We have previously introduced a high throughput multiplexing computational spectral imaging device. The device measures scalar projections of pseudo-arbitrary spectral filters at each spatial pixel. This paper discusses simulation and initial experimental progress in performing computational spectral unmixing by taking advantage of the natural sparsity commonly found in the fractional abundances. The simulation results show a lower unmixing error compared to traditional spectral imaging devices. Initial experimental results demonstrate the ability to directly perform spectral unmixing with less error than multiplexing alone.

Wetlands are the most important and valuable ecosystems on Earth. They are called “kidneys of the Earth”. Vegetation change detection is necessary to understand the condition of a wetland and to support ecosystem sustainable management and utilization. It has been a great challenge to estimate vegetation (including bald cypress trees) coverage of the wetland because it is difficult to access directly. Satellite remote sensing technology can be one important feasible method to map and monitor changes of wetland forest vegetation and land cover over large areas. Remote sensing mapping techniques have been applied to detect and map vegetation changes in wetlands. To address spectral mixture issues associated with moderate resolution remote sensing images, many spectral mixture methods have been developed and applied to unmix the mixed pixels in order to accurately map endmembers (e.g., different land cover types and different materials within pixels) fractions or abundance. Of them, Mixture Tuned Matched Filtering (MTMF) is an advanced spectral unmixing method that has attracted many researchers to test it for mapping land cover types including mapping tree species with medium or coarse remote sensing image data. MTMF is a partial unmixing method that suppresses background noise and estimates the subpixel abundance of a single target material. In this study, to understand impacts of anthropogenic (e.g., urbanization) and natural forces/climate change on the bald cypress tree dynamic change, the bald cypress trees cover change in University of South Florida Forest Preserve Area was mapped and analysed by using MTMF tool and multitemporal Landsat imagery over 30 years from 1984 to 2015. To evaluate the MTMF’s performance, a tradition spectral unmixing method, Linear Spectral Unmixing (LSU), was also tested. The experimental results indicate that (1) the bald cypress tree cover percentage in the study area has generally increased during the 30 years from 1984 to 2015, but over the time period from 1994 to 2005, the bald cypress tree cover percentage reduced; (2) MTMF tool outperformed the LSU method in mapping the change of the bald cypress trees over the 30 years to demonstrate its powerful capability; and (3) there potentially exists an impact of human activities on the change of the bald cypress trees although a further quantitative analysis is needed in the future research.

Remote sensing has the potential to monitor spatial variation in water quality over large areas. While ocean colour work has developed analytical bio-optical water quality retrieval algorithms for medium spatial resolution platforms, remote sensing of lake water is often limited to high spatial resolution satellites such as Landsat, which have limited spectral resolution. This thesis presents the results of an investigation into satellite monitoring of lake water quality. The aim of this investigation was to ascertain the feasibility of estimating water quality and its spatial distribution using Landsat 7 ETM+ imagery combined with in situ data from Rotorua and Waikato lakes. For the comparatively deep Rotorua lakes, r² values of 0.91 (January 2002) and 0.83 (March 2002) were found between in situ chlorophyll (chl) a and the Band1/Band3 ratio. This technique proved useful for analysing the spatial distribution of phytoplankton, especially in lakes Rotoiti and Rotoehu. For the more bio-optically complex shallow lakes of the Waikato, a linear spectral unmixing (LSU) approach was investigated where the water surface reflectance spectrum is defined by the contribution from pure pixels or endmembers. The model estimates the percentage of the endmember within the pixel, which is then used in a final regression with in situ data to map water quality in all pixels. This approach was used to estimate the concentration of chl a (r² = 0.84). Total suspended solid (TSS) concentration was mapped using the traditional Band 3 regression with in situ data, which combined atmospherically corrected reflectance for both images into a single relationship (r² = 0.98). The time difference between in situ data collection and satellite data capture is a potential source of error. Other potential sources of error include sample location accuracy, the influence of dissolved organic matter, and masking of chl a signatures by high concentrations of TSS. The results from this investigation suggest that remote sensing of water quality provides meaningful and useful information with a range of applications and could provide information on temporal spatial variability in water quality.

The majority of the impact of climate and human induced changes on forest are related to climate variability and deforestation. Similarly, changes in forest phenology due to climate variability and deforestation has been recognized as being among the most important early indicators of the impact of environmental change on forest ecosystem functioning. Comprehensive data on baseline forest cover changes including deforestation is required to provide background information needed for governments to make decision on Reducing Emissions from Deforestation and Forest Degradation (REED). Despite the fact that Nigeria ranks among the countries with highest deforestation rates based on Food and Agricultural Organization estimates, only a few studies have aimed at mapping forest cover changes at country scales. However, recent attempts to map baseline forest cover and deforestation in Nigeria has been based on global scale remote sensing techniques which do not confirm with ground based observations at country level. The aim of this study is two-fold: firstly, baseline forest cover was estimated using an ‘adaptive’ remote sensing model that classified forest cover with high accuracies at country level for the savanna and rainforest zones. The first part of this study also compared the potentials of different MODIS data in detecting forest cover changes at regional (cluster level) scale. The second part of this study explores the trends and response of forest phenology to rainfall across four forest clusters from 2002 to 2012 using vegetation index data from the MODIS and rainfall data obtained from the TRMM.

The master thesis deals with subpixel classification of hyperspectral data from senzor APEX. In the first part there is research from the literature describing algorithms of the subpixel classifications and spectral characteristics of the vegetation. In the practical part there is a work focusing on the classification of the areas with the cover of Norway Spruce trees at eight areas in the Krkonoše national park. Three methods of supervised classification were used - Linear Spectral Unmixing, Support Vector Machine and Spectral Angle Mapper. Field data, spectral curves for exact trees from the eight areas obtained by the contact probe ASD FieldSpec 4 Wide-Res, were used for the extraction of endmembers of the spruces. For each research area maps of land cover were produced by means of the classification methods described above and the accuracies of the classifications were evaluated. Powered by TCPDF (www.tcpdf.org)

Les acquisitions de télédétection ont des caractéristiques complémentaires en termes de résolution spatiale et temporelle et peuvent mesurer différents aspects de la couverture neigeuse (propriétés physiques de surface, type de neige, etc.). En combinant plusieurs acquisitions, il devrait être possible d'obtenir un suivi précis et continu de la neige. Cependant, cet objectif se heurte à la complexité du traitement des images satellites et à la confusion possible entre les différents matériaux observés. Plus particulièrement, l’accès à l’information fractionnelle, c’est-à-dire à la proportion de neige dans chaque pixel, nécessite de retrouver la proportion de l’ensemble des matériaux qui se trouvent dans celui-ci. Ces proportions sont accessibles via des méthodes d’inversions ou démélange spectral se basant sur la résolution spectrale des images obtenues. Le défi général est alors d’arriver à exploiter correctement les différentes informations de natures différentes qui nous sont apportées par les différentes acquisitions afin de produire des cartes d’enneigement précises. Les objectifs de la thèse sont alors au nombre de trois et peuvent se résumer par trois grandes interrogations qui permettent de traiter les différents points évoqués:- Quelles sont les limitations actuelles de l’état de l'art pour l’observation spatiale optique de la neige ?- Comment exploiter les séries temporelles pour s’adapter à la variabilité spectrale des matériaux ?- Est-il possible de généraliser la fusion de données pour une acquisition multimodale à partir de capteurs optiques ?Une étude complète des différents produits de neige issus du satellite MODIS est ainsi proposée, permettant l’identification des nombreuses limitations dont la principale est le haut taux d’erreurs lors de la reconstitution de la fraction (environ 30%). Parmi ces résultats sont notamment identifiés des problèmes liés aux méthodes de démélange face à la variabilité spectrale des matériaux. Face à ces limitations nous avons exploité les séries temporelles MODIS pour proposer une nouvelle approche d’estimation des endmembers, étape critique du démélange spectral. La faible évolution temporelle du milieu (hors neige) est alors utilisée pour contraindre l’estimation des endmembers non seulement sur l’image d’intérêt, mais également sur les images des jours précédents. L’efficacité de cette approche bien que démontrée ici reste sujette aux limitations de résolution spatiale intrinsèques au capteur. Des expérimentations sur la fusion de donnée, à même de pouvoir améliorer la qualité des images, ont par conséquent été réalisées. Devant les limitations de ces méthodes dans le cas des capteurs multispectraux utilisés, une nouvelle approche de fusion a été proposée. Via la formulation d’un nouveau modèle et sa résolution, la fusion entre des capteurs optiques de tous types peut être réalisée sans considération de recouvrement spectral. Les différentes expérimentations sur l’estimation de cartes de neige montrent un intérêt certain d’une meilleure résolution spatiale pour isoler les zones enneigées. Ce travail montre ainsi les nouvelles possibilités de développement pour l’observation de la neige, mais également les évolutions de l’utilisation combinée des images satellites pour l’observation de la Terre en général. / Remote sensing acquisitions have complementary characteristics in terms of spatial and temporal resolution and can measure different aspects of snow cover (e.g., surface physical properties and snow type). By combining several acquisitions, it should be possible to obtain a precise and continuous monitoring of the snow. However, this task has to face the complexity of processing satellite images and the possible confusion between different materials observed. In particular, the estimation of fractional information, i.e., the amount of snow in each pixel, requires to know the proportion of the materials present in a scene. These proportions can be obtained performing spectral unmixing. The challenge is then to effectively exploit the information of different natures that are provided by the multiple acquisitions in order to produce accurate snow maps.Three main objectives are addressed by this thesis and can be summarized by the three following questions:- What are the current limitations of state-of-the-art techniques for the estimation of snow cover extent from optical observations?- How to exploit a time series for coping with the spectral variability of materials?- How can we take advantage of multimodal acquisitions from optical sensors for estimating snow cover maps?A complete study of the various snow products from the MODIS satellite is proposed. It allows the identification of numerous limitations, the main one being the high rate of errors during the estimation of the snow fraction (approximately 30%).The experimental analysis allowed to highlight the sensitivity of the spectral unmixing methods against the spectral variability of materials.Given these limitations, we have exploited the MODIS time series to propose a new endmembers estimation approach, addressing a critical step in spectral unmixing. The low temporal evolution of the medium (except snow) is then used to constrain the estimation of the endmembers not only on the image of interest, but also on images of the previous days. The effectiveness of this approach, although demonstrated here, remains limited by the spatial resolution of the sensor.Data fusion has been considered aiming at taking advantage of multiple acquisitions with different characteristics in term of resolution available on the same scene. Given the limitations of the actual methods in the case of multispectral sensors, a new fusion approach has been proposed. Through the formulation of a new model and its resolution, the fusion between optical sensors of all types can be achieved without consideration of their characteristics. The various experiments on the estimation of snow maps show a clear interest of a better spatial resolution to isolate the snow covered areas. The improvement in spectral resolution will improve future approaches based on spectral unmixing.This work explores the new possibilities of development for the observation of snow, but also for the combined use of the satellite images for the observation of the Earth in general.

The aim of this study is to extract information about mineral distribution and percentages of Oymaaga&ccedil / granitoid (Beypazari-Ankara) by using remote sensing techniques. Two methods are applied during the studies which are spectral analysis and Crosta techniques. Spectral measurements are done for fresh and weathered samples collected at 32 locations. Mineral percentages are calculated using spectral mixture analysis for each sample by considering main, accessory and secondary mineral content of granodiorite. A total of 10 endmembers for fresh samples and 15 for weathered samples are used. USGS spectral library data is utilized through the analyses. For Crosta technique (image analysis) the multispectral ASTER satellite image is used. Five alteration minerals are discriminated and their maps are generated during this analysis. Interpretation and comparison of the results of both methods and testing these results with the existing petrographical and geochemical data indicate that: 1) according to the results of both spectral analyses and Crosta technique a zonation is not observed in the granitoid, 2) comparison of the results for alteration minerals of these two analyses are partly compatible but not exactly similar, 3) Results of spectral analysis do not fit geochemical nor modal analyses because of inconsistency of the data sets.

Le démélange spectral est un des sujets majeurs de l’analyse d’images hyperspectrales. Ce problème consiste à identifier les composants macroscopiques présents dans une image hyperspectrale et à quantifier les proportions (ou abondances) de ces matériaux dans tous les pixels de l’image. La plupart des algorithmes de démélange suppose un modèle de mélange linéaire qui est souvent considéré comme une approximation au premier ordre du mélange réel. Cependant, le modèle linéaire peut ne pas être adapté pour certaines images associées par exemple à des scènes engendrant des trajets multiples (forêts, zones urbaines) et des modèles non-linéaires plus complexes doivent alors être utilisés pour analyser de telles images. Le but de cette thèse est d’étudier de nouveaux modèles de mélange non-linéaires et de proposer des algorithmes associés pour l’analyse d’images hyperspectrales. Dans un premier temps, un modèle paramétrique post-non-linéaire est étudié et des algorithmes d’estimation basés sur ce modèle sont proposés. Les connaissances a priori disponibles sur les signatures spectrales des composants purs, sur les abondances et les paramètres de la non-linéarité sont exploitées à l’aide d’une approche bayesienne. Le second modèle étudié dans cette thèse est basé sur l’approximation de la variété non-linéaire contenant les données observées à l’aide de processus gaussiens. L’algorithme de démélange associé permet d’estimer la relation non-linéaire entre les abondances des matériaux et les pixels observés sans introduire explicitement les signatures spectrales des composants dans le modèle de mélange. Ces signatures spectrales sont estimées dans un second temps par prédiction à base de processus gaussiens. La prise en compte d’effets non-linéaires dans les images hyperspectrales nécessite souvent des stratégies de démélange plus complexes que celles basées sur un modèle linéaire. Comme le modèle linéaire est souvent suffisant pour approcher la plupart des mélanges réels, il est intéressant de pouvoir détecter les pixels ou les régions de l’image où ce modèle linéaire est approprié. On pourra alors, après cette détection, appliquer les algorithmes de démélange non-linéaires aux pixels nécessitant réellement l’utilisation de modèles de mélange non-linéaires. La dernière partie de ce manuscrit se concentre sur l’étude de détecteurs de non-linéarités basés sur des modèles linéaires et non-linéaires pour l’analyse d’images hyperspectrales. Les méthodes de démélange non-linéaires proposées permettent d’améliorer la caractérisation des images hyperspectrales par rapport au méthodes basées sur un modèle linéaire. Cette amélioration se traduit en particulier par une meilleure erreur de reconstruction des données. De plus, ces méthodes permettent de meilleures estimations des signatures spectrales et des abondances quand les pixels résultent de mélanges non-linéaires. Les résultats de simulations effectuées sur des données synthétiques et réelles montrent l’intérêt d’utiliser des méthodes de détection de non-linéarités pour l’analyse d’images hyperspectrales. En particulier, ces détecteurs peuvent permettre d’identifier des composants très peu représentés et de localiser des régions où les effets non-linéaires sont non-négligeables (ombres, reliefs,...). Enfin, la considération de corrélations spatiales dans les images hyperspectrales peut améliorer les performances des algorithmes de démélange non-linéaires et des détecteurs de non-linéarités. / Spectral unmixing is one the major issues arising when analyzing hyperspectral images. It consists of identifying the macroscopic materials present in a hyperspectral image and quantifying the proportions of these materials in the image pixels. Most unmixing techniques rely on a linear mixing model which is often considered as a first approximation of the actual mixtures. However, the linear model can be inaccurate for some specific images (for instance images of scenes involving multiple reflections) and more complex nonlinear models must then be considered to analyze such images. The aim of this thesis is to study new nonlinear mixing models and to propose associated algorithms to analyze hyperspectral images. First, a ost-nonlinear model is investigated and efficient unmixing algorithms based on this model are proposed. The prior knowledge about the components present in the observed image, their proportions and the nonlinearity parameters is considered using Bayesian inference. The second model considered in this work is based on the approximation of the nonlinear manifold which contains the observed pixels using Gaussian processes. The proposed algorithm estimates the relation between the observations and the unknown material proportions without explicit dependency on the material spectral signatures, which are estimated subsequentially. Considering nonlinear effects in hyperspectral images usually requires more complex unmixing strategies than those assuming linear mixtures. Since the linear mixing model is often sufficient to approximate accurately most actual mixtures, it is interesting to detect pixels or regions where the linear model is accurate. This nonlinearity detection can be applied as a pre-processing step and nonlinear unmixing strategies can then be applied only to pixels requiring the use of nonlinear models. The last part of this thesis focuses on new nonlinearity detectors based on linear and nonlinear models to identify pixels or regions where nonlinear effects occur in hyperspectral images. The proposed nonlinear unmixing algorithms improve the characterization of hyperspectral images compared to methods based on a linear model. These methods allow the reconstruction errors to be reduced. Moreover, these methods provide better spectral signature and abundance estimates when the observed pixels result from nonlinear mixtures. The simulation results conducted on synthetic and real images illustrate the advantage of using nonlinearity detectors for hyperspectral image analysis. In particular, the proposed detectors can identify components which are present in few pixels (and hardly distinguishable) and locate areas where significant nonlinear effects occur (shadow, relief, ...). Moreover, it is shown that considering spatial correlation in hyperspectral images can improve the performance of nonlinear unmixing and nonlinearity detection algorithms.

abstract: Hyperspectral unmixing is an important remote sensing task with applications including material identification and analysis. Characteristic spectral features make many pure materials identifiable from their visible-to-infrared spectra, but quantifying their presence within a mixture is a challenging task due to nonlinearities and factors of variation. In this thesis, physics-based approaches are incorporated into an end-to-end spectral unmixing algorithm via differentiable programming. First, sparse regularization and constraints are implemented by adding differentiable penalty terms to a cost function to avoid unrealistic predictions. Secondly, a physics-based dispersion model is introduced to simulate realistic spectral variation, and an efficient method to fit the parameters is presented. Then, this dispersion model is utilized as a generative model within an analysis-by-synthesis spectral unmixing algorithm. Further, a technique for inverse rendering using a convolutional neural network to predict parameters of the generative model is introduced to enhance performance and speed when training data are available. Results achieve state-of-the-art on both infrared and visible-to-near-infrared (VNIR) datasets as compared to baselines, and show promise for the synergy between physics-based models and deep learning in hyperspectral unmixing in the future. / Dissertation/Thesis / Masters Thesis Electrical Engineering 2020

In recent years, the substantial increase in the number of spectral channels in optical remote sensing sensors allows more detailed spectroscopic analysis of objects on the Earth surface. Modern hyperspectral sensors are able to sample the sunlight reflected from a target on the ground with hundreds of adjacent narrow spectral channels. However, the increased spectral resolution comes at the price of a lower spatial resolution, e.g. the forthcoming German hyperspectral sensor Environmental Mapping and Analysis Program (EnMAP) which will have 244 spectral channels and a pixel size on ground as large as 30 m x 30 m. The main aim of this thesis is dealing with the problem of reduced spatial resolution in hyperspectral sensors. This is addressed first as an unmixing problem, i.e., extraction and quantification of the spectra of pure materials mixed in a single pixel, and second as a resolution enhancement problem based on fusion of multispectral and hyperspectral imagery. This thesis proposes novel methods for hyperspectral unmixing using sparse approximation techniques and external spectral dictionaries, which unlike traditional least squares-based methods, do not require pure material spectrum selection step and are thus able to simultaneously estimate the underlying active materials along with their respective abundances. However, in previous works it has been shown that these methods suffer from some drawbacks, mainly from the intra dictionary coherence. To improve the performance of sparse spectral unmixing, the use of derivative transformation and a novel two step group unmixing algorithm are proposed. Additionally, the spatial homogeneity of abundance vectors by introducing a multi-look model for spectral unmixing is exploited. Based on the above findings, a new method for fusion of hyperspectral images with higher spatial resolution multispectral images is proposed. The algorithm exploits the spectral information of the hyperspectral image and the spatial information from the multispectral image by means of sparse spectral unmixing to form a new high spatial and spectral resolution hyperspectral image. The introduced method is robust when applied to highly mixed scenarios as it relies on external spectral dictionaries. Both the proposed sparse spectral unmixing algorithms as well as the resolution enhancement approach are evaluated quantitatively and qualitatively. Algorithms developed in this thesis are significantly faster and yield better or similar results to state-of-the-art methods.