Ben Somers

Automated Extraction of Image-Based Endmember Bundles for Improved Spectral Unmixing

Spectral unmixing is an important task in hyperspectral data exploitation. It amounts to estimating the abundance of pure spectral constituents (endmembers) in each (possibly mixed) observation collected by the imaging instrument. In recent years, several endmember extraction algorithms (EEAs) have been proposed for automated endmember extraction from hyperspectral data sets. Traditionally, EEAs extract/select only one single standard endmember spectrum for each of the presented endmember classes or scene components. The use of fixed endmember spectra, however, is a simplification since in many cases the conditions of the scene components are spatially and temporally variable. As a result, variation in endmember spectral signatures is not always accounted for and, hence, spectral unmixing can lead to poor accuracy of the estimated endmember fractions. Here, we address this issue by developing a simple strategy to adapt available EEAs to select multiple endmembers (or bundles) per scene component. We run the EEAs in randomly selected subsets of the original hyperspectral image, and group the extracted samples of pure materials in a bundle using a clustering technique. The output is a spectral library of pure materials, extracted automatically from the input scene. The proposed technique is applied to several common EEAs and combined with an endmember variability reduction technique for unmixing purposes. Experiments with both simulated and real hyperspectral data sets indicate that the proposed strategy can significantly improve fractional abundance estimations by accounting for endmember variability in the original hyperspectral data.

MUSIC-CSR: Hyperspectral Unmixing via Multiple Signal Classification and Collaborative Sparse Regression

Spectral unmixing aims at finding the spectrally pure constituent materials (also called endmembers) and their respective fractional abundances in each pixel of a hyperspectral image scene. In recent years, sparse unmixing has been widely used as a reliable spectral unmixing methodology. In this approach, the observed spectral vectors are expressed as linear combinations of spectral signatures assumed to be known a priori and presented in a large collection, termed spectral library or dictionary, usually acquired in laboratory. Sparse unmixing has attracted much attention as it sidesteps two common limitations of classic spectral unmixing approaches, namely, the lack of pure pixels in hyperspectral scenes and the need to estimate the number of endmembers in a given scene, which are very difficult tasks. However, the high mutual coherence of spectral libraries, jointly with their ever-growing dimensionality, strongly limits the operational applicability of sparse unmixing. In this paper, we introduce a two-step algorithm aimed at mitigating the aforementioned limitations. The algorithm exploits the usual low dimensionality of the hyperspectral data sets. The first step, which is similar to the multiple signal classification array signal processing algorithm, identifies a subset of the library elements, which contains the endmember signatures. Because this subset has cardinality much smaller than the initial number of library elements, the sparse regression we are led to is much more well conditioned than the initial one using the complete library. The second step applies collaborative sparse regression, which is a form of structured sparse regression, exploiting the fact that only a few spectral signatures in the library are active. The effectiveness of the proposed approach, termed MUSIC-CSR, is extensively validated using both simulated and real hyperspectral data sets.

A weighted linear spectral mixture analysis approach to address endmember variability in agricultural production systems

The least squares error (LSE) technique is frequently used to estimate abundance fractions in linear spectral mixture analysis (LSMA). The LSE is typically equally weighted for all wavebands, assuming equally important effects. This is, however, not always the case and therefore traditional LSMA often results in suboptimal fraction estimates. This study presents a weighted LSMA approach that prioritises wavebands with minor or no negative effects on fraction estimates. Synthetic mixed pixel spectra compiled from in situ measured spectra of bare soil, citrus tree and weed canopies were used for validation. The results show markedly improved fraction estimates obtained for the weighted approach, with a mean absolute gain of 0.24 in R 2 and a mean absolute reduction in fraction abundance error of 0.06.

A Quantitative Analysis of Virtual Endmembers' Increased Impact on the Collinearity Effect in Spectral Unmixing

In the past decades, spectral unmixing has been studied for deriving the fractions of spectrally pure materials in a mixed pixel. However, limited attention has been given to the collinearity problem in spectral mixture analysis. In this paper, quantitative analysis and detailed simulations are provided, which show that the high correlation between the endmembers, including the virtual endmembers introduced in a nonlinear model, has a strong impact on unmixing errors through inflating the Gaussian noise. While distinctive spectra with low correlations are often selected as true endmembers, the virtual endmembers formed by their product terms can be highly correlated. It is found that a virtual-endmember-based nonlinear model generally suffers more from collinearity problems compared to linear models and may not perform as expected when the Gaussian noise is high, despite its higher modeling power. Experiments were conducted on a set of in situ measured data, and the results show that the linear mixture model performs better in 61.5% of the cases.

An automated waveband selection technique for optimized hyperspectral mixture analysis

Linear spectral mixture analysis (SMA) has been used extensively in remote sensing studies to estimate the sub-pixel composition of spectral mixtures. The lack of ability to account for sufficient temporal and spatial variability between and among ground component or endmember spectra has been acknowledged as a major shortcoming of conventional SMA approaches. In an attempt to overcome this problem, a novel and automated linear spectral mixture protocol, referred to as stable zone unmixing (SZU), is presented and evaluated. Stable spectral features (i.e. least sensitive to spectral variability) are automatically selected for use in the mixture analysis based on a minimum InStability Index (ISI) criterion. ISI is defined as the ratio of the spectral variability within and the spectral variability among the endmember classes that are present within the mixture. The algorithm was tested on a set of scenarios, generated from in situ measured hyperspectral data. The scenarios covered both urban and natural environments under differing conditions. SZU provided reliable endmember cover distribution maps in all scenarios. On average, an absolute gain in R2—the coefficient of determination of the modelled versus the observed sub-pixel cover fractions—of 0.14 over the traditional SMA approaches was observed while the absolute gain in fraction abundance error was 0.06. It was concluded that the SZU protocol has potential to be an effective and efficient SMA algorithm for generating optimal cover fraction estimates regardless of the scenario considered. Moreover, the subset selection protocol, as implemented in SZU, can be regarded as complementary to conventional SMA approaches resulting in a further reduction of spectral variability.

Hyperspectral indices to diagnose leaf biotic stress of apple plants, considering leaf phenology

Novel and existing hyperspectral vegetation indices were evaluated in this study, with the aim of assessing their utility for accurate tracking of leaf spectral changes due to differences in biophysical indicators caused by apple scab. Novel indices were extracted from spectral profiles by means of narrow‐waveband ratioing of all possible two‐band combinations between 350 nm and 2500 nm at nanometer intervals (2 311 250 combinations) and all possible two‐band derivative combinations. Narrow‐waveband ratios consisting of wavelengths of approximately 1500 nm and 2250 nm, associated with water content, have proven to be the most appropriate for detecting apple scab at early developmental stages. Logistic regression c‐values ranged from 0.80 to 0.88. At a more developed infection stage, vegetation indices such as R440/R690 and R695/R760 exhibited superior distinction between non‐infected and infected leaves. Identified derivative indices were located in similar regions. It therefore was concluded that the most appropriate indices at early stages of infection are ratios of wavelengths situated at the water band slopes. The choice of appropriate indices and their discriminatory performances, however, depended on the phenological stage of the leaves. Hence, an undisturbed 20‐day growth profile was examined to assess the effect of physiological changes on spectral variations at consecutive growth stages of leaves. Results suggested that an accurate distinction could be made between different leaf developmental stages using the 570 nm, 1460 nm, 1940 nm and 2400 nm wavelengths, and the red‐edge inflection point. These results are useful to crop managers interested in an early warning system to aid proactive system management and steering.

Spectral mixture analysis to monitor defoliation in mixed-aged Eucalyptus globulus Labill plantations in southern Australia using Landsat 5-TM and EO-1 Hyperion data

Defoliation is a key parameter of forest health and is associated with reduced productivity and tree mortality. Assessing the health of forests requires regular observations over large areas. Satellite remote sensing provides a cost-effective alternative to traditional ground-based assessment of forest health, but assessing defoliation can be difficult due to mixed pixels where vegetation cover is low or fragmented. In this study we apply a novel spectral unmixing technique, referred to as weighted Multiple Endmember Spectral Mixture Analysis (wMESMA), to Landsat 5-TM and EO-1 Hyperion data acquired over a Eucalyptus globulus (Labill.) plantation in southern Australia. This technique combines an iterative mixture analysis cycle allowing endmembers to vary on a per pixel basis (MESMA) and a weighting algorithm that prioritizes wavebands based on their robustness against endmember variability. Spectral mixture analysis provides an estimate of the physically interpretable canopy cover, which is not necessarily correlated with defoliation in mixed-aged plantations due to natural variation in canopy cover as stands age. There is considerable variability in the degree of defoliation as well as in stand age among sites and in this study we found that results were significantly improved by the inclusion of an age correction algorithm for both the multi-spectral (R2no age correction = 0.55 vs R2age correction = 0.73 for Landsat) and hyperspectral (R2no age correction = 0.12 vs R2age correction = 0.50 for Hyperion) image data. The improved accuracy obtained from Landsat compared to the Hyperion data illustrates the potential of applying SMA techniques for analysis of multi-spectral datasets such as MODIS and SPOT-VEGETATION.

A conceptual framework for the simultaneous extraction of sub-pixel spatial extent and spectral characteristics of crops.

The sub-pixel spectral contribution of background soils and shadows hampers the accurate site-specific monitoring of agricultural crop characteristics from aerial or satellite images. To address this problem, the present study combines measured in situ and hyperspectral data in an alternative unmixing algorithm. The proposed algorithm, referred to as Soil Modeling Mixture Analysis (SMMA), incorporates a soil reflectance model in a traditional unmixing algorithm and as such opens up the opportunity to simultaneously extract the sub-pixel spatial extent of crops as well as its pure spectral information. The performance of the algorithm is evaluated using a soil moisture reflectance model, calibrated for an in situ measured Albic Luvisol dataset. Synthetic mixtures, i.e., compiled from in situ measured hyperspectral bare soil and citrus tree canopy spectra, were decomposed and the sub-pixel crop cover fractions (R 2 � 0.94, RMSE � 0.03) and pure vegetation signals (average extraction error 350 to 2,500 nm � 0.017, RMSE � 0.02) were adequately extracted from the mixtures.

Hyperspectral Time Series Analysis of Native and Invasive Species in Hawaiian Rainforests

Abstract: The unique ecosystems of the Hawaiian Islands are progressively being threatened following the introduction of exotic species. Operational implementation of remote sensing for the detection, mapping and monitoring of these biological invasions is currently hampered by a lack of knowledge on the spectral separability between native and invasive species. We used spaceborne imaging spectroscopy to analyze the seasonal dynamics of the canopy hyperspectral reflectance properties of four tree species: (i) Metrosideros polymorpha , a keystone native Hawaiian species; (ii) Acacia koa , a native Hawaiian nitrogen fixer; (iii) the highly invasive Psidium cattleianum ; and (iv) Morella faya , a highly invasive nitrogen fixer. The species specific separability of the reflectance and derivative-reflectance signatures extracted from an Earth Observing-1 Hyperion time series, composed of 22 cloud-free images spanning a period of four years and was quantitatively evaluated using the Separability Index (SI). The analysis revealed that the Hawaiian native trees were universally unique from the invasive trees in their near-infrared-1 (700–1,250 nm) reflectance (0.4 > SI > 1.4). Due to its higher leaf area index, invasive trees generally had a higher near-infrared reflectance. To a lesser extent, it could also be demonstrated that nitrogen-fixing trees were spectrally unique from non-fixing trees. The higher leaf nitrogen content of nitrogen-fixing trees was expressed through slightly increased separabilities in visible and shortwave-infrared reflectance wavebands (SI = 0.4). We also found phenology to be key to spectral separability analysis. As such, it was shown that the spectral separability in the near-infrared-1 reflectance

Invasive Species Mapping in Hawaiian Rainforests Using Multi-Temporal Hyperion Spaceborne Imaging Spectroscopy

We evaluated the potential of multi-temporal Multiple Endmember Spectral Mixture Analysis (MESMA) of Earth Observing-1 Hyperion data for detection of invasive tree species in the montane rainforest area of the Hawaii Volcanoes National Park, Island of Hawaii. We observed a clear seasonal trend in invasive species detection success when unmixing results were cross-referenced to ground observations; with Kappa coefficients (indicating detection success, 0-1) ranging between 0.66 (summer) and 0.69 (winter) and 0.51-0.53 during seasonal transition periods. An increase of Kappa to 0.80 was observed when spectral features extracted from September, August and January were integrated into MESMA. Multi-temporal unmixing improved the detection success of invasive species because spectral information acquired over different portions of the growing season allowed us to capture species-specific phenology, thereby reducing spectral similarity among species.