Band Selection and Dimension Estimation for Hyperspectral Imagery—a New Approach Based on Invasive Weed Optimization

Currently, hyperspectral images have potential applications in many scientific areas due to the high spectral resolution. Extracting suitable and adequate bands/features from high dimensional data is a crucial task to classify such data. To overcome this issue, dimension reduction techniques have direct effects to improve the efficiency of classifiers on hyperspectral images. One common approach for decreasing the dimensionality is the feature/band selection by considering the optimum dimensionality of the hyperspectral imagery. In this paper, a new method was proposed to select optimal band for classification application, based on a metaheuristic Invasive Weed Optimization (IWO) algorithm. In this regard, the K-nearest neighbour (K-NN) technique was used as the classifier. Moreover, as a by-product of our band selection method, a new method was proposed to estimate an optimum dimension of the reduced hyperspectral images for better classification. Experimental results over three real-world hyperspectral datasets clearly showed that the proposed IWO-based band selection algorithm of this study led to the significant progress in selecting suitable bands for classification applications and estimation of optimum dimensionality of these datasets. In this regard, the overall accuracy (OA) of classification of the proposed IWO-based band selection algorithm was 92.02, 93.57, and 89.72 % for each dataset, respectively. Moreover, results reveal the superiority of the proposed IWO-based band selection algorithm against the other algorithms including GA, SA, ACO, and PSO for band selection purpose.     

高光谱图像子空间的波段选择

为降低高光谱遥感数据光谱空间的冗余度,提出一种快速的波段选择方法。该方法在波段子空间下进行,依次选择各子空间中方差最大的波段作为初始波段,设定目标函数,然后逐子空间替换波段使得目标性能更加优化,直至没有替换可以使得目标更优为止。在两个公开高光谱影像数据集上对比3种常用波段选择方法(ABC、AP、ABS)来验证提出方法的有效性,实验结果表明:(1)在印第安纳数据上,本文方法与ABC、AP、ABS所选波段子集相比平均相关性分别降低22.04%、52.61%、55.71%,最佳指数分别提高0.58%、51.73%、0.95%,总体分类精度分别提高0.16%、1.39%、23.07%,在搜索效率上与同类型的ABC方法相比提高6.61%—69.02%;(2)在帕维亚大学数据上,本文方法与ABC、AP、ABS所选波段子集相比平均相关性分别降低2.38%、0.51%、32.83%,最佳指数分别提高1.34%、17.97%、12.92%,总体分类精度分别提高0.31%、0.69%、8.53%,在搜索效率上与同类型的ABC方法相比提高19.13%—86.34%。本文提出的波段选择方法能够选择合适的波段子集满足不同的应用需要,是一种有效的波段选择方法。     

鲁棒多特征谱聚类的高光谱影像波段选择

传统谱聚类的高光谱影像波段选择模型中,采用的波段相似矩阵受到噪声或异常值的影响且仅能表征波段的单一相似特征,导致波段子集的选取结果受到限制.本文从波段选择的目的 出发,提出鲁棒多特征谱聚类方法,整合多个特征的波段相似矩阵来形成综合相似矩阵以解决上述问题.该方法假设4种相似性度量包括光谱信息散度、光谱角度距离、波段相关性...     

利用多时相的高光谱航空图像监测冬小麦条锈病

冬小麦发生锈病 ,叶绿素被大量破坏 ,水分蒸滕量大大增加 ,叶片细胞大小、形态、叶片结构发生了改变 ,从而改变了叶片和冠层的光学特性 ,使得遥感探测与评价成为可能。利用多时相的高光谱航空飞行图像数据 ,了解、分析和发现条锈病病害对作物光谱的影响及其光谱特征 ;设计了病害光谱指数 ,成功地监测了冬小麦条锈病病害程度与范围。对比 3个生育期的条锈病与正常生长冬小麦的PHI图像光谱及光谱特征 ,发现 :5 6 0— 6 70nm黄边、红谷波段 ,条锈病病害冬小麦的冠层反射率高于正常生长的冬小麦光谱反射率 ;近红外波段 ,条锈病病害的冠层反射率低于正常生长的冬小麦光谱反射率 ;条锈病冬小麦冠层光谱红谷吸收深度和绿峰的反射峰高度都会减小     

高光谱图像波段选择的改进二进制布谷鸟算法

波段选择是高光谱遥感图像分类的重要前提，本文提出了一种用于高光谱遥感图像波段选择的改进二进制布谷鸟算法，通过使用混合二进制编码算法更新子代鸟巢和使用遗传算法交叉方式更新被发现鸟巢两个方面对二进制布谷鸟算法进行改进，找出在图像中起主要作用且相关性低的波段，实现对高光谱遥感图像降维。将本文算法运用于PaviaU数据集和AVIRIS数据集，并与二进制布谷鸟算法、二进制粒子群算法、最小冗余最大相关算法、Relief算法等进行对比分析。结果表明，改进二进制布谷鸟算法波段特征选择效率更高，且选取的波段更具代表性，能够较好地提高后续分类精度。     

Examining the utility of random forest and AISA Eagle hyperspectral image data to predict Pinus patula age in KwaZulu-Natal,South Africa

High-data dimensionality is a common problem in hyperspectral data processing. Consequently, remote sensing techniques that reduce the number of bands are considered essential tools for most hyperspectral applications. The aim of this study was to examine the utility of the random forest ensemble to select the optimal subset of hyperspectral bands to predict the age of Pinus patula stands. Airborne AISA Eagle hyperspectral image data were collected over the study area. The random forest ensemble was used to test whether the forward or backward variable selection methods could identify the optimal subset of bands. Results indicate that both the selection methods produced high-predictive accuracies (root mean square error = 3.097 years). However, the backward variable selection method utilized 206 bands for the final model, while the forward variable selection utilized only a small subset of non-redundant bands (n = 9) while preserving the highest model accuracy (R ² = 0.6).     

Spectral Deconvolution and Non-Overlap Bands Sampling for IMS-1 Hyperspectral Imager Data

In this paper, we attempt to realize ‘near true’ spectral signatures of the ground features achievable from IMS-1 hyperspectral Imager (HySI) by employing spectral deconvolution method. From spectral response function of 64 spectral bands, it is first inferred that their spectral bandwidths are higher than the Nyquist sampling, suggesting that the spectral details of ground features are likely to be affected. The proposed approach tries artificially to reduce effective bandwidth of each band, thereby resolving individual band spectrum separable for further spectroscopic investigation. The spectral bands are further sampled to non-overlapping 17 spectral bands at a sampling interval of 25 nm to enable extraction of unique spectral signatures of the ground features.     

Multi-Feature Classification Approach for High Spatial Resolution Hyperspectral Images

High spatial resolution hyperspectral images not only contain abundant radiant and spectral information, but also display rich spatial information. In this paper, we propose a multi-feature high spatial resolution hyperspectral image classification approach based on the combination of spectral information and spatial information. Three features are derived from the original high spatial resolution hyperspectral image: the spectral features that are acquired from the auto subspace partition technique and the band index technique; the texture features that are obtained from GLCM analysis of the first principal component after principal component analysis is performed on the original image; and the spatial autocorrelation features that contain spatial band X and spatial band Y, with the grey level of spatial band X changing along columns and the grey level of spatial band Y changing along rows. The three features are subsequently combined together in Support Vector Machine to classify the high spatial resolution hyperspectral image. The experiments with a high spatial resolution hyperspectral image prove that the proposed multi-feature classification approach significantly increases classification accuracies.     

Spectral Characteristics of Weathered Oil Films on Water Surface and Selection of Potential Sensitive Bands in Hyper-Spectral Images

This study was an experiment to measure the reflectance data of water and oil films at the 0, 1st, 3rd, 12th, and 48th h of weathering and analyzed the spectral reflectance characteristics of the weathered oil films. Based on the experiment and analysis, the measured data were resampled through the spectral response function of the hyperspectral data. The environmental noise equivalent reflectance difference of the image data was then used to assess the sensitivity of each band for the detection of oil films. The bands that were most effective in detecting oil films in various stages of weathering were assessed. Results indicated that the spaceborne hyperspectral sensor employed in this study had the potential to discriminate among oil films in different stages of weathering. At the 0, 1st, 3rd, and 12th h of weathering, discrimination between water and oil films and among oil films in varying stages of weathering was achieved at the visible and near-infrared bands. After 48 h, the potential sensitive bands were distributed mainly in the near-infrared region. This experiment can provide a method to select bands of hyperspectral data for monitoring oil films.     

Dimensionality Reduction and Classification of Hyperspectral Images Using Object-Based Image Analysis

Object-based image analysis (OBIA) has attained great importance for the delineation of landscape features, particularly with the accessibility to satellite images with high spatial resolution acquired by recent sensors. Statistical parametric classifiers have become ineffective mainly due to their assumption of normal distribution, vast increase in the dimensions of the data and availability of limited ground sample data. Despite pixel-based approaches, OBIA takes semantic information of extracted image objects into consideration, and thus provides more comprehensive image analysis. In this study, Indian Pines hyperspectral data set, which was recorded by the AVIRIS hyperspectral sensor, was used to analyse the effects of high dimensional data with limited ground reference data. To avoid the dimensionality curse, principal component analysis (PCA) and feature selection based on Jeffries–Matusita (JM) distance were utilized. First 19 principal components representing 98.5% of the image were selected using the PCA technique whilst 30 spectral bands of the image were determined using JM distance. Nearest neighbour (NN) and random forest (RF) classifiers were employed to test the performances of pixel- and object-based classification using conventional accuracy metrics. It was found that object-based approach outperformed the traditional pixel-based approach for all cases (up to 18% improvement). Also, the RF classifier produced significantly more accurate results (up to 10%) than the NN classifier.     

基于波段子集的独立分量分析的特征提取的高光谱遥感影像分类

针对高光谱影像数据具有波段众多、数据量较大的特点,本文提出了一种基于波段子集的独立分量分析(ICA)特征提取的高光谱遥感影像分类的新方法。以北京昌平小汤山地区的高光谱影像为例,根据高光谱遥感影像的相邻波段的相关性进行子空间划分,在各个波段子集上采用ICA算法进行特征提取,将各个子空间提取的特征合并组成特征向量,采用支持向量机(SVM)分类器进行分类。结果表明:该方法分类精度最佳(分类精度89.04%,Kappa系数0.8605,明显优于其它特征提取方法的SVM分类,有效地提高了高光谱数据的分类精度。     

Using in situ hyperspectral reflectance data to distinguish nine aquatic plant species

In situ hyperspectral reflectance data were studied at 50 bands (10 nm bandwidth) over the 400–900 nm spectral range to determine their potential for distinguishing among nine aquatic plant species: American lotus [Nelumbo lutea (Willd.) Pers.], American pondweed (Potamogeton nodusus Poir.), giant duckweed [Spirodela polyrrhiza (L.) Schleid.], Mexican waterlily (Nymphaea mexicana Zucc.), white waterlily (Nymphaea odorata Aiton), spatterdock [Nuphar lutea (L.) Sm.], giant salvinia (Salvinia molesta Mitchell), waterhyacinth [Eichhornia crassipes (Mart.) Solms] and waterlettuce (Pistia stratiotes L.). The species were studied on three dates: 30 May, 1 July and 3 August 2009. All nine species were studied in July and August, while only eight species were studied in May; giant duckweed was not studied in May due to insufficient availability. Two procedures were used to determine the optimum bands for discriminating among species: multiple comparison range tests and stepwise discriminant analysis. Multiple comparison range tests results for May showed that most separations among species occurred at bands 795–865 nm in the near-infrared (NIR) spectral region where up to six species could be distinguished. For July, few species could be distinguished amongthe 50 bands; most separations occurred at the 715 nm red-NIR edge band where four species could be differentiated. The optimum bands in August occurred in the green (525–595 nm), red (605–635 nm) and red-NIR edge (695–705 nm) spectral regions where up to six species could be distinguished. Stepwise discriminant analysis identified 11 bands in the blue, green, red-NIR edge and NIR spectral regions to be significant to discriminate among the eight species in May. For July and August, stepwise discriminant analysis identified 15bands and 13 bands, respectively, from the blue to NIR regions to be significant for discriminating among the nine species.     

A Gaussian elimination based fast endmember extraction algorithm for hyperspectral imagery

A fast endmember-extraction algorithm based on Gaussian Elimination Method (GEM) is proposed in this paper under the fact that a pixel is an endmember if it has the maximum value in any spectral band of a hyperspectral image when based on linear mixing model. Applying Gaussian elimination is much like performing a lower triangular matrix to transform the hyperspectral image. As more endmembers have been extracted, fewer bands are needed to be involved in the Gaussian elimination process, thus greatly reducing the computing time. The experimental results with both simulated and real hyperspectral images indicate that the method proposed here is much faster than the vertex component analysis (VCA) method, and can provide a similar performance with VCA.     

Estimation of Sugar Beet Aboveground Biomass by Band Depth Optimization of Hyperspectral Canopy Reflectance

Aboveground biomass of sugar beet influences tuber growth and sugar accumulation. Thus, accurate, rapid, and non-destructive technique of biomass estimation is important to optimize the crop management practices to attain the required aboveground biomass to support high tuber yields and optimal sugar content. The current research aimed to evaluate the performance of hyperspectral indices and band depth analysis, to remotely assess the aboveground biomass in sugar beet. The biomass and hyperspectral reflectance were collected at different growth stages in experimental and farmers’ fields. The model development was based on sugar beet plants sampled at various times during the growing period subject to seven nitrogen rates. The results showed that accuracy of biomass estimation was greater when using vegetation indices involving red edge bands (680–740 nm) as compared to that using the red light-based indices. Four types of optimized band depth information (band depth, band depth ratio, normalized band depth index, and band depth normalized to band area) involving the red edge further increased the accuracy of biomass estimation. This study demonstrated as the sugar beet biomass increased towards later growing period, biomass estimation using red light-based vegetation indices were less accurate as compared to that using band depth analysis in the vicinity of the red edge.     

矢量C-V模型的高光谱遥感影像分割

高光谱遥感影像除了包含普通2维影像所具有的空间信息还包含了1维光谱信息,传统的针对2维影像的分割方法不能很好地应用于高光谱遥感影像。为此,本文提出一种能够同时处理多波段影像的高光谱遥感影像矢量C-V模型分割方法。首先选出高光谱遥感影像中目标与背景对比度较大的波段,并通过计算波段相关系数,去除其中的冗余信息形成新的波段组合,进而根据所确定的波段组合构建高光谱遥感影像矢量矩阵;在此基础上,构造基于该矢量矩阵的矢量C-V分割模型。模型中通过引入基于梯度的边缘引导函数,在保留传统C-V模型基于区域信息进行影像分割的基础上,利用影像的边缘细节信息,增强了模型在异质区域和复杂背景情况下对目标边缘的捕捉能力,提高了对高光谱遥感影像的分割精度和速度。最后利用HYPERION数据进行仿真实验,并将实验结果和传统C-V模型和相关方法进行了对比,结果表明,本文方法能够在短时间内有效地分割高光谱遥感影像,与传统方法相比,具有分割精度更高运算速度更快的特点。     

Spectral studies of lunar equivalent rocks — A prelude to lunar mineral mapping

Hyperspectral remote sensing technique is widely applied for geological studies including the study of extra-terrestrial rocks. Since it has many spectral bands, discrimination between rocks and minerals can be done more precisely. To perform chemical and mineralogical mapping and to study the rocks on the lunar surface, India has proposed to launch its first lunar remote sensing satellite Chandrayaan-1 in the year 2008. For mineralogical mapping, the mission will carry a Hyperspectral Imager (HySI) instrument, which operates in the VNIR region. This paper presents-an attempt to study the spectral response of lunar-akin terrestrial rocks, in the VNIR region (as in the case of the proposed HySI on-board Chandrayaan-1). For this purpose, rocks similar to those present on the lunar surface were collected and their spectral response in the 64 simulated bands of HySI sensor were studied using a spectro-radiometer. Petrographic studies and modal analysis were carried out using thin sections of the rock samples. On studying the spectral response of the lunar-like rock samples in the 64 HySI bands, it is seen that there are distinct absorption features in bands 58 (923.75nm-927.5nm) and 63 (942.5nm-946.25nm) of the NIR wavelength ranges, for basalt rocks; distinct reflectance features in band 20 (590nm to 600nm) for ganmbbro: distinct reflectance features in band 19 (580nm to 590nm) and absorption in band 18 (570-580nm) for gabbroic anorthosite and distinct reflection features in band 63 (942.5nm to 946.25nm) for anorthosite. Thus, this study demonstrates the possibility of identifying the minerals and rocks on lunar surface using the hyperspectral approach and the spectral signatures of lunar-like rocks present on Earth.     

面向对象的高光谱遥感影像稀疏表示分类

稀疏表示用于高光谱遥感影像分类多是基于像素层次来处理的。文中提出一种面向对象的高光谱遥感影像稀疏表示分类方法。首先从高光谱影像中提取4个波段组成标准的多波段影像,进行面向对象的影像分割;然后计算各对象在各波段上的光谱均值,并选取少量样本进行训练;最后利用基于Fisher字典学习的稀疏表示进行高光谱遥感影像的分类。实验结果表明,该方法可以利用较少的样本得到较好的分类效果,与基于像素层的稀疏分类相比较,分类精度与效率均有所提高,分类结果更接近真实地物,避免了零碎图斑。     

波段选择协同表达高光谱异常探测算法

高光谱遥感影像具有光谱分辨率极高的特点,承载了大量可区分不同类型地物的诊断性光谱信息以及区分亚类相似地物之间细微差别的光谱信息,在目标探测领域具有独特的优势。与此同时,高光谱遥感影像也带来了数据维数高、邻近波段之间存在大量冗余信息的问题,高维度的数据结构往往使得高光谱影像异常目标类和背景类之间的可分性降低。为了缓解上述问题,本文提出了一种基于波段选择的协同表达高光谱异常探测算法。首先,使用最优聚类框架对高光谱波段进行选择,获得一组波段子集来表示原有的全部波段,使得高光谱影像异常目标类与背景类之间的可分性增强。然后使用协同表达对影像上的像元进行重建,由于异常目标类和背景类之间的可分性增强,对异常目标像元进行协同表达时将会得到更大的残差,异常目标像元的输出值增大,可以更好地实现异常目标和背景类的分离。本文使用了3组高光谱影像数据进行异常目标探测实验,实验结果表明,该方法与其他现有高光谱异常目标探测算法对比,曲线下面积AUC(Area Under Curve)值更高,可以更好地实现异常目标与背景分离,能够更有效地对高光谱影像进行异常目标探测。     

A Multiple SVM System for Classification of Hyperspectral Remote Sensing Data

With recent technological advances in remote sensing sensors and systems, very high-dimensional hyperspectral data are available for a better discrimination among different complex land-cover classes. However, the large number of spectral bands, but limited availability of training samples creates the problem of Hughes phenomenon or ‘curse of dimensionality’ in hyperspectral data sets. Moreover, these high numbers of bands are usually highly correlated. Because of these complexities of hyperspectral data, traditional classification strategies have often limited performance in classification of hyperspectral imagery. Referring to the limitation of single classifier in these situations, Multiple Classifier Systems (MCS) may have better performance than single classifier. This paper presents a new method for classification of hyperspectral data based on a band clustering strategy through a multiple Support Vector Machine system. The proposed method uses the band grouping process based on a modified mutual information strategy to split data into few band groups. After the band grouping step, the proposed algorithm aims at benefiting from the capabilities of SVM as classification method. So, the proposed approach applies SVM on each band group that is produced in a previous step. Finally, Naive Bayes (NB) as a classifier fusion method combines decisions of SVM classifiers. Experimental results on two common hyperspectral data sets show that the proposed method improves the classification accuracy in comparison with the standard SVM on entire bands of data and feature selection methods.     

Testing the capability of spectral resolution of the new multispectral sensors on detecting the severity of grey leaf spot disease in maize crop

Abstract

In this study, we tested whether GLS field symptoms on maize can be detected using hyperspectral data re-sampled to WorldView-2, Quickbird, RapidEye and Sentinel-2 resolutions. To achieve this objective, Random Forest algorithm was used to classify the 2013 re-sampled spectra to represent the three identified disease severity categories. Results showed that Sentinel-2, with 13 spectral bands, achieved the highest overall accuracy and kappa value of 84% and 0.76, respectively, while the WorldView-2, with eight spectral bands, yielded the second highest overall accuracy and kappa value of 82% and 0.73, respectively. Results also showed that the 705 and 710 nm red edge bands were the most valuable in detecting the GLS for Sentinel-2 and RapidEye, respectively. On the re-sampled WorldView 2 and Quickbird sensor resolutions, the respective 608 and 660 nm in the yellow and red bands were identified as the most valuable for discriminating all categories of infection.