面向亚像素级高光谱海洋溢油探测的端元变异鲁棒解混方法研究

In recent years, with the maturity and deployment of hyperspectral satellites in various countries, their application in remote sensing of marine oil spill has become a hot topic. For hyperspectral imaging remote sensing, spectral resolution and spatial resolution are always contradictory. Especially in marine remote sensing applications such as spaceborne and airborne, it is inevitable that single-pixels contain multiple spectral features, resulting in “mixed pixels”. Unmixing technology is the solution to this problem, but it still faces serious "endmember variability" problems, especially in the marine environment. The variation caused by the influence of ocean waves and solar flare is very different from ordinary endmember variation, which leads to the failure of traditional unmixing methods, and makes hyperspectral techniques still unable to play a role in sub-pixel scale oil spill detection. The purpose of this research is to propose an endmember variability robust unmixing model based on Gaussian mixture model (GMM). On this basis, low-rank and non-linear constraints are studied to further improve the unmixing accuracy. Furthermore, a detection method based on GMM unmixing and abundance low-rank decomposition is proposed to form a complete sub-pixel verification chain of marine oil spill detection methods. The ultimate goal of this research is to form a robust unmixing method for endmember variability problem, to realize the sub-pixel level of hyperspectral oil spill detection and sensing capabilities, and to provide reliable technical support for marine oil spill monitoring.

近年来随着各国高光谱卫星的陆续成熟和部署，其在海洋溢油遥感中的应用已成为热点。对高光谱成像系统而言，光谱分辨率和空间分辨率始终是矛盾关系，尤其在星载、机载等遥感应用中，必然使单像元包含多种地物光谱，导致“混合像元”产生。光谱解混技术是该问题的解决途径，但仍面临严重的“端元变异”问题，尤其是海洋环境中，海浪、太阳耀光的影响导致的端元变异现象与普通端元变异有极大差异，进而导致传统解混方法失效，至今令高光谱遥感仍无法在亚像素尺度溢油探测中发挥作用。本项目旨在提出一种基于高斯混合模型（GMM）的端元变异鲁棒解混模型，在此基础上开展低秩、非线性约束研究，力图使解混精度进一步提升；并提出基于GMM解混和丰度低秩分解的探测方法，以形成完整的亚像素级海洋溢油探测方法验证链条。本项目最终目标是形成一套端元变异鲁棒的解混方法，实现亚像素级的高光谱海洋溢油探测感知能力，为海洋溢油监测提供可靠的技术支撑。

课题面向亚像素级海洋溢油探测需求，采用高光谱技术手段，以解混为主导思想开展研究。课题突破了端元变异解混技术、亚像素目标探测技术等若干关键技术，形成一套端元变异鲁棒的解混方法，实现亚像素级的高光谱海洋溢油探测感知能力，为海洋溢油监测提供可靠的技术支撑。针对目前高光谱解混方法存在的对海浪、太阳耀光造成的特殊端元变异问题难以有效表示的问题，提出了一种基于 GMM 的端元变异解混建模方法。针对 GMM 模型在解混应用中面临的特殊问题，在 GMM 解混模型基础上开展低秩、非线性约束设计研究，进一步提升解混精度；最后，提出基于 GMM 解混与丰度低秩分解的目标探测方法，形成一套完整的面向亚像素级海洋溢油探测方法的验证链条。.项目成果在TNNLS、JAS、TGRS、GRSL、RS等遥感或红外权威期刊发表SCI论文17篇，遥感顶级会议IEEE International Geoscience and Remote Sensing Symposium (IGARSS)上发表口头报告论文3篇；授权发明专利2项；成果针对海洋溢油探测的端元变异问题，实现了亚像素级的高光谱海洋溢油解混与探测方法，有效提升了海洋溢油检测的技术水平。