高光谱遥感影像复杂非线性条带噪声的去除方法研究

Hyperspectral remote sensing images are important for earth observation, possessing urgent needs and wide application value. However, due to the constraints of the sensor performance or the impacts from the imaging environment, the acquired images are vulnerable to structurally distributed stripe noise, which seriously limits the subsequent applications. Worse still, the non-linearity and complexity of stripe noise increase significantly with the improvement of space and spectral resolution in imaging. The existing methods are far from meeting the needs of actual image destriping. Therefore, in view of the technical bottleneck of complex non-linear stripe elimination in the hyperspectral images, this project is conducted with the main line of feature characterization, adaptive modeling, and fast solving. First of all, the differences between stripe noise and other image components are analyzed and used to construct the separability characteristics of complex stripes. Then, an adaptive hyperspectral stripe removal framework is designed after full consideration of hyperspectral mixed noise, noise differences among different bands, complex stripe characteristics, and space-spectrum fidelity requirements. Finally, the multi-strategy accelerated optimization approach including block processing, numerical optimization, and parallel computing is developed to process the massive remote sensing data. The results of this project can break through the technical limitations of existing destriping methods, and realize the unification of accurate stripe removal and spatial-spectral detail preservation, which effectively enhances the application potential of hyperspectral remote sensing images.

高光谱遥感影像是对地观测的重要数据，具有迫切需求和广泛应用价值。但是受传感器性能及成像环境的影响，影像在获取时易受结构化分布的条带噪声污染，严重制约后续应用。特别当影像空间、光谱分辨率同时提升时，条带噪声的非线性化、复杂化问题尤为突出，现有方法远不能满足实际处理的需求。因此，本项目拟针对高光谱影像复杂非线性条带处理时的技术瓶颈，以“特征刻画-自适应建模-快速求解”为研究主线：解析条带噪声和其他影像成分的差异，刻画复杂条带的可分性特征；综合高光谱混合噪声情形、谱段噪声差异、复杂条带特征和空-谱保真需求，构建顾及多因素的自适应条带去除模型；面向海量遥感数据的处理任务，发展“分块处理-数值优化-并行计算”多策略加速的优化求解方法。本项目的研究成果可突破现有方法在遥感影像空间、光谱分辨率同时提升时条带处理的技术局限，实现条带去除与空-谱细节保持的统一，有效提升高光谱遥感影像的实际应用潜力。

高光谱遥感影像是对地观测的重要数据，具有迫切需求和广泛应用价值。但受传感器性能及成像环境的影响，影像在获取时易受结构化分布的条带噪声污染，严重制约后续应用。针对高光谱影像复杂条带噪声的处理问题，本项目围绕“特征表示-模型构建-快速求解”的研究主线，从高光谱噪声分析入手，稳步开展了复杂条带的可分特征表示、噪声去除模型构建和快速数值求解三方面内容的研究。在条带特征表示方面，发展了基于排序域的条带特征表示方法，并提出了顾及乘性因素的条带噪声建模方法；在条带去除模型构建方面，就条带可能影响的数据维度，发展了稳健的一维、二维及三维噪声去除模型；在快速数值求解方面，依据条带的物理特性和统计规律，提出了相应的快速数值求解策略。本项目的研究成果可解决现有方法在复杂条带去除与地物细节协同保持上的技术局限，有效提升高光谱遥感影像的应用潜力。. 按照既定的研究计划，项目顺利完成了课题设定的研究内容，实现了预期目标。在项目的支持下，项目组发表了科研论文8篇，其中SCI检索6篇；授权专利1项；培养在读博士1人、硕士2人。