基于加权张量分解的高光谱混合噪声去除方法  

Hyperspectral Mixed Noise Removal Method Based on Weighted Tensor Decomposition

作　　者：李东益 沈焕锋[1] 管小彬 储栋 LI Dongyi;SHEN Huanfeng;GUAN Xiaobin;CHU Dong(College of Resources and Environmental Sciences,Wuhan University,Wuhan 430079,China;College of Geography and Tourism,Anhui Normal University,Wuhu,Anhui 241000,China)

机构地区：[1]武汉大学资源与环境科学学院,武汉430079 [2]安徽师范大学地理与旅游学院,安徽芜湖241000

出　　处：《遥感信息》2025年第1期1-9,共9页Remote Sensing Information

基　　金：国家自然科学基金(U23A2021);湖北省重大科技专项资金(2023BCA003)。

摘　　要：高光谱影像不可避免地会受到高斯噪声、脉冲噪声、条带噪声等不同类型的混合噪声影响,极大地限制了影像的后续应用。各国学者已经发展了系列高光谱去噪方法,但仍难以处理多种类型的混合噪声,在抑制噪声的同时往往难以兼顾高频细节的保留,特别是对条带噪声的去除效果不佳。为此,文章提出一种联合加权非局部低秩张量与条带低秩正则化约束的高光谱去噪模型,通过设计的加权自适应收缩算法实现更精确的低秩张量奇异值分解,能在影像细节保真的前提下有效去除严重场景噪声。另外,利用低秩矩阵分解对条带噪声进行建模,增强了模型对条带噪声的去除能力,从而有效去除不同类型的混合噪声。模拟实验和真实实验结果显示,该方法在定性和定量上均优于对比方法,能够在去除不同高光谱传感器各类噪声的同时更好地保留空间细节。Hyperspectral images are inevitably affected by different types of mixed noises such as Gaussian noise,impulse noise,and stripe noise,which greatly limits the subsequent application of images.Scholars from various countries have developed a series of HSI denoising methods,but it is still difficult to deal with various types of mixed noises.It is often difficult to preserve high-frequency details while suppressing noise,especially the removal of stripe noise is not effective.To this end,this paper proposes a HSI denoising model that combines weighted non-local low-rank tensor and stripe low-rank regularization constraints.The designed weighted adaptive shrinkage algorithm achieves a more accurate low-rank tensor singular value decomposition,which can effectively remove severe scene noise while preserving image details.In addition,the stripe noise is modeled using low-rank matrix decomposition,which enhances the model’s ability to remove stripe noise,thereby effectively removing different types of mixed noise.The results of simulation experiments and real experiments show that the proposed method is superior to the comparative methods in both qualitative and quantitative aspects,and can better retain spatial details while removing various types of noise from different hyperspectral sensors.

关 键 词：高光谱影像去噪 张量奇异值分解 条带噪声 非局部模型 低秩表示 

分 类 号：P237[天文地球—摄影测量与遥感]