A new kernel method for hyperspectral image feature extraction

Abstract

Hyperspectral image provides abundant spectral information for remote discrimination of subtle differences in ground covers. However, the increasing spectral dimensions, as well as the information redundancy, make the analysis and interpretation of hyperspectral images a challenge. Feature extraction is a very important step for hyperspectral image processing. Feature extraction methods aim at reducing the dimension of data, while preserving as much information as possible. Particularly, nonlinear feature extraction methods (e.g. kernel minimum noise fraction (KMNF) transformation) have been reported to benefit many applications of hyperspectral remote sensing, due to their good preservation of high-order structures of the original data. However, conventional KMNF or its extensions have some limitations on noise fraction estimation during the feature extraction, and this leads to poor performances for post-applications. This paper proposes a novel nonlinear feature extraction method for hyperspectral images. Instead of estimating noise fraction by the nearest neighborhood information (within a sliding window), the proposed method explores the use of image segmentation. The approach benefits both noise fraction estimation and information preservation, and enables a significant improvement for classification. Experimental results on two real hyperspectral images demonstrate the efficiency of the proposed method. Compared to conventional KMNF, the improvements of the method on two hyperspectral image classification are 8 and 11%. This nonlinear feature extraction method can be also applied to other disciplines where high-dimensional data analysis is required.     

Hyperspectral Image Classification Using Relevance Vector Machines

This letter presents a hyperspectral image classification method based on relevance vector machines (RVMs). Support vector machine (SVM)-based approaches have been recently proposed for hyperspectral image classification and have raised important interest. In this letter, it is genuinely proposed to use an RVM-based approach for the classification of hyperspectral images. It is shown that approximately the same classification accuracy is obtained using RVM-based classification, with a significantly smaller relevance vector rate and, therefore, much faster testing time, compared with SVM-based classification. This feature makes the RVM-based hyperspectral classification approach more suitable for applications that require low complexity and, possibly, real-time classification.     

An improved atmospheric correction algorithm for hyperspectral remotely sensed imagery

There is an increased trend toward quantitative estimation of land surface variables from hyperspectral remote sensing. One challenging issue is retrieving surface reflectance spectra from observed radiance through atmospheric correction, most methods for which are intended to correct water vapor and other absorbing gases. In this letter, methods for correcting both aerosols and water vapor are explored. We first apply the cluster matching technique developed earlier for Landsat-7 ETM+ imagery to Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) data, then improve its aerosol estimation and incorporate a new method for estimating column water vapor content using the neural network technique. The improved algorithm is then used to correct Hyperion imagery. Case studies using AVIRIS and Hyperion images demonstrate that both the original and improved methods are very effective to remove heterogeneous atmospheric effects and recover surface reflectance spectra.     

小样本的高光谱图像降噪与分类

在样本数目稀少情况下实现高光谱图像精细分类是个挑战性的问题。高光谱图像信噪比提高比较困难,噪声大小对分类结果有最直接的影响。利用高光谱图像相邻波段之间的相关性和相邻像素之间的相关性,提出多级降噪滤波的高光谱图像分类方法,通过改进的两阶段稀疏与低秩矩阵分解方法,去除高光谱图像中能量较高的噪声,利用主成分分析方法去除高光谱图像中能量较低的噪声,引导滤波方法去除分类结果图中的"椒盐噪声"。选取两幅真实高光谱图像进行实验,结果表明,两阶段稀疏与低秩矩阵分解法和主成分分析法两种降噪方法具有较强的互补性;引导滤波方法使得分类图更加平滑且分类精度更高。与其他光谱空间分类方法相比,本文方法分类精度更高,且在样本极少时能获得很高的分类精度。     

Hyperspectral image classification by a variable interval spectral average and spectral curve matching combined algorithm

Classification of hyperspectral images has been receiving considerable attention with many new applications reported from commercial and military sectors. Hyperspectral images are composed of a large number of spectral channels, and have the potential to deliver a great deal of information about a remotely sensed scene. However, in addition to high dimensionality, hyperspectral image classification is compounded with a coarse ground pixel size of the sensor for want of adequate sensor signal to noise ratio within a fine spectral passband. This makes multiple ground features jointly occupying a single pixel. Spectral mixture analysis typically begins with pixel classification with spectral matching techniques, followed by the use of spectral unmixing algorithms for estimating endmembers abundance values in the pixel. The spectral matching techniques are analogous to supervised pattern recognition approaches, and try to estimate some similarity between spectral signatures of the pixel and reference target. In this paper, we propose a spectral matching approach by combining two schemes—variable interval spectral average (VISA) method and spectral curve matching (SCM) method. The VISA method helps to detect transient spectral features at different scales of spectral windows, while the SCM method finds a match between these features of the pixel and one of library spectra by least square fitting. Here we also compare the performance of the combined algorithm with other spectral matching techniques using a simulated and the AVIRIS hyperspectral data sets. Our results indicate that the proposed combination technique exhibits a stronger performance over the other methods in the classification of both the pure and mixed class pixels simultaneously.     

属类概率距离构图的半监督高光谱图像分类

提出一种利用属类概率距离构图的半监督学习算法,并应用于高光谱图像分类。首先,该算法利用基于分类的稀疏表达方法来预估未标记样本的属类概率向量,然后,利用这个概率向量对描述数据相似性的距离函数进行改造,改造后的距离函数能有效扩大异类样本点之间的距离,在新的距离函数的度量下,每个样本点的邻域中可包含更多同类的样本点。最后,将该距离函数应用于半监督学习线性邻域传播算法和标签传播算法中。在Hyperion 和AVIRIS高光谱遥感图像上的实验结果表明：相比于传统的基于图的半监督学习算法,该算法能有效提高高光谱遥感图像分类精度。     

De-striping hyperspectral imagery using wavelet transform and adaptive frequency domain filtering

Hyperspectral imagers are built line-by-line similar to images acquired by pushbroom sensors. They can experience striping artifacts due to variations in detector response to incident imagery. In this research, a method for hyperspectral image de-striping based on wavelet analysis and adaptive Fourier zero-frequency amplitude normalization has been developed. The algorithm was tested against three other de-striping algorithms. Hyperspectral image bands of different scenes with significant striping and random noise, as well as an image with simulated noise, were used in the testing. The results were assessed visually and quantitatively using frequency domain Signal-to-Noise Ratio (SNR), Root Mean Square Error (RMSE) and/or Peak Signal-to-Ratio (PSNR). The results demonstrated the superiority of our proposed algorithm in de-striping hyperspectral images without introducing unwanted artifacts, yet preserving image details. In the noise-induced image results, the proposed method reduced RMSE error and improved PSNR by 3.5 dB which is better than other tested methods. A Combined method, integrating the proposed algorithm with a generic wavelet-based de-noising algorithm, showed significant random noise suppression in addition to stripe reduction with a PSNR value of 4.3 dB. These findings make the algorithm a candidate for practical implementation on remote sensing images including high resolution hyperspectral images contaminated with stripe and random noise.     

高光谱图像空间光谱维去相关噪声评估

高光谱图像噪声评估既是评价图像质量的重要内容,也是衡量传感器性能的重要指标。一般噪声评估方法通过对图像规则分割或利用某种距离准则对图像进行连续性分割,计算图像子块的局部标准差或多元线性回归的残差来实现对图像噪声的估计。但这些方法获取的图像子块并不是完全均匀的,图像子块中仍然会存在地物边界,导致图像噪声评估的结果不准确。为了有效提取图像中的均匀子块,本文提出了一种优化的空间光谱维去相关（OSSDC）方法,基于光谱角距离和欧氏距离双重判定,从光谱曲线的形状和数值上寻找相似像元,获取图像中的均匀子块,然后利用多元线性回归计算残差实现对图像噪声的估算。利用模拟图像和实际航空飞行实验获取的高光谱图像对优化算法进行检验,同时与几种常用噪声评估方法进行对比分析,结果表明优化后的算法计算结果更准确,稳定性和适用性优于其他方法。     

结合NSCT和KPCA的高光谱遥感图像去噪

针对高光谱遥感图像易受噪声干扰,本文提出了一种基于非下采样Contourlet变换NSCT(Nonsubsampled Contourlet Transform)和核主成分分析KPCA(Kernel Principal Component Analysis)的去噪方法。首先对高光谱各波段图像进行NSCT分解;然后利用KPCA对NSCT系数进行处理,并在KPCA重构时依据各类噪声的特性选取合适的主成分;最后用处理过的系数进行逆变换得到去噪图像。实验结果表明,本文方法抑制了高光谱遥感图像中的噪声干扰,较完整地保留了原始数据的有效信息。     

高光谱数据截断加权核范数稀疏解混

受仪器和观测条件限制,高光谱数据易受噪声污染,给数据解译带来挑战。针对传统稀疏解混模型抗噪性能差的问题,本文提出一种截断加权核范数稀疏解混方法,利用高光谱图像像元之间的相关性减轻噪声对丰度估计的干扰。该方法借助低秩表示在挖掘数据内在低维结构方面的优势,在稀疏解混中加入基于截断加权核范数的低秩约束,并结合加权稀疏技术,在稀疏正则项中引入空间邻域权重。截断加权核范数对丰度矩阵的奇异值向量分段处理,可以更好地实现丰度矩阵的低秩逼近,使丰度图像保持空间一致性并保留更多细节信息,空间加权策略则增强了丰度图像的空间连续性。模拟高光谱数据、Cuprite矿区真实数据和红树林高光谱数据实验表明,与其他先进的稀疏解混方法相比,所提方法具有更好的抗噪性,能够提高解混精度。     

空间与光谱维度的高光谱图像噪声估计

高光谱图像能够获取地物精细的光谱诊断特征,但受限于多谱段分光的成像机制,图像各个谱段上光成像的能量不足,信噪比难以提升。高光谱图像噪声类型与强度的准确估计,是提升高光谱图像去噪性能的关键,也是优化其成像系统设计的重要依据。现有高光谱图像噪声估计算法通常将不同类型的图像噪声作为一个整体,并未充分考虑不同类型噪声的区别。本文从高光谱图像获取的机理出发,提出了一种联合空间与光谱维度分析的高光谱图像噪声估计方法。首先,建立了高光谱图像噪声退化模型,将图像中的主要噪声定义为两类:条带与高斯噪声。然后,基于条带噪声在空间维度上独特的频率特性,提出了基于傅里叶变换与局部均值滤波的条带噪声估计方法。最后,基于在光谱维度上高光谱图像相邻波段间的高相关性,通过多元回归分析估计高斯噪声的均值与标准差。本文在模拟高光谱噪声数据上进行算法验证的同时,深入分析了高分五号短波红外高光谱相机、机载Nano-Hyperspec成像仪等国内外成像仪获取的真实高光谱数据。实验结果表明,本文提出的噪声估计方法能够有效的估计出高光谱图像不同谱段条带与高斯噪声的量化指标。实验结果可用于分析高光谱图像在不同传感器与不同成像场景下的退化原因,从而设计更优的图像去噪方法与成像系统。     

Automatic subpixel registration for a tunable hyperspectral imaging system

Hyperspectral imagery of the Earth's surface is increasingly being acquired from aerial platforms. The many bands acquired by typical hyperspectral instruments are collected either in a push-broom, scanning, or staring fashion. Staring methods can be used in ground- and aerial-based applications and have the advantage of readily producing coherent images. Staring remote sensing instruments need some form of coregistration to match band-to-band pixel locations because it takes some time for the instrument to acquire images and save them as the aerial platform moves above the target scene. A well-known method for registration is the phase correlation (PC), which may be used to register images to an accuracy of /spl plusmn/1 pixels. This letter reports an enhancement to the PC method that allows for subpixel registration of hyperspectral images. The x-y location at which the maximum correlation function occurs is fitted with a cubic interpolation to find the maximum. This method was implemented to recover subpixel rotation and translation accuracy from an airborne hyperspectral imaging system, dubbed the Portable Hyperspectral Tunable Imaging System. Results showed that the approach improves up to 9.5% of the normalized cross correlation between wavebands in comparison with the PC method alone.     

A Theoretical Framework for Hyperspectral Anomaly Detection Using Spectral and Spatial A Priori Information

This letter presents a new theoretical approach for anomaly detection using a priori information about targets. This a priori knowledge deals with the general spectral behavior and the spatial distribution of targets. In this letter, we consider subpixel and isolated targets that are spectrally anomalous in one region of the spectrum but not in another. This method is totally different from matched filters that suffer from a relative sensitivity to low errors in the target spectral signature. We incorporate the spectral a priori knowledge in a new detection distance, and we propose a Bayesian approach with a Markovian regularization to suppress the potential targets that do not respect the spatial a priori. The interest of the method is illustrated on simulated data consisting in realistic anomalies that are superimposed on a real HyMap hyperspectral image.     

基于灰度归一化的HJ-1A星HSI图像条带噪声去除方法

针对HJ -1A星HSI图像上的条带噪声,在分析传统条带噪声去除方法的基础上,提出了一种新的基于灰度归一化去除条带噪声的方法.首先,设计孤立点检测窗口,将随机斑点噪声与条带噪声分离;然后,利用灰度归一化方法建立图像各列像元灰度级与基准列像元灰度级对应的查找表,根据查找表对HSI图像条带噪声进行消除.实验结果表明,该方法在很好地保留图像光谱信息的情况下,能够有效地消除图像条带噪声和随机斑点噪声.     

高光谱遥感影像优化判别局部对齐特征提取

目前,高光谱遥感特征提取方法往往因受到噪声的干扰而导致降维效果欠佳。近年来,判别局部对齐DLA (Discriminative Locality Alignment)由于可以处理非线性分布样本、保留局部判别信息,同时避免矩阵奇异性问题,受到了很多学者的关注;但该方法无法有效估计和减少噪声对高光谱遥感影像的影响。针对以上问题,本文提出了最小噪声判别局部对齐MDLA (Minimum-noise Discriminative Locality Alignment)的线性特征提取方法和核最小噪声判别局部对齐KMDLA (Kernel Minimum-noise Discriminative Locality Alignment)的非线性特征提取方法。充分利用最小噪声分离MNF (Minimum Noise Fraction)的去噪能力,将MNF与DLA算法结合提出了MDLA方法,该方法首先利用MNF对高光谱遥感影像进行降维,减少图像中的噪声,然后再在子空间进行DLA变换得到最终的投影数据。为提高样本分布的非线性判别能力,基于KMNF与DLA算法将核方法引入MDLA,提出了KMDLA方法,该方法首先通过KMNF将原始空间的数据映射到新的特征空间,然后在特征空间中进行DLA变换得到最终的投影数据。实验部分首先利用3组高光谱遥感数据对提出算法的性能进行评价,并与相关特征提取算法进行了对比分析,最后分析了图像噪声对不同降维方法性能的影响。结果表明:提出的算法对高光谱遥感影像特征提取效果较好,且可有效减少噪声对影像的影响并提升其分类准确度。     

高光谱影像光谱-空间多特征加权概率融合分类

提出了一种基于光谱-空间多特征加权概率融合的高光谱影像分类方法。首先,利用最小噪声分离(minimum noise fraction,MNF)方法对高光谱影像进行降维和特征提取,并以得到的MNF特征影像作为光谱特征,联合灰度共生矩阵(gray level co-occurrence matrix,GLCM)提取的纹理特征、基于OFC算子建立的多尺度形态学特征以及采用连续最大角凸锥(sequential maximum angle convex cone,SMACC)提取的端元组分特征,组成3组光谱-空间特征;然后利用支持向量机(support vector machine,SVM)对每一组光谱-空间特征进行分类,得到每组特征的概率输出结果;最后,建立多特征加权概率融合模型,应用该模型将不同特征的概率输出结果进行加权融合,得到最终分类结果。为了验证该方法的有效性,利用ROSIS和 AVIRIS影像进行试验,总体分类精度分别达到97.65%和96.62%。结果表明本文的方法不但较好地克服了传统基于单一特征高光谱影像分类的局限性,而且其分类效果也优于常规矢量叠加(vector stacking,VS)和概率融合的多特征分类方法,有效地改善了高光谱影像的分类结果。     

高光谱与高空间分辨率遥感数据的亚热带森林生物量反演

精确估算森林生物量对全球碳平衡以及气候变化的研究有重要意义。以亚热带天然次生林为研究对象,借助地面实测样地数据,通过对机载LiCHy(LiDAR,CCD and Hyperspectral)传感器同时获取的高光谱和高空间分辨率数据进行信息提取和数据融合,建模反演森林生物量。首先通过面向对象分割方法进行单木冠幅提取,然后融合从高光谱数据提取的光谱特征变量和从高空间分辨率数据提取的单木冠幅统计变量,构建多元回归模型估算地上、地下生物量,最后利用地面实测生物量经交叉验证评价模型精度。结果表明,综合模型的精度(R~2为0.54—0.62)高于高光谱模型(R~2为0.48—0.57);在高光谱模型中地上生物量模型精度(R~2为0.57)高于地下生物量模型(R~2为0.48);在综合模型中地上生物量模型精度(R~2为0.62)同样高于地下生物量模型(R~2为0.54)。交叉验证结果表明,与仅使用高光谱数据(单一数据源)相比,通过集成高光谱和高空间分辨率数据的生物量反演效果有所提升,可以更加有效地估算亚热带森林生物量。     

光谱角—欧氏距离的高光谱图像辐射归一化

辐射归一化旨在减小不同时相遥感影像间因获取条件不一致而导致的非地表辐射变化的差异,是土地覆盖变化监测的重要前提条件。本文根据高光谱图像上同类地物的谱形及数值的相似性,利用光谱角距离(SAD)和欧氏距离(ED)双重判定选取不变特征点,提出了一种基于光谱角—欧氏距离的辐射归一化方法。在评价指标中除了常用的均方根误差和相对偏差,更增加了高光谱特色的衡量光谱保真性指标:皮尔森系数、光谱扭曲程度。利用高光谱遥感CHRIS图像对本文提出方法进行验证,并与基于多元变化检测(MAD)的辐射归一化方法比较。结果表明,本文方法不仅在辐射特性上优于基于多元变化检测(MAD)的方法,而且具有保持光谱特性的优势,具有较好的应用前景。     

Unsupervised Segmentation of Hyperspectral Images Using Modified Phase Correlation

This letter presents hyperspectral image segmentation based on the phase-correlation measure of subsampled hyperspectral data, which is referred to as modified phase correlation. The hyperspectral spectrum of each pixel is initially subsampled to gain robustness against noise and spatial variability, and phase correlation is applied to determine spectral similarity. Similar and dissimilar pixels are decided according to the peak value of the phase correlation result to determine pixels that fall into the same segments. The approach can be regarded as a region-growing technique. The total number of segments is determined automatically according to the similarity threshold