基于光易变性低秩正交先验的高光谱解混

Hyperspectral unmixing based on low rank orthogonal priors for spectral variability

高光谱解混是通过图像分解提取端元及丰度特征的过程,然而由光照、大气等因素引起的光谱类内易变性,或者由环境变化、设备等非线性因素导致的谱间易变性,会导致特征提取精度下降。为了全面考虑解混过程中光谱变化的问题,本文引入光谱易变性的低秩正交先验提出了一种增强型的光谱解混优化模型。首先,在线性解混模型基础上引入易变性数据拟合项来同时考虑光谱类内和类间变化,利用缩放因子来解决光谱类内易变性,同时增加光谱易变性扰动矩阵来解决谱间易变性。其次,该模型利用正交先验约束来实现原光谱字典与易变性项的空间低相干性,通过采用核范数对数松弛来强化丰度矩阵的低秩特性,抑制微小分量及噪声。最后,采用交替优化法及向量-矩阵算子降低求解算法复杂度。通过模拟数据集和真实数据集仿真测试结果表明,本文所提算法取得了优于对比算法的良好性能,验证了该优化模型的有效性。

Hyperspectral unmixing is the process of extracting endmembers and abundance features through image decomposition.However,intra spectral variability caused by factors such as illumination and atmosphere,or in ter spectral variability caused by non linear factors such as environmental changes and equipment,can lead to a decrease in feature extraction accuracy.To comprehensively consider the issue of spectral changes during the unmixing process,an enhanced spectral unmixing optimization model is proposed in this paper by introducing a low rank orthogonal prior for spectral variability.Firstly,a variability data fitting term is introduced on top of the linear unmixing model to account for both intra class and inter class spectral variations.And a scaling factor is used to address intra class variability in the spectrum,while a spectral variability perturbation matrix is added to address inter class variability.Secondly,the model utilises orthogonal prior constraints to achieve the low spatial coherence between the original spectral dictionary and the variability term,and suppresses tiny tiny components and noise by employing kernel norm logarithmic relaxation to strengthen the low rank property of the abundance matrix.Finally,the alternating optimization method and vector matrix operator are used to reduce the complexity of the solution algorithm.The results of simulation tests on both simulated and real datasets show that the proposed algorithm achieves better performance than the comparison algorithm,which verifies the effectiveness of the optimization model.