结合空间光谱预处理和约束非负矩阵分解的高光谱图像混合像元分解

发布时间：2018-06-26 04:10

本文选题：高光谱图像 + 混合像元分解　；参考：《中国图象图形学报》2017年04期

【摘要】：目的混合像元问题在高光谱遥感图像处理分析中普遍存在,非负矩阵分解的方法被引入到高光谱图像解混中。本文提出结合空间光谱预处理和约束非负矩阵分解的混合像元分解流程。方法结合空间光谱预处理的约束非负矩阵分解,如最小体积约束、流行约束等,通过加入邻域的空间和光谱信息进行预处理获得更优的预选端元,从而对非负矩阵分解的解混结果进行优化。结果在5组不同信噪比的模拟数据实验中,空间预处理(SPP)和空间光谱预处理(SSPP)均能够有效提高约束非负矩阵分解(最小体积约束的非负矩阵分解和图正则非负矩阵分解)的解混结果,其中SPP在不同信噪比的情况下都能优化约束非负矩阵分解的结果,而SSPP在低信噪比的情况下,预处理效果更佳。利用美国内华达州Cuprite矿区数据进行真实数据实验,SPP提高了约束非负矩阵分解的解混精度,而SSPP在复杂场景下,解混精度更佳。模拟数据和真实数据的实验均表明,空间光谱预处理能够有效地提高约束非负矩阵分解的解混精度,特别是对于信噪比较低的情况下,融合空间和光谱信息对噪声有很好的鲁棒性。结论本文对约束非负矩阵分解的解混算法添加空间光谱预处理,利用高光谱遥感数据的空间和光谱信息,优化预选端元,加入空间光谱预处理的非负矩阵解混实验流程,在复杂场景情况下,对噪声具有较好的鲁棒性。
[Abstract]:Aim the mixed pixel problem is widely used in hyperspectral remote sensing image processing and analysis, and the method of nonnegative matrix decomposition is introduced into hyperspectral image de-mixing. In this paper, a mixed pixel decomposition process combining spatial spectral preprocessing and constrained nonnegative matrix decomposition is proposed. Methods combining the constraint nonnegative matrix factorization of spatial spectral preprocessing, such as minimum volume constraint, popular constraint and so on, preprocessing the spatial and spectral information of the neighborhood to obtain a better pre-selected end element. Thus, the unmixing results of nonnegative matrix decomposition are optimized. Results in five groups of simulated data with different signal-to-noise ratio, Both spatial preprocessing (SPP) and spatial spectral preprocessing (SSPP) can effectively improve the unmixing results of constrained nonnegative matrix decomposition (minimum volume constrained nonnegative matrix decomposition and graph regular nonnegative matrix factorization). SPP can optimize the results of constrained nonnegative matrix decomposition under different SNR, while SSPP has better preprocessing effect under low SNR. The real data experiment using Cuprite mining area data in Nevada, USA, has improved the resolution accuracy of constrained nonnegative matrix decomposition, while SSPP is better in complex scenarios. The experiments of simulated data and real data show that spatial spectral preprocessing can effectively improve the unmixing accuracy of constrained nonnegative matrix decomposition, especially in the case of low signal-to-noise ratio (SNR). The fusion space and spectral information are robust to noise. Conclusion in this paper, spatial spectral pretreatment is added to the unmixing algorithm of constrained nonnegative matrix decomposition. By using spatial and spectral information of hyperspectral remote sensing data, the pre-selected endelements are optimized and the experimental flow of non-negative matrix unmixing with spatial spectral preprocessing is optimized. In the case of complex scene, it has good robustness to noise.
【作者单位】：中国科学院遥感与数字地球研究所;郑州大学信息工程学院;
【基金】：国家自然科学基金项目(41325004,41571349)~~
【分类号】：TP751

【相似文献】