerequisite. It decomposes the hyperspectral data to ascertain the abundance corresponding to each endmember in the spectral library. Currently, the majority of sparse unmixing methods are inadequate in high-noise environments due to their failure to comprehensively account for hyperspectral characteristics. Addressing this challenge, this paper introduces an innovative sparse unmixing approach for hyperspectral images named spectral weighted sparse unmixing based on adaptive total variation and low-rank constraints (SWSU-ATVLR). Initially, the sparse unmixing algorithm is introduced in detail. Subsequently, we present the our method. This method seamlessly integrates the low-rank, adaptive TV and spectral weighting characteristics of hyperspectral data. While preserving the low-rank attributes and sparsity of abundance, the adaptively adjusted abundance matrix exhibits a regularized horizontal and vertical difference ratio across various structures and fully utilizes spectral information, enhancing denoising efficiency. Subsequently, the ADMM algorithm is employed to solve the new model. To validate our proposed algorithm, SWSU-ATVLR method is compared and analysed in detailed experiments with several current state-of-the-art methods through simulated and real data experiments. Experimental results prove that our proposed method is superior to these state-of-the-art methods.
高光谱稀疏解混是一种图像处理技术，它以一个富含端元光谱信息的光谱库为前提条件。它对高光谱数据进行分解，以确定光谱库中每个端元对应的丰度。目前，大多数稀疏解混方法在高噪声环境下存在不足，因为它们未能全面考虑高光谱特征。针对这一挑战，本文介绍了一种用于高光谱图像的创新型稀疏解混方法，即基于自适应全变差和低秩约束的光谱加权稀疏解混（SWSU - ATVLR）。首先，详细介绍了稀疏解混算法。随后，阐述了我们的方法。该方法无缝集成了高光谱数据的低秩、自适应全变差和光谱加权特性。在保留丰度的低秩属性和稀疏性的同时，自适应调整后的丰度矩阵在各种结构上呈现出正则化的水平和垂直差异比，并充分利用光谱信息，提高了去噪效率。随后，采用交替方向乘子法（ADMM）算法求解新模型。为验证我们提出的算法，通过模拟和真实数据实验，将SWSU - ATVLR方法与当前几种先进方法在详细实验中进行了比较和分析。实验结果证明，我们提出的方法优于这些先进方法。
