Hyperspectral Applications Research Articles

Hyperspectral unmixing based on iteratively re-weighted smoothed l<SUB align="right">0 regularisation

Hyperspectral unmixing (HU) plays a fundamental role in a wide range of hyperspectral applications. Sparse-based approach has recently received much attention in hyperspectral unmixing area. Sparse unmixing is based on the assumption that each mixed pixel in the hyperspectral image can be expressed in the form of linear combination of a number of pure spectral signatures that are known in advance. Despite the success of sparse unmixing based on the L0, L1/2 or L0 regulariser, the limitation of this approach on its computational complexity or sparsity affects the efficiency or accuracy. In this paper, as iteratively re-weighted algorithm can help to find the much sparser solution in optimal problems, we proposed a sparse unmixing method based on iteratively re-weighted smoothed L0 regularisation, named RSL0SU. Then a variable splitting and augmented Lagrangian algorithm is introduced to solve the optimisation problem. Comparing to SUnSAL method and SL0SU method, our experimental results with both simulated and real hyperspectral data sets demonstrate that the RSL0SU method is an effective and accurate spectral unmixing algorithm for hyperspectral remote sensing imagery.

Evaluation of Six Algorithms to Monitor Wheat Leaf Nitrogen Concentration

The rapid and non-destructive monitoring of the canopy leaf nitrogen concentration (LNC) in crops is important for precise nitrogen (N) management. Nowadays, there is an urgent need to identify next-generation bio-physical variable retrieval algorithms that can be incorporated into an operational processing chain for hyperspectral satellite missions. We assessed six retrieval algorithms for estimating LNC from canopy reflectance of winter wheat in eight field experiments. These experiments represented variations in the N application rates, planting densities, ecological sites and cultivars and yielded a total of 821 samples from various places in Jiangsu, China over nine consecutive years. Based on the reflectance spectra and their first derivatives, six methods using different numbers of wavelengths were applied to construct predictive models for estimating wheat LNC, including continuum removal (CR), vegetation indices (VIs), stepwise multiple linear regression (SMLR), partial least squares regression (PLSR), artificial neural networks (ANNs), and support vector machines (SVMs). To assess the performance of these six methods, we provided a systematic evaluation of the estimation accuracies using the six metrics that were the coefficients of determination for the calibration (R2C) and validation (R2V) sets, the root mean square errors of prediction (RMSEP) for the calibration and validation sets, the ratio of prediction to deviation (RPD), the computational efficiency (CE) and the complexity level (CL). The following results were obtained: (1) For the VIs method, SAVI(R1200, R705) produced a more accurate estimation of the LNC than other indices, with R²C, R²V, RMSEP, RPD and CE values of 0.844, 0.795, 0.384, 2.005 and 0.10 min, respectively; (2) For the SMLR, PLSR, ANNs and SVMs methods, the SVMs using the first derivative canopy spectra (SVM-FDS) offered the best accuracy in terms of R²C, R²V, RMSEP, RPD, and CE, at 0.96, 0.78, 0.37, 2.02, and 21.17, respectively; (3) The PLSR-FDS, ANN-OS and SVM-FDS methods yield similar accuracies if the CE and CL are not considered, however, ANNs and SVMs performed better on calibration set than the validation set which indicate that we should take more caution with the two methods for over-fitting. Except PLS method, the performance for most methods did not enhance when the spectrum were operated by the first derivative. Moreover, the evaluation of the robustness demonstrates that SVM method may be better suited than the other methods to cope with potential confounding factors for most varieties, ecological site and growth stage; (4) The prediction accuracy was found to be higher when more wavelengths were used, though at the cost of a lower CE. The findings are of interest to the remote sensing community for the development of improved inversion schemes for hyperspectral applications concerning other types of vegetation. The examples provided in this paper may also serve to illustrate the advantages and shortcomings of empirical hyperspectral models for mapping important vegetation biophysical properties of other crops.

On Diverse Noises in Hyperspectral Unmixing

Traditional spectral unmixing methods are usually based on the linear mixture model (LMM) or nonlinear mixture model (NLMM), in which only the additive noise is considered. However, in hyperspectral applications, the additive, multiplicative, and mixed noises play important roles. In this paper, we propose an antinoise model for hyperspectral unmixing. In the antinoise model, all the additive, multiplicative and mixed noises are addressed. To deal with the problems faced by LMM or NLMM and to tackle the antinoise model, an antinoise model based hyperspectral unmixing method is presented, where block coordinate descent is employed to solve an approximated $L_0$ norm constraint, then a nonnegative matrix factorization (NMF) method is presented, which is based on the bounded Itakura–Saito divergence. The experimental results on both synthetic and real hyperspectral data sets demonstrate the efficacy of the proposed model and the corresponding method.

Robust Hyperspectral Unmixing With Correntropy-Based Metric.

Hyperspectral unmixing is one of the crucial steps for many hyperspectral applications. The problem of hyperspectral unmixing has proved to be a difficult task in unsupervised work settings where the endmembers and abundances are both unknown. In addition, this task becomes more challenging in the case that the spectral bands are degraded by noise. This paper presents a robust model for unsupervised hyperspectral unmixing. Specifically, our model is developed with the correntropy-based metric where the nonnegative constraints on both endmembers and abundances are imposed to keep physical significance. Besides, a sparsity prior is explicitly formulated to constrain the distribution of the abundances of each endmember. To solve our model, a half-quadratic optimization technique is developed to convert the original complex optimization problem into an iteratively reweighted nonnegative matrix factorization with sparsity constraints. As a result, the optimization of our model can adaptively assign small weights to noisy bands and put more emphasis on noise-free bands. In addition, with sparsity constraints, our model can naturally generate sparse abundances. Experiments on synthetic and real data demonstrate the effectiveness of our model in comparison to the related state-of-the-art unmixing models.

HYPERSPECTRAL HYPERION IMAGERY ANALYSIS AND ITS APPLICATION USING SPECTRAL ANALYSIS

Abstract. Rapid advancement in remote sensing open new avenues to explore the hyperspectral Hyperion imagery pre-processing techniques, analysis and application for land use mapping. The hyperspectral data consists of 242 bands out of which 196 calibrated/useful bands are available for hyperspectral applications. Atmospheric correction applied to the hyperspectral calibrated bands make the data more useful for its further processing/ application. Principal component (PC) analysis applied to the hyperspectral calibrated bands reduced the dimensionality of the data and it is found that 99% of the data is held in first 10 PCs. Feature extraction is one of the important application by using vegetation delineation and normalized difference vegetation index. The machine learning classifiers uses the technique to identify the pixels having significant difference in the spectral signature which is very useful for classification of an image. Supervised machine learning classifier technique has been used for classification of hyperspectral image which resulted in overall efficiency of 86.6703 and Kappa co-efficient of 0.7998.

A novel two-step method for winter wheat-leaf chlorophyll content estimation using a hyperspectral vegetation index

Remote sensing estimation of leaf chlorophyll content is of importance to crop nutrition diagnosis and yield assessment, yet the feasibility and stability of such estimation has not been assessed thoroughly for mixed pixels. This study analyses the influence of spectral mixing on leaf chlorophyll content estimation using canopy spectra simulated by the PROSAIL model and the spectral linear mixture concept. It is observed that the accuracy of leaf chlorophyll content estimation would be degraded for mixed pixels using the well-accepted approach of the combination of transformed chlorophyll absorption index (TCARI) and optimized soil-adjusted vegetation index (OSAVI). A two-step method was thus developed for winter wheat chlorophyll content estimation by taking into consideration the fractional vegetation cover using a look-up-table approach. The two methods were validated using ground spectra, airborne hyperspectral data and leaf chlorophyll content measured the same time over experimental winter wheat fields. Using the two-step method, the leaf chlorophyll content of the open canopy was estimated from the airborne hyperspectral imagery with a root mean square error of 5.18 μg cm−2, which is an improvement of about 8.9% relative to the accuracy obtained using the TCARI/OSAVI ratio directly. This implies that the method proposed in this study has great potential for hyperspectral applications in agricultural management, particularly for applications before crop canopy closure.

Quantitative hyperspectral analysis for characterization of the coastal water from Damietta to Port Said, Egypt

The coastal zone of the Nile Delta is wealthy of natural resources that currently accommodate and support not less than one third of the population. However, it is still a promising area for energy resources and industrial activities that initiated the country to develop new action plans for industrial estates and services; it is under extreme environmental pressure. The newly developed techniques of remote sensing offer effective tool to monitor such environment on a near-real time basis at minimum in situ sampling. Detection of water quality conditions and parameters is one of the major advantages of hyperspectral remote sensing technologies; which has been widely used to assess water quality conditions of many open water mass and aquatic ecosystems. This research looked at characterization of the water quality using the different functions of in situ measurements of hyperspectral sensors in conjunction with water quality multi-probe sensor. The spectral behaviour of the water at the outlets is quite different from the other sampling sites of the coastal water. Turbidity was recorded at high level of 65 (NTU) and chlorophyll was up to 147mg/l at the outlets, which were positively correlated with the hyperspectral curve of the spectroradiometer and Hyperion data. The archived Hyperion data showed some similarities and differences with the in situ data that were mostly due to the timing differences between both datasets; it, however, enabled for understanding of the hyperspectral application in water quality.

Parallel Hyperspectral Unmixing on GPUs

This letter presents a new parallel method for hyperspectral unmixing composed by the efficient combination of two popular methods: vertex component analysis (VCA) and sparse unmixing by variable splitting and augmented Lagrangian (SUNSAL). First, VCA extracts the endmember signatures, and then, SUNSAL is used to estimate the abundance fractions. Both techniques are highly parallelizable, which significantly reduces the computing time. A design for the commodity graphics processing units of the two methods is presented and evaluated. Experimental results obtained for simulated and real hyperspectral data sets reveal speedups up to 100 times, which grants real-time response required by many remotely sensed hyperspectral applications.

Continuous iterative guided spectral class rejection classification algorithm for hyperspectral data

This paper discusses a hyperspectral application of the continuous iterative guided spectral class rejection (CIGSCR) classification method based on the iterative guided spectral class rejection (IGSCR) classification method for remotely sensed data. Both CIGSCR and IGSCR use semisupervised clustering to locate clusters that are associated with classes in a classification scheme. In CIGSCR and IGSCR, training data are used to evaluate the strength of the association between a particular cluster and a class, and a statistical hypothesis test is used to determine which clusters should be associated with a class and used for classification and which clusters should be rejected and possibly refined. Experimental results indicate that the soft classification output by CIGSCR is reasonably accurate (when compared to IGSCR), and the fundamental algorithmic changes in CIGSCR (from IGSCR) result in CIGSCR being less sensitive to input parameters that influence iterations, as well as being highly parallelizable. Furthermore, evidence is presented that the semisupervised clustering in CIGSCR produces more accurate classifications than classification based on clustering without supervision.

A Background Self-Learning Framework for Unstructured Target Detectors

Unstructured background model based detectors have been successfully applied in various hyperspectral target detection applications. The background statistics of an image can be estimated in a global way or a local way. The global approach involves modeling the background directly from the whole image, which can prove to be inaccurate due to target contamination of the background information. The local approach usually involves estimating the background statistics using a spatially sliding local window. However, this approach can also fail to reflect reality, due to sensitive parameters, like the window size, and presents high computational costs. This letter proposes a self-learning method to adaptively determine the background statistics for unstructured detectors, with the consideration of exploiting both the spatial and spectral information, and accelerating the computation speed. The experimental results with two real hyperspectral images confirm the superior performance when compared to the other two approaches to modeling background statistics.

Research on the Band Selection Algorithm Based on Noise Evaluation of Hyperspectral Image

Along with the development of hyperspectral remote sensing technology, band selection algorithm of hyperspectral image has become the research focus of hyperspectral application. A band selection algorithm based on noise evaluation of hyperspectral image and named as LMLSD is proposed in this paper. All the bands of hyperspectral image are ranked according to the SNR calculated based on local mean and local standard deviation of every band. Simulation results show that the performance of LMLSD is better than classical band selection algorithms of ABS and SDAA. The algorithm of LMLSD has validity and feasibility in practical application.

Adaptive noise variance estimation and intrinsic order selection for low SNR hyperspectral signals

In hyperspectral applications, signal vectors belong to a much lower dimensional subspace than the observed data. The true dimensionality of hyperspectral data is difficult to determine in practice. In the presence of powerful noise, estimation of the number of spectrally distinct signal sources that characterize the hyperspectral data is a challenge. Existing methods mostly assume some prior knowledge of the noise and signal structure. In practice, there is no a priori knowledge of the noise or signal statistics. Eigenvalue decomposition based methods, as the most commonly used methods, assume that the contribution of noise to the signal is extractable. In a noisy hyperspectral application this assumption is questionable. In this paper, we propose the Adaptive Noise Variance Estimation and Intrinsic Order Selection method that exploits the concept of residual autocorrelation power to adaptively estimate the noise variance and then simultaneously denoise and estimate the rank of the hyperspectral data. Rank conjecture is obtained by locating an optimum subset that best represents the noiseless signal. The algorithm was applied to both synthetically simulated data and to a real hyperspectral image. Comparing the results with those of existing methods indicates that this method will substantially improve the accuracy of the rank estimation in extremely noisy hyperspectral applications.

Broadband QWIP FPAs for hyperspectral applications

Is a quantum well infrared photodetector (QWIP) suitable for hyperspectral applications? Yes, provided its spectral response is broad enough, the signal-to-noise ratio in narrow spectral slices across the entire response band is high enough, and the operating temperature is reasonable. We present the performance of a 640×512 QWIP focal plane array (FPA) with nearly uniform spectral response from 8 to 12μm that achieves temporal noise equivalent temperature difference (NETD) lower than 200mK in 20nm wide spectral slices across the entire 8–12μm band with the FPA operating at a temperature of 50K or colder. A hyperspectral camera would feature a grating or prism to disperse 8–12μm light across such a FPA or a hyperspectral filter on top of the FPA. Operating in pushbroom mode on an earth-observing satellite, the camera can be used for spectral mapping of vegetation, crops and forests, for pollution monitoring, and for studies in atmospheric chemistry.

Parameters optimization for wavelet denoising based on normalized spectral angle and threshold constraint machine learning

Wavelet parameters (e.g., wavelet type, level of decomposition) affect the perfor- mance of the wavelet denoising algorithm in hyperspectral applications. Current studies select the best wavelet parameters for a single spectral curve by comparing similarity criteria such as spectral angle (SA). However, the method to find the best parameters for a spectral library that contains multiple spectra has not been studied. In this paper, a criterion named normalized spectral angle (NSA) is proposed. By comparing NSA, the best combination of parameters for a spectral library can be selected. Moreover, a fast algorithm based on threshold constraint and machine learning is developed to reduce the time of a full search. After several iterations of learning, the combination of parameters that constantly surpasses a threshold is selected. The experiments proved that by using the NSA criterion, the SA values decreased significantly, and the fast algorithm could save 80% time consumption, while the denoising performance was not obviously impaired. © 2012 Society of Photo-Optical Instrumentation Engineers (SPIE). (DOI: 10.1117/ 1.JRS.6.063579)

Hyperspectral target detection via discrete wavelet-based spectral fringe-adjusted joint transform correlation

Spectral variability remains a major challenge for target detection in hyperspectral imagery (HSI). Recently, the spectral fringe-adjusted joint transform correlation (SFJTC) technique has been used effectively for hyperspectral target detection applications. In this paper, we propose to use discrete wavelet transform (DWT) coefficients of the signatures as features for detection in order to make the SFJTC technique more insensitive to spectral variability. We devised a supervised training algorithm that uses the pure target signature and randomly selected samples from input scenery to select an optimal set of DWT coefficients for detection. We have inserted target signatures into urban and vegetative hyperspectral scenery with varying levels of spectral variability to explore the performance of our DWT-based SFJTC technique in different operating conditions. Detection results in the form of receiver-operating-characteristic (ROC) curves and area-under-the-ROC (AUROC) curves show that the proposed scheme yields the largest mean AUROC values compared to SFJTC using the original signatures and traditional hyperspectral detection algorithms.

Simultaneous Denoising and Intrinsic Order Selection in Hyperspectral Imaging

In this paper, we address the problem of order selection in noisy hyperspectral applications. In conventional unmixing methods, this problem has been divided into two separate processes of order selection and unmixing. Order selection methods generally use a denoising approach at the beginning stage. The data in this case pass through three stages: denoising, order selection, and unmixing. Each of these steps mainly aims to optimize a different criterion independently. In addition, any error created in the denoising process will be propagated not only to the order selection stage but also consequently to the unmixing results. Commonly used denoising methods such as eigenvalue-decomposition-based methods, e.g., singular-value-decomposition-based methods, provide a threshold value to separate the noise from the signal. These approaches are heavily sensitive to the threshold value and signal-to-noise ratio (SNR). Moreover, these methods tend to lose their efficiency rapidly for lower SNRs. Note that both the denoising step and the dimension estimation step aim to provide the optimum estimate of the same noiseless data. Consequently, adopting a simultaneous denoising and dimension estimation method with a goal to provide the optimum estimate of the desired noiseless data is rational. This process not only avoids possible error propagations from the denoising stage to the dimension estimation stage but also unifies the optimization criteria that were used in each of these steps. In this paper, a simultaneous denoising and dimension estimation method is introduced. The approach is based on minimizing the estimated mean square error. Minimization is done by comparing the estimated data in a range of subspaces dictated by a simultaneous process. Minimizing the error at once, the proposed method denoises the data and provides the optimum dimension simultaneously. Owing to the parallel processing of denoising and dimension estimation, the simulation results show the advantages of the proposed method over some of the state-of-the-art approaches and illustrate a substantial performance, particularly for cases with a lower SNR.

Comparison of measured and satellite-derived spectral diffuse attenuation coefficients for the Arabian Sea

We present here the results of our study comparing the spectral diffuse attenuation coefficients K d(λ) measured in the Arabian Sea with those derived from the Sea-Viewing Wide Field-of-View Sensor (SeaWiFS) using three algorithms, of which two are empirical-data-driven and one is semi-analytical. The measurements were carried out in all water types and the mean values of the measured spectral K d(λ) are 0.105, 0.092, 0.077, 0.082, 0.110 and 0.490 m−1 for wavelength λ at 412, 443, 490, 510, 555 and 670 nm, respectively. This profile corresponds to a chlorophyll value of about 1 mg m−3. The maximum values of the measured K d correspond to waters with chlorophyll of about 8 mg m−3. Though the satellite-derived K d(λ) are found to be overestimated in all bands, we have observed good correlations between the measured and satellite-derived values in all bands, and excluding the band at 670 nm, the mean absolute percent deviations are observed to be less than 50% in all bands. The performance of the data-driven empirical methods was found to be consistent in all the bands, except at the red band of 670 nm, which is uncorrelated with the measured values and has large errors. The performances of the empirical methods depend on the accuracy of the band ratios of the retrieved remote sensing reflectance. Though the performance of the semi-analytical algorithm is found to be spectrally varying, with large positive bias observed in the blue regions, this algorithm is recommended for hyperspectral applications. The performance of the semi-analytical algorithm could be improved by having a robust algorithm to accurately derive spectral inherent optical properties of absorption and backscattering coefficients from the satellite data.

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).

Hyperspectral opportunities with next-generation QSIP arrays

Hyperspectral applications that employ gratings typically exploit only one grating order, since other orders that emerge from the grating at the same angle of dispersion would corrupt the spectrum measured with conventional FPAs. Previously, dualband infrared focal plane arrays (FPA), developed for multi-spectral imaging applications, have demonstrated advantages over conventional multi-FPA sensor configurations in compactness and band-to-band pixel registration. In addition, dualband and multiband FPA architectures of Quantum Structure Infrared Photodetector (QSIP) technology would enable applications beyond simple multi-spectral imaging. For example, In the case of dual- (or multi-) waveband FPAs, the different grating orders can be paired with the FPA wavebands, allowing high efficiency hyperspectral imaging over very broad wavelength regions. Exploiting the “third dimension” of FPA detecting layers for this type of hyperspectral application has been demonstrated previously. As time progresses, multi-waveband FPAs are expected to provide an increase in spectral information at the pixel level without the need for external (e.g. dispersive) optical elements. As the number of wavebands increases to the point of providing spectral overlap of adjacent spectral resolution elements, hyperspectral capability is then achieved by the FPA acting alone. This technology may someday become possible through advanced QSIP architectures having photons of different wavelength continuously absorbed at different depths, and their resulting photocurrents isolated with a vertical grid of contacts or an equivalent mechanism for transporting depth-dependent signal photocurrent to a read-out circuit unit cell.

Spectral and Spatial Complexity-Based Hyperspectral Unmixing

Hyperspectral unmixing, which decomposes pixel spectra into a collection of constituent spectra, is a preprocessing step for hyperspectral applications like target detection and classification. It can be considered as a blind source separation (BSS) problem. Independent component analysis, which is a widely used method for performing BSS, models a mixed pixel as a linear mixture of its constituent spectra weighted by the correspondent abundance fractions (sources). The sources are assumed to be independent and stationary. However, in many instances, this assumption is not valid. In this paper, a complexity-based BSS algorithm is introduced, which studies the complexity of sources instead of the independence. We extend the 1-D temporal complexity, which is called complexity pursuit that was proposed by Stone, to the 2-D spatial complexity, which is named spatial complexity BSS (SCBSS), to describe the spatial autocorrelation of each abundance fraction. Further, the temporal complexity of spectrum is combined into SCBSS to account for the spectral smoothness, which is termed spectral and spatial complexity BSS. More importantly, a strict theoretic interpretation is given, showing that the complexity-based BSS is very suitable for hyperspectral unmixing. Experimental results on synthetic and real hyperspectral data demonstrate the advantages of the proposed two algorithms with respect to other methods.