Enhanced hyperspectral imagery representation via diffusion geometric coordinates

The concise and informative representation of hyperspectral imagery is achieved via the introduced diffusion geometric coordinates derived from nonlinear dimension reduction maps - diffusion maps. The huge-volume high- dimensional spectral measurements are organized by the affinity graph where each node in this graph only connects to its local neighbors and each edge in this graph represents local similarity information. By normalizing the affinity graph appropriately, the diffusion operator of the underlying hyperspectral imagery is well-defined, which means that the Markov random walk can be simulated on the hyperspectral imagery. Therefore, the diffusion geometric coordinates, derived from the eigenfunctions and the associated eigenvalues of the diffusion operator, can capture the intrinsic geometric information of the hyperspectral imagery well, which gives more enhanced representation results than traditional linear methods, such as principal component analysis based methods. For large-scale full scene hyperspectral imagery, by exploiting the backbone approach, the computation complexity and the memory requirements are acceptable. Experiments also show that selecting suitable symmetrization normalization techniques while forming the diffusion operator is important to hyperspectral imagery representation.

DIMENSIONALITY REDUCTION FOR HYPERSPECTRAL IMAGERY BASED ON FASTICA

The high dimensions of hyperspectral imagery have caused burden for further processing. A new Fast Independent Component Analysis (FastICA) approach to dimensionality reduction for hyperspectral imagery is presented. The virtual dimensionality is introduced to determine the number of dimensions needed to be preserved. Since there is no prioritization among independent components generated by the FastICA,the mixing matrix of FastICA is initialized by endmembers,which were extracted by using unsupervised maximum distance method. Minimum Noise Fraction (MNF) is used for preprocessing of original data,which can reduce the computational complexity of FastICA significantly. Finally,FastICA is performed on the selected principal components acquired by MNF to generate the expected independent components in accordance with the order of endmembers. Experimental results demonstrate that the proposed method outperforms second-order statistics-based transforms such as principle components analysis.

Abundance quantification by independent component analysis of hyperspectral imagery for oil spill coverage calculation

The estimation of oil spill coverage is an important part of monitoring of oil spills at sea.The spatial resolution of images collected by airborne hyper-spectral remote sensing limits both the detection of oil spills and the accuracy of estimates of their size.We consider at-sea oil spills with zonal distribution in this paper and improve the traditional independent component analysis algorithm.For each independent component we added two constraint conditions:non-negativity and constant sum.We use priority weighting by higher-order statistics,and then the spectral angle match method to overcome the order nondeterminacy.By these steps,endmembers can be extracted and abundance quantified simultaneously.To examine the coverage of a real oil spill and correct our estimate,a simulation experiment and a real experiment were designed using the algorithm described above.The result indicated that,for the simulation data,the abundance estimation error is 2.52% and minimum root mean square error of the reconstructed image is 0.030 6.We estimated the oil spill rate and area based on eight hyper-spectral remote sensing images collected by an airborne survey of Shandong Changdao in 2011.The total oil spill area was 0.224 km^2,and the oil spill rate was 22.89%.The method we demonstrate in this paper can be used for the automatic monitoring of oil spill coverage rates.It also allows the accurate estimation of the oil spill area.

Multi-Scale Feature Extraction for Joint Classification of Hyperspectral and LiDAR Data

With the development of sensors,the application of multi-source remote sensing data has been widely concerned.Since hyperspectral image(HSI)contains rich spectral information while light detection and ranging(LiDAR)data contains elevation information,joint use of them for ground object classification can yield positive results,especially by building deep networks.Fortu-nately,multi-scale deep networks allow to expand the receptive fields of convolution without causing the computational and training problems associated with simply adding more network layers.In this work,a multi-scale feature fusion network is proposed for the joint classification of HSI and LiDAR data.First,we design a multi-scale spatial feature extraction module with cross-channel connections,by which spatial information of HSI data and elevation information of LiDAR data are extracted and fused.In addition,a multi-scale spectral feature extraction module is employed to extract the multi-scale spectral features of HSI data.Finally,joint multi-scale features are obtained by weighting and concatenation operations and then fed into the classifier.To verify the effective-ness of the proposed network,experiments are carried out on the MUUFL Gulfport and Trento datasets.The experimental results demonstrate that the classification performance of the proposed method is superior to that of other state-of-the-art methods.

Mapping spatial variation in acorn production from airborne hyperspectral imagery

Masting is a well-marked variation in yields of oak forests. In Japan, this phenomenon is also related to wildlife management and oak regeneration practices. This study demonstrates the capability of integrating remote sensing techniques into map- ping spatial variation of acorn production. The hyperspectral images in 72 wavelengths （407-898 nm） were acquired over the study area ten times over a period of three years （2003-2005） during the early growing season of Quercus serrata using the Airborne Im- aging Spectrometer Application （AISA） Eagle System. With the canopy spectral reflectance values of 22 sample trees extracted from the images, yield estimation models were developed via multiple linear regression （MLR） analyses. Using the object-oriented classi- fication approach in eCognition, canopies representative of individual oak trees （Q. serrata） were identified from the corresponding hyperspectral imagery and combined with the fitted estimation models developed, acorn yield over the entire forest were estimated and visualized into maps. Three estimation models, obtained for June 27 in 2003, July 13 in 2004 and June 21 in 2005, showed good performance in acorn yield estimation both for the training and validation datasets, all with R2 〉 0.4, p 〈 0.05 and RRMSE 〈 1 （the relative root mean square of error）. The present study shows the potential of airborne hyperspectral imagery not only in estimating acorn yields during early growing seasons, but also in identifying Q. serrata from other image objects, based on which of the spatial distribution patterns of acorn production over large areas could be mapped. The yield map can provide within-stand abundance and valuable information for the size and spatial synchrony of acorn production.

Hypergraph Regularized Deep Autoencoder for Unsupervised Unmixing Hyperspectral Images

Deep learning(DL)has shown its superior performance in dealing with various computer vision tasks in recent years.As a simple and effective DL model,autoencoder(AE)is popularly used to decompose hyperspectral images(HSIs)due to its powerful ability of feature extraction and data reconstruction.However,most existing AE-based unmixing algorithms usually ignore the spatial information of HSIs.To solve this problem,a hypergraph regularized deep autoencoder(HGAE)is proposed for unmixing.Firstly,the traditional AE architecture is specifically improved as an unsupervised unmixing framework.Secondly,hypergraph learning is employed to reformulate the loss function,which facilitates the expression of high-order similarity among locally neighboring pixels and promotes the consistency of their abundances.Moreover,L_(1/2)norm is further used to enhance abundances sparsity.Finally,the experiments on simulated data,real hyperspectral remote sensing images,and textile cloth images are used to verify that the proposed method can perform better than several state-of-the-art unmixing algorithms.

Denoising of hyperspectral imagery by cubic smoothing spline in the wavelet domain

The acquired hyperspectral images （HSIs） are inherently attected by noise wlm Dano-varylng level, which cannot be removed easily by current approaches. In this study, a new denoising method is proposed for removing such kind of noise by smoothing spectral signals in the transformed multi- scale domain. Specifically, the proposed method includes three procedures： 1 ） applying a discrete wavelet transform （DWT） to each band; 2） performing cubic spline smoothing on each noisy coeffi- cient vector along the spectral axis; 3 ） reconstructing each band by an inverse DWT. In order to adapt to the band-varying noise statistics of HSIs, the noise covariance is estimated to control the smoothing degree at different spectra｜ positions. Generalized cross validation （GCV） is employed to choose the smoothing parameter during the optimization. The experimental results on simulated and real HSIs demonstrate that the proposed method can be well adapted to band-varying noise statistics of noisy HSIs and also can well preserve the spectral and spatial features.

Inversion of maize leaf nitrogen using UAV hyperspectral imagery in breeding fields

Nitrogen(N)as a pivotal factor in influencing the growth,development,and yield of maize.Monitoring the N status of maize rapidly and non-destructive and real-time is meaningful in fertilization management of agriculture,based on unmanned aerial vehicle(UAV)remote sensing technology.In this study,the hyperspectral images were acquired by UAV and the leaf nitrogen content(LNC)and leaf nitrogen accumulation(LNA)were measured to estimate the N nutrition status of maize.24 vegetation indices(VIs)were constructed using hyperspectral images,and four prediction models were used to estimate the LNC and LNA of maize.The models include a single linear regression model,multivariable linear regression(MLR)model,random forest regression(RFR)model,and support vector regression(SVR)model.Moreover,the model with the highest prediction accuracy was applied to invert the LNC and LNA of maize in breeding fields.The results of the single linear regression model with 24 VIs showed that normalized difference chlorophyll(NDchl)had the highest prediction accuracy for LNC(R^(2),RMSE,and RE were 0.72,0.21,and 12.19%,respectively)and LNA(R^(2),RMSE,and RE were 0.77,0.26,and 14.34%,respectively).And then,24 VIs were divided into 13 important VIs and 11 unimportant VIs.Three prediction models for LNC and LNA were constructed using 13 important VIs,and the results showed that RFR and SVR models significantly enhanced the prediction accuracy of LNC and LNA compared to the multivariable linear regression model,in which RFR model had the highest prediction accuracy for the validation dataset of LNC(R^(2),RMSE,and RE were 0.78,0.16,and 8.83%,respectively)and LNA(R^(2),RMSE,and RE were 0.85,0.19,and 9.88%,respectively).This study provides a theoretical basis for N diagnosis and precise management of crop production based on hyperspectral remote sensing in precision agriculture.

A background refinement method based on local density for hyperspectral anomaly detection

For anomaly detection,anomalies existing in the background will affect the detection performance.Accordingly,a background refinement method based on the local density is proposed to remove the anomalies from thebackground.In this work,the local density is measured by its spectral neighbors through a certain radius which is obtained by calculating the mean median of the distance matrix.Further,a two-step segmentation strategy is designed.The first segmentation step divides the original background into two subsets,a large subset composed by background pixels and a small subset containing both background pixels and anomalies.The second segmentation step employing Otsu method with an aim to obtain a discrimination threshold is conducted on the small subset.Then the pixels whose local densities are lower than the threshold are removed.Finally,to validate the effectiveness of the proposed method,it combines Reed-Xiaoli detector and collaborative-representation-based detector to detect anomalies.Experiments are conducted on two real hyperspectral datasets.Results show that the proposed method achieves better detection performance.

An Intelligent Deep Learning Based Xception Model for Hyperspectral Image Analysis and Classification

Due to the advancements in remote sensing technologies,the generation of hyperspectral imagery(HSI)gets significantly increased.Accurate classification of HSI becomes a critical process in the domain of hyperspectral data analysis.The massive availability of spectral and spatial details of HSI has offered a great opportunity to efficiently illustrate and recognize ground materials.Presently,deep learning(DL)models particularly,convolutional neural networks(CNNs)become useful for HSI classification owing to the effective feature representation and high performance.In this view,this paper introduces a new DL based Xception model for HSI analysis and classification,called Xcep-HSIC model.Initially,the presented model utilizes a feature relation map learning(FRML)to identify the relationship among the hyperspectral features and explore many features for improved classifier results.Next,the DL based Xception model is applied as a feature extractor to derive a useful set of features from the FRML map.In addition,kernel extreme learning machine(KELM)optimized by quantum-behaved particle swarm optimization(QPSO)is employed as a classification model,to identify the different set of class labels.An extensive set of simulations takes place on two benchmarks HSI dataset,namely Indian Pines and Pavia University dataset.The obtained results ensured the effective performance of the XcepHSIC technique over the existing methods by attaining a maximum accuracy of 94.32%and 92.67%on the applied India Pines and Pavia University dataset respectively.

Low-Rank and Sparse Representation with Adaptive Neighborhood Regularization for Hyperspectral Image Classification

Low-Rank and Sparse Representation(LRSR)method has gained popularity in Hyperspectral Image(HSI)processing.However,existing LRSR models rarely exploited spectral-spatial classification of HSI.In this paper,we proposed a novel Low-Rank and Sparse Representation with Adaptive Neighborhood Regularization(LRSR-ANR)method for HSI classification.In the proposed method,we first represent the hyperspectral data via LRSR since it combines both sparsity and low-rankness to maintain global and local data structures simultaneously.The LRSR is optimized by using a mixed Gauss-Seidel and Jacobian Alternating Direction Method of Multipliers(M-ADMM),which converges faster than ADMM.Then to incorporate the spatial information,an ANR scheme is designed by combining Euclidean and Cosine distance metrics to reduce the mixed pixels within a neighborhood.Lastly,the predicted labels are determined by jointly considering the homogeneous pixels in the classification rule of the minimum reconstruction error.Experimental results based on three popular hyperspectral images demonstrate that the proposed method outperforms other related methods in terms of classification accuracy and generalization performance.

Unsupervised hyperspectral unmixing based on robust nonnegative dictionary learning

Considering the sparsity of hyperspectral images(HSIs),dictionary learning frameworks have been widely used in the field of unsupervised spectral unmixing.However,it is worth mentioning here that existing dictionary learning method-based unmixing methods are found to be short of robustness in noisy contexts.To improve the performance,this study specifically puts forward a new unsupervised spectral unmixing solution.For the reason that the solution only functions in a condition that both endmembers and the abundances meet non-negative con-straints,a model is built to solve the unsupervised spectral un-mixing problem on the account of the dictionary learning me-thod.To raise the screening accuracy of final members,a new form of the target function is introduced into dictionary learning practice,which is conducive to the growing robustness of noisy HSI statistics.Then,by introducing the total variation(TV)terms into the proposed spectral unmixing based on robust nonnega-tive dictionary learning(RNDLSU),the context information under HSI space is to be cited as prior knowledge to compute the abundances when performing sparse unmixing operations.Ac-cording to the final results of the experiment,this method makes favorable performance under varying noise conditions,which is especially true under low signal to noise conditions.

New Results in Perceptually Lossless Compression of Hyperspectral Images

Hyperspectral images (HSI) have hundreds of bands, which impose heavy burden on data storage and transmission bandwidth. Quite a few compression techniques have been explored for HSI in the past decades. One high performing technique is the combination of principal component analysis (PCA) and JPEG-2000 (J2K). However, since there are several new compression codecs developed after J2K in the past 15 years, it is worthwhile to revisit this research area and investigate if there are better techniques for HSI compression. In this paper, we present some new results in HSI compression. We aim at perceptually lossless compression of HSI. Perceptually lossless means that the decompressed HSI data cube has a performance metric near 40 dBs in terms of peak-signal-to-noise ratio (PSNR) or human visual system (HVS) based metrics. The key idea is to compare several combinations of PCA and video/ image codecs. Three representative HSI data cubes were used in our studies. Four video/image codecs, including J2K, X264, X265, and Daala, have been investigated and four performance metrics were used in our comparative studies. Moreover, some alternative techniques such as video, split band, and PCA only approaches were also compared. It was observed that the combination of PCA and X264 yielded the best performance in terms of compression performance and computational complexity. In some cases, the PCA + X264 combination achieved more than 3 dBs than the PCA + J2K combination.

基于多尺度非对称密集网络的高光谱图像分类

近年来,基于有限标记样本的高光谱图像(HSI)分类方法取得了重大进展。然而,由于高光谱图像的特殊性,冗余的信息和有限的标记样本给提取强判别特征带来了巨大挑战。此外,由于各类别像素分布不均,如何强化中心像素的作用,减弱不同类别的周围像素的负面影响也是提高分类性能的关键。为了克服上述局限性,该文提出一种基于多尺度非对称密集网络(MS-ADNet)的高光谱图像分类方法。首先,提出一个多尺度样本构建模块,通过在每个像素周围提取多个尺度的图像块,并进行反卷积和拼接以构建输入样本,使其既包含详细的结构区域,又包含较大的同质区域;然后,提出一个非对称密集连接结构,在空间和光谱特征联合提取中实现核骨架增强,即增强了方形卷积核的中心十字区域部分提取的特征,有效地促进了特征重用。此外,为了提高光谱特征的鉴别性,提出一种精简的元素光谱注意力机制,并将其置于密集连接网络的前端和后端。在每类仅采用5个样本进行网络训练的情况下,该方法在Indiana Pines, Pavia University和Salinas数据集上的总体准确率分别达到了77.66%, 84.54%和92.39%,取得了极具竞争力的分类结果。

星载LiDAR与HJ-1A/HSI高光谱数据联合估测区域森林冠层高度

【目的】将ICESat-GLAS波形数据与HJ-1A/HSI高光谱数据联合,借助HSI高光谱数据提供的连续高分辨率光谱信息,实现区域森林冠层高度的估测,降低由于GLAS光斑呈离散条带状分布无法覆盖整个研究区造成的估测误差。【方法】首先,从平滑后的ICESat-GLAS波形数据中提取波形参数(波形长度W和地形坡度参数TS),基于W和TS建立GLAS森林冠层高度估测模型,并利用此模型计算研究区所有GLAS光斑内的森林冠层高度;然后,采用最小噪声分离法(MNF)对HJ-1A/HSI高光谱数据进行降维,提取前3个MNF分量(MNF1,MNF2,MNF3);最后,基于支持向量回归机(SVR)算法,利用GLAS估测的森林冠层高度和3个MNF分量建立区域森林冠层高度SVR估测模型,并估测研究区内无GLAS光斑覆盖区域的森林冠层高度,生成森林冠层高度分布图。【结果】从ICESat-GLAS波形数据中提取的地形坡度参数TS与野外实测地形坡度具有显著线性关系(R2=0.78);基于W和TS建立的GLAS森林冠层高度估测模型的R^2=0.78,RMSE=2.51 m,模型验证的R^2=0.85,RMSE=1.67 m;基于支持向量回归机算法建立的SVR模型建模的R2=0.70,RMSE=3.62 m,模型验证的R2=0.67,RMSE=4.42 m。采用野外数据对最终得到的森林冠层高度分布图的估测误差进行分析,结果估测误差最大值为7.10 m,最小值为0.07 m,平均值为1.78 m,估测误差的标准差为1.49 m,Q1为0.75 m,Q3为2.31 m。【结论】从ICESat-GLAS波形数据中提取的地形坡度参数TS能够很好地反映地形坡度的变化,本研究建立的线性关系模型可克服对数关系模型在平坦地区解释困难的问题。基于支持向量回归机算法,将ICESat-GLAS波形数据与HJ-1A/HSI高光谱数据联合,可克服ICESat-GLAS由于光斑呈离散条带状分布无法实现区域森林冠层高度估测的不足,实现对区域森林冠层高度的高精度估测。

基于HSI高光谱和TM图像的土地盐渍化信息提取方法

选择黄河三角洲垦利县代表性盐碱化区域为研究区,以2011年3月15日HJ-1A卫星HSI高光谱影像和2011年3月22日TM影像为信息源,经几何纠正、图像裁剪、大气校正等预处理,分析不同盐渍化程度土地、水体、滩涂等主要地类的光谱特征,确定地类信息提取特征波段。结合土壤盐分含量,采用定量与定性相结合规则,构建地类信息提取模型,以决策树分类方法进行图像分类,提取土地盐渍化信息。利用地表点位土壤含盐量数据对地表土地盐渍化程度的化学分析结果,对遥感解译数据进行精度验证,并对高光谱和多光谱影像的分类精度进行比较分析。结果表明:HSI图像的总体分类精度达96.43%,Kappa系数为95.59%,而TM图像的总体分类精度为89.17%,Kappa系数为86.74%,说明相比多光谱TM数据,基于高光谱图像可以更为准确有效地提取土地盐渍化信息;由分类结果图可以看出,高光谱影像土地盐渍化的区分度高于多光谱影像。该研究探索了高光谱图像土地盐渍化信息的提取技术方法,提供了不同盐渍化土地的分布比例数据,可为黄河三角洲滨海盐碱土地资源的科学利用与管理提供决策依据。

卷积神经网络与视觉Transformer联合驱动的跨层多尺度融合网络高光谱图像分类方法

高光谱图像(HSI)分类是地球科学和遥感影像处理任务中最受关注的研究热点之一。近年来,卷积神经网络(CNN)和视觉Transformer相结合的方法,通过综合考虑局部-全局信息,在HSI分类任务中取得了成功。然而,HSI中地物具有丰富的纹理信息和复杂多样的结构,且不同地物之间存在尺度差异。现有的二者结合的方法通常对多尺度地物目标的纹理和结构信息的提取能力有限。为了克服上述局限性,该文提出CNN与视觉Transformer联合驱动的跨层多尺度融合网络HSI分类方法。首先,从结合CNN与视觉Transformer的角度出发,设计了跨层多尺度局部-全局特征提取模块分支,其主要由卷积嵌入的视觉Transformer和跨层特征融合模块构成。具体来说,卷积嵌入的视觉Transformer通过深度融合多尺度CNN与视觉Transformer实现了多尺度局部-全局特征信息的有效提取,从而增强网络对不同尺度地物的关注。进一步地,跨层特征融合模块深度聚合了不同层次的多尺度局部-全局特征信息,以综合考虑地物的浅层纹理信息和深层结构信息。其次,构建了分组多尺度卷积模块分支来挖掘HSI中密集光谱波段潜在的多尺度特征。最后,为了增强网络对HSI中局部波段细节和整体光谱信息的挖掘,设计了残差分组卷积模块对局部-全局光谱特征进行提取。Indian Pines, Houston 2013和Salinas Valley 3个HSI数据集上的实验结果证实了所提方法的有效性。

新一代国产高光谱ZY1-02E卫星在内陆水体水质参数反演中的应用

2021年12月26日,我国成功发射资源一号02E(ZY1-02E)卫星,其搭载了新一代高光谱相机(Advanced Hyperspectral Imager,AHSI),拥有30 m空间分辨率的可见光到短波红外范围内的166个波段,在内陆水体水质参数反演方面具有重要潜力。本研究以北京市沙河水库和金海湖为研究区,开展基于ZY1-02E AHSI影像数据的叶绿素a浓度和透明度反演研究,以评价其实际应用效果。基于京津冀地区12个湖库的遥感反射率和叶绿素a浓度实测数据,构建叶绿素a反演半经验模型。将该模型和基于准解析算法(Quasi-Analytical Algorithm,QAA)的透明度半分析模型应用于ZY1-02E AHSI影像,并利用在沙河水库和金海湖两个研究区获取的星地同步实测水质数据对反演结果进行精度评价。结果表明,基于670 nm和705 nm波长的归一化指数的叶绿素a反演半经验模型的精度最高,拟合度和平均相对误差分别为0.79和21.70%;基于QAA-V6的透明度半分析模型的精度最高,拟合度和平均相对误差分别为0.93和13.90%。该研究结果初步证明了ZY1-02E高光谱数据在内陆水体水质参数反演中的潜力。

实测高光谱和HSI影像的区域土壤含水量遥感监测研究

基于典型研究区植被冠层实测高光谱数据和HSI高光谱影像数据,通过相关分析选择与不同深度土壤含水量响应敏感波段,建立两者的土壤含水量反演模型,并用实测高光谱土壤含水量反演模型校正HSI影像土壤含水量反演的模型。结果表明:土壤含水量响应敏感波段区域为450~650 nm和850~920 nm;两种土壤含水量反演模型对土壤深度为0~10 cm的土壤含水量估算效果最好,其中实测冠层高光谱土壤含水量反演模型精度高于HSI影像土壤含水量反演模型,判定系数(R^2)分别为0.659和0.557;经过校正的HSI影像土壤含水量反演模型精度有了较大的提高,判定系数(R^2)从0.557提升到0.719,均方根误差(RMSE)为0.043 5,较好地提高了区域尺度条件下土壤含水量监测精度,因此运用该方法进行土壤含水量遥感监测是可行的,为进一步提高区域尺度下土壤含水量定量遥感监测提供参考借鉴。

基于环境一号HSI高光谱数据提取叶绿素a浓度的混合光谱分解模型研究

随着遥感技术的应用推广以及对研究精度的要求提高,越来越多的研究注意到混合像元的问题。在水质遥感监测中传感器探测的水体辐射亮度值是纯水和各种水质参数辐射亮度值的叠加,混合像元问题严重影响了水质定量遥感反演的准确性。基于环境一号HSI高光谱数据,首先分析了混合光谱分解模型的物理基础,然后基于采样点浓度大小和PPI(纯净像元指数)方法在遥感影像上提取纯水和叶绿素a的端元波谱,并利用线性光谱分解方法得到叶绿素a的丰度值找丰度值与叶绿素a浓度值之间的统计关系,建立了叶绿素a浓度反演的混合光谱分解模型,且反演精度较高。本文为水质定量遥感提供了一种新的思路。