This work consists of three original articles and is focused on the overall question: How can hyperspectral imaging contribute to patient safety and improve outcomes during abdominal surgery? The hypothesis was that HSI is suitable for the intraoperative assessment of tissue structures and decision support in routine clinical use. Spectral imaging was performed with the TIVITA Tissue for open surgery or TIVITA Mini system for laparoscopic HSI from Diaspective Vision GmbH (Am Salzhaff-Pepelow, Germany). Both HSI systems use pushbroom mode and provide 100 spectral channels in the visible and near-infrared spectral range from 500 - 1000 nm. The Number of Effective Pixels is at least 640 × 480 (x-, y-axis), while the field of view and spatial resolution depend on the measurement distance and the used focal length of the objective. Illumination is done by halogen spots for open surgery and broadband LED in the laparoscopic system. The first part of this work aimed to evaluate HSI for the measurement of ischemic conditioning effects of the gastric conduit during esophagectomy. Ischemic preconditioning by dividing major blood vessels of the stomach prior to gastric pull-up is performed to improve the perfusion at the later esophagogastric anastomosis to reduce the risk of leaks. Intraoperative hyperspectral records of the gastric tube were acquired from 22 patients through the mini-thoracotomy. Fourteen of them underwent ischemic conditioning of the stomach several days before the two-step transthoracic esophagectomy and gastric pull-up with intrathoracic anastomosis was performed. The tip of the gastric tube (later esophago-gastric anastomosis) was measured with HSI. These in vivo records showed that the tissue oxygenation of the gastric conduit was significantly higher in patients who underwent ischemic conditioning (78% vs. 66%; p = 0.03). In the second part of this work, a novel hyperspectral imaging system for MIS is described and evaluated to address the requirements for clinical use and high-resolution spectral imaging. Reference objects and resected human tissue were used to show spectral conformity with the approved HSI device for open surgery. Furthermore, varying object distances were investigated and the signal-to-noise ratio (SNR) for different light sources were measured. Measurements with both systems were performed on a human tissue resectate and compared quantitatively. It was shown that the handheld design of the laparoscopic HSI system enables the processing and visualization of spectral data in parallel during acquisition within a few seconds. The obtained measurements from both spectral imaging devices were consistent and a mean SNR of 30 to 43 dB (500 to 950 nm) was found using a standard rigid laparoscope in combination with a broadband LED light source. Finally, in the third part of this work, different image registration methods were investigated to compensate for small movements of the laparoscope and tissue deformations. The obtained image transformation is used to augment the laparoscopic color video with the static HSI data to support intraoperative localization. Multiple feature-based algorithms and a pre-trained deep homography neural network (DH-NN) were evaluated for the estimation of appropriate image transformations (single and multi-homography). The methods were validated with a ground truth dataset of 750 annotated laparoscopic images, that was created during this work, and in vivo data from the TIVITA Mini system. All feature-based single homography methods outperformed the fine-tuned DH-NN in terms of reprojection error, Structural Similarity Index Measure (SSIM), and processing time. The feature detector and descriptor ORB1000 enabled video-rate registration of laparoscopic images on standard hardware with submillimeter accuracy. Therefore, all initially stated research questions could be confirmed with the applied methods. Although technical limitations have been identified, the non-invasive and contact-free measurement principle makes HSI attractive for a variety of surgical disciplines.:1 Introduction 1.1 Interaction of light and biological tissue 1.2 Spectral imaging systems 1.3 Medical applications of spectral imaging 1.4 Intraoperative visualization of spectral data 2 Original Articles 2.1 Evaluation of hyperspectral imaging (HSI) for the measurement of ischemic conditioning effects of the gastric conduit during esophagectomy 2.2 Laparoscopic system for simultaneous high-resolution video and rapid hyperspectral imaging in the visible and near-infrared spectral range 2.3 Comparison of image registration methods for combining laparoscopic video and spectral image data 3 Summary 3.1 Conclusions and Outlook 4 References

Melanomas account for more than 1.7% of global cancer diagnoses and about 1% of all skin cancer diagnoses in the United States. This type of cancer occurs in the melanin-producing cells in the epidermis and exhibits distinctive variations in melanin and blood concentration values in the form of skin lesions. The current approach for evaluating skin cancer lesions involves visual inspection with a dermatoscope, typically followed by biopsy and histopathological analysis. However, to decrease the risk of misdiagnosis in this process requires invasive biopsies, contributing to the emotional and financial distress of patients. The implementation of a non-invasive imaging technique to aid the analysis of skin lesions in the early stages can potentially mitigate these consequences. Hyperspectral imaging (HSI) has shown promise as a non-invasive technique to analyze skin lesions. Images taken of human skin using a hyperspectral camera are a result of numerous elements in the skin. Being a turbid, inhomogeneous material, the skin has chromophores and scattering agents, which interact with light and produce characteristic back-scattered energy that can be harnessed and examined with an HSI camera. To achieve this in this study, a mathematical model of the skin is used to extract meaningful information from the hyperspectral data in the form of parameters such as melanin concentration, blood volume fraction and blood oxygen saturation in the skin. The human skin is modelled as a bi-layer planar system, whose surface reflectance is theoretically calculated using the Kubelka-Munk theory and absorption laws by Beer and Lambert. The model is evaluated for its sensitivity to the parameters and then fitted to measured hyperspectral data of four volunteer subjects in different conditions. Mean values of melanin, blood volume fraction and oxygen saturation obtained for each of the subjects are reported and compared with theoretical values from literature. Sensitivity analysis revealed wavelengths and wavelength groups which resulted in maximum change in percentage reflectance calculated from the model were 450 and 660 nm for melanin, 500 - 520 nm and 590 - 625 nm for blood volume fraction and 606, 646 and 750 nm for blood oxygen saturation. / Master of Science / Melanoma, the most serious type of skin cancer, develops in the melanin-producing cells in the epidermis. A characteristic marker of skin lesions is the abrupt variations in melanin and blood concentration in areas of the lesion. The present technique to inspect skin cancer lesions involves dermatoscopy, which is a qualitative visual analysis of the lesion’s features using a few standardized techniques such as the 7-point checklist and the ABCDE rule. Typically, dermatoscopy is followed by a biopsy and then a histopathological analysis of the biopsy. To reduce the possibility of misdiagnosing actual melanomas, a considerable number of dermoscopically unclear lesions are biopsied, increasing emotional, financial, and medical consequences. A non-invasive imaging technique to analyze skin lesions during the dermoscopic stage can help alleviate some of these consequences. Hyperspectral imaging (HSI) is a promising methodology to non-invasively analyze skin lesions. Images taken of human skin using a hyperspectral camera are a result of numerous elements in the skin. Being a turbid, inhomogeneous material, the skin has chromophores and scattering agents, which interact with light and produce characteristic back-scattered energy that can be harnessed and analyzed with an HSI camera. In this study, a mathematical model of the skin is used to extract meaningful information from the hyperspectral data in the form of melanin concentration, blood volume fraction and blood oxygen saturation. The mean and standard deviation of these estimates are reported and compared with theoretical values from the literature. The model is also evaluated for its sensitivity with respect to these parameters to identify the most relevant wavelengths.

Thesis (PhD)--Stellenbosch University, 2013. / ENGLISH ABSTRACT: Maize (Zea mays L.) is used for human and animal consumption in diverse forms, from specialised foods in developed countries, to staple food in developing countries. Unfortunately, maize is prone to infection by different Fusarium species that can produce harmful mycotoxins. Fusarium verticillioides is capable of asymptomatic infection, where infected kernels show no sign of fungal growth, but are contaminated with mycotoxins. If fungal contamination is not detected early on, mycotoxins can enter the food chain. Rapid and accurate methods are required to detect, identify and distinguish between pathogens to enable swift decisions regarding the fate of a batch or consignment of cereal. Near infrared (NIR) hyperspectral imaging and multivariate image analysis (MIA) were evaluated to investigate the fungal development in maize kernels over time. When plotting principal component (PC) 4 against PC5, with percentages sum of squares (%SS) 0.49% and 0.34%, three distinct clusters were apparent in the score plot and this was associated with degree of infection. Prominent peaks at 1900 nm and 2136 nm confirmed that the source of variation was due to changes in starch and protein. Variable importance plots (VIP) confirmed the peaks observed in the PCA loading line plots. Early detection of fungal contamination and activity (20 h after inoculation) was possible before visual symptoms of infection appeared. Using NIR hyperspectral imaging and MIA it was possible to differentiate between species of Fusarium associated with maize. It was additionally applied to examine the fungal growth kinetics on culture media. Partial least squares discriminant analysis (PLS-DA) prediction results showed that it was possible to discriminate between species, with F. verticillioides the least correctly predicted (between 16-47% pixels correctly predicted). For F. subglutinans 78-100% and for F. proliferatum 60-80% pixels were correctly predicted. Three prominent bands at 1166, 1380 and 1918 nm were considered to be responsible for the differences between the growth zones. Variations in the bands at 1166 and 1380 nm were correlated with the depletion of carbohydrates as the fungus grew while the band at 1918 nm was a possible indication of spore and new mycelial formation. By plotting the pixels from the individual growth zones as a function of time, it was possible to visualise the emergence and interaction of the growth zones as separate growth profiles. The microstructure of fungal infected maize kernels was studied over time using high resolution X-ray micro-computed tomography (μCT). The presence of voids and airspaces could be seen in two dimensional (2D) X-ray transmission images and in the three dimensional (3D) tomograms. Clear differences were detected between kernels imaged after 20 and 596 h of inoculation. This difference in voids as the fungus progressed showed the effect of fungal damage on the microstructure of the maize kernels. Imaging techniques are important for rapid, accurate and objective evaluation of products for quality and safety. NIR hyperspectral imaging offers rapid chemical evaluation of samples in 2D images while μCT offers 3D microstructural information. By combining these image techniques more value was added and this led to a comprehensive evaluation of Fusarium infection in maize. / AFRIKAANSE OPSOMMING: Mielies (Zea mays L.) word in verskeie vorms deur mens en dier verbruik, van gespesialiseerde voedsel in ontwikkelde lande, tot stapelvoedsel in ontwikkelende lande. Ongelukkig is mielies onderhewig aan besmetting deur verskeie Fusarium spesies wat skadelike mikotoksiene kan produseer. Fusarium verticilloioides is in staat tot asimptomatiese infeksie waar die besmette pit geen teken van fungusgroei toon nie, maar (reeds) met mikotoksiene besmet is. Indien fungusbesmetting nie vroegtydig opgespoor word nie, kan mikotoksiene die voedselketting betree. Vinnige en akkurate metodes word benodig om patogene op te spoor, te identifiseer en ook om onderskeid tussen patogene te tref om sodoende (effektiewe) besluite aangaande die gebruik van ‘n lot of besending graan te neem. Naby-infrarooi (NIR) hiperspektrale beelding en meerveranderlike beeld ontleding (MIA) is geëvalueer om fungusontwikkeling in mieliepitte oor tyd te ondersoek. Wanneer hoofkomponent (PC) 4 teenoor PC5 gestip word, met persentasies som van kwadrate (%SS) 0.49% en 0/34%, is drie afsonderlike groepein die telling grafiek waargeneem. Dit is geassosieer met die graad van besmetting. Prominente pieke by 1900 nm en 2136 nm het bevestig dat veranderinge in stysel en proteïene die bron van die variasie was. Veranderlike belangrikheidsgrafieke (VIP) het die pieke wat in die PCA beladingslyngrafieke waargeneem is, bevestig. Vroegtydige opsporing (bespeuring) van fungusbesmetting en aktiwiteit (20 h na inokulasie) was moontlik voor visuele besmettingsimptome verskyn het. Onderskeid tussen Fusarium spesies wat met mielies geassosieer word, was moontlik deur gebruik te maak van NIR hiperspektrale beelding en MIA. Dit is bykomend toegepas om fungusgroeikinetika op kwekingsmedia te bestudeer. Parsiële kleinste kwadrate diskriminantanalise (PLS-DA) voorspellingsresultate het getoon dat dit moontlik was om tussen spesies te onderskei, met F. verticillioides die minste korrek voorspel (tussen 19-47% beeldelemente korrek voorspel). Vir F. subglutinans is 78-100% en vir F. proliferatum is 60-80% beeldelemente korrek voorspel. Drie prominente bande by 1166, 1380 en 1918 nm is oorweeg as oorsaak vir die verskille tussen die groeisones. Variasies in die bande by 1166 en 1380 nm is gekorreleer met die vermindering van koolhidrate soos die fungus groei, terwyl die band by 1918 nm ‘n moontlike aanduiding van spoor en nuwe miseliale vorming is. Deur die beeldelemente van die individuele groeisones as ‘n funksie van tyd te stip, was dit moontlik om die verskyning en interaksie van die groeisones as aparte groeiprofiele te visualiseer. Hoë-resolusie X-straal mikro-berekende tomografie (μCT) is gebruik om die mikrostruktuur van fungusbesmette mieliepitte oor tyd te ondersoek. Die voorkoms van leemtes en lugruimtes kon in die twee-dimensionele (2D) X-straal transmissie beelde en in die drie-dimensionele (3D) tomogramme gesien word. Duidelike verskille is waargeneem tussen pitte wat na 20 en 596 h na inokulasie verbeeld is. Hierdie verskil in leemtes soos die fungus vorder, het die effek van fungusskade op die mikrostruktuur van mieliepitte getoon. Beeldingstegnieke is belangrik vir vinnige, akkurate en objektiewe evaluasie van produkte vir kwaliteit en veiligheid. NIR hiperspektrale beelding bied vinnige chemiese evaluering van monsters in 2D beelde, terwyl μCT 3D mikrostrukturele inligting gee. Meer waarde is toegevoeg deur hierdie beeldingstegnieke te kombineer en dit het gelei tot ‘n omvangryke evaluering van Fusarium besmetting in mielies.

A major goal in biological imaging is to visualize interactions of different tissues, often fluorescently labeled, during dynamic processes. Only a few of these labels fit into the available spectral range without overlap, but can be separated computationally if the full spectrum of every single pixel is known. In medical imaging, hyperspectral techniques show promise to identify different tissue types without any staining. Yet, microscopists still commonly acquire spectral information either with filters, thus integrating over a few broad bands only, or point-wise, dispersing the spectra onto a multichannel detector, which is inherently slow. Light sheet fluorescence microscopy (LSFM) and optical projection tomography (OPT) are two techniques to acquire 3D microscopic data fast, photon-efficiently and gently on the specimen. LSFM works in fluorescence mode and OPT in transmission. Both are based on a fast widefield detection scheme where a 2D detector records the spatial information but leaves no room to acquire dispersed spectra. Hyperspectral imaging had not yet been demonstrated for either technique. In this work, I developed a line-scanning hyperspectral LSFM and an excitation scanning OPT to acquire 5D data (3D spatial, 1D temporal, 1D spectral) and optimized the performance of both setups to minimize acquisition times without sacrificing image contrast, spatial or spectral information. I implemented and assessed different evaluation pipelines to classify and unmix relevant features. I demonstrate the efficiency of my workflow by acquiring up to five fluorescent markers and the autofluorescence in \\zf and fruit fly embryos on my hyperspectral LSFM. I extracted both concentration maps and spectra for each of these fluorophores from the multidimensional data. The same methods were applied to investigate the transmission data from my spectral OPT, where I found evidence that OPT image formation is governed by refraction, whereas scattering and absorption only play a minor role. Furthermore, I have implemented a robust, educational LSFM on which laymen have explored the working principles of modern microscopies. This eduSPIM has been on display in the Technische Sammlungen Dresden for one year during the UNESCO international year of light. / Ein wichtiges Ziel biologischer Bildgebung ist die Visualisierung des Zusammenspiels von verschiedenen, meist fluoreszent markierten, Geweben bei dynamischen Prozessen. Nur wenige dieser Farbstoffe passen ohne Überlapp in das zur Verfügung stehende Spektrum. Sie können jedoch rechnerisch getrennt werden, wenn das gesamte Spektrum jedes Pixels bekannt ist. In medizinischen Anwendungen versprechen hyperspektrale Techniken, verschiedene Gewebetypen markierungsfrei zu identifizieren. Dennoch ist es in der Mikroskopie noch immer üblich, spektrale Information entweder mit Filtern über breiten Bändern zu integrieren, oder Punktspektren mithilfe von Dispersion zu trennen und auf einem Multikanaldetektor aufzunehmen, was inhärent langsam ist. Light Sheet Fluorescence Microscopy (LSFM) und Optical Projection Tomography (OPT) nehmen 3D Mikroskopiedaten schnell, photoneneffizient und sanft für die Probe auf. LSFM arbeitet mit Fluoreszenz, OPT in Transmission. Beide basieren auf schneller Weitfelddetektion, wobei die räumliche Information mit einem 2D Detektor aufgenommen wird, der keinen Raum lässt, um die getrennten Spektren zu messen. Hyperspektrale Bildgebung wurde bis jetzt für keine der zwei Techniken gezeigt. Ich habe ein hyperspektrales LSFM mit Linienabtastung und ein OPT mit Wellenlängenabtastung entwickelt, um 5D Daten (3D räumlich, 1D zeitlich, 1D spektral) aufzunehmen. Beide Aufbauten wurden hinsichtlich minimaler Aufnahmezeit optimiert, ohne dabei Kontrast, räumliche oder spektrale Auflösung zu opfern. Ich habe verschiedene Abläufe zum Klassifizieren und Trennen der Hauptkomponenten implementiert. Ich nehme bis zu fünf Fluorophore und Autofluoreszenz in Zebrafisch- und Fruchtfliegenembryos mit dem hyperspektralen LSFM auf und zeige die Effizienz des gesamten Ablaufes, indem ich Spektren und räumliche Verteilung aller Marker extrahiere. Die Transmissionsdaten des spektralen OPT werden mit denselben Methoden untersucht. Ich konnte belegen, dass die Bildformation im OPT massgeblich von Brechung bestimmt ist, und Streuung und Absorption nur einen geringen Beitrag leisten. Außerdem habe ich ein robustes, didaktisches LSFM gebaut, damit Laien die Funktionsweise moderner Mikroskopie erkunden können. Dieses eduSPIM war ein Jahr lang in den Technischen Sammlungen Dresden ausgestellt.

L'imagerie hyperspectrale consiste à acquérir une scène spatiale à plusieurs longueurs d'onde, e.g. en microscopie. Cependant, lorsque l'image est observée à une résolution suffisamment fine, elle est dégradée par un flou (convolution) et une procédure de déconvolution doit être utilisée pour restaurer l'image originale. Ce problème inverse, par opposition au problème direct modélisant la dégradation de l'image observée, est étudié dans la première partie . Un autre problème inverse important en imagerie, la séparation de sources, consiste à extraire les spectres des composants purs de l'image (sources) et à estimer les contributions de chaque source à l'image. La deuxième partie propose des contributions algorithmiques en restauration d'images hyperspectrales. Le problème est formulé comme la minimisation d'un critère pénalisé et résolu à l'aide d'une structure de calcul rapide. La méthode est adaptée à la prise en compte de différents a priori sur l'image, tels que sa positivité ou la préservation des contours. Les performances des techniques proposées sont évaluées sur des images de biocapteurs bactériens en microscopie confocale de fluorescence. La troisième partie est axée sur le problème de séparation de sources, abordé dans un cadre géométrique. Nous proposons une nouvelle condition suffisante d'identifiabilité des sources à partir des coefficients de mélange. Une étude innovante couplant le modèle d'observation avec le mélange de sources permet de montrer l'intérêt de la déconvolution comme étape préliminaire de la séparation. Ce couplage est validé sur des données acquises en spectroscopie Raman / Hyperspectral imaging refers to the acquisition of spatial images at many spectral bands, e.g. in microscopy. Processing such data is often challenging due to the blur caused by the observation system, mathematically expressed as a convolution. The operation of deconvolution is thus necessary to restore the original image. Image restoration falls into the class of inverse problems, as opposed to the direct problem which consists in modeling the image degradation process, treated in part 1 of the thesis. Another inverse problem with many applications in hyperspectral imaging consists in extracting the pure materials making up the image, called endmembers, and their fractional contribution to the data or abundances. This problem is termed spectral unmixing and its resolution accounts for the nonnegativity of the endmembers and abundances. Part 2 presents algorithms designed to efficiently solve the hyperspectral image restoration problem, formulated as the minimization of a composite criterion. The methods are based on a common framework allowing to account for several a priori assumptions on the solution, including a nonnegativity constraint and the preservation of edges in the image. The performance of the proposed algorithms are demonstrated on fluorescence confocal images of bacterial biosensors. Part 3 deals with the spectral unmixing problem from a geometrical viewpoint. A sufficient condition on abundance coefficients for the identifiability of endmembers is proposed. We derive and study a joint observation model and mixing model and demonstrate the interest of performing deconvolution as a prior step to spectral unmixing on confocal Raman microscopy data

<p>Low-cost unmanned aerial vehicles (UAVs) utilizing push-broom hyperspectral scanners are poised to become a popular alternative to conventional remote sensing platforms such as manned aircraft and satellites. In order to employ this emerging technology in fields such as high-throughput phenotyping and precision agriculture, direct georeferencing of hyperspectral data using onboard integrated global navigation satellite systems (GNSS) and inertial navigation systems (INS) is required. Directly deriving the scanner position and orientation requires the spatial and rotational relationship between the coordinate systems of the GNSS/INS unit and hyperspectral scanner to be evaluated. The spatial offset (lever arm) between the scanner and GNSS/INS unit can be measured manually. However, the angular relationship (boresight angles) between the scanner and GNSS/INS coordinate systems, which is more critical for accurate generation of georeferenced products, is difficult to establish. This research presents three alternative calibration approaches to estimate the boresight angles relating hyperspectral push-broom scanner and GNSS/INS coordinate systems. For reliable/practical estimation of the boresight angles, the thesis starts with establishing the optimal/minimal flight and control/tie point configuration through a bias impact analysis starting from the point positioning equation. Then, an approximate calibration procedure utilizing tie points in overlapping scenes is presented after making some assumptions about the flight trajectory and topography of covered terrain. Next, two rigorous approaches are introduced – one using Ground Control Points (GCPs) and one using tie points. The approximate/rigorous approaches are based on enforcing the collinearity and coplanarity of the light rays connecting the perspective centers of the imaging scanner, object point, and the respective image points. To evaluate the accuracy of the proposed approaches, estimated boresight angles are used for ortho-rectification of six hyperspectral UAV datasets acquired over an agricultural field. Qualitative and quantitative evaluations of the results have shown significant improvement in the derived orthophotos to a level equivalent to the Ground Sampling Distance (GSD) of the used scanner (namely, 3-5 cm when flying at 60 m).</p>

L'imagerie hyperspectrale (HSI) repose sur le fait que, pour un matériau donné, la quantité de rayonnement émis varie avec la longueur d'onde. Les capteurs HSI mesurent donc le rayonnement des matériaux au sein de chaque pixel pour un très grand nombre de bandes spectrales contiguës et fournissent des images contenant des informations à la fois spatiale et spectrale. Les méthodes classiques de détection adaptative supposent généralement que le fond est gaussien à vecteur moyenne nul ou connu. Cependant, quand le vecteur moyen est inconnu, comme c'est le cas pour l'image hyperspectrale, il doit être inclus dans le processus de détection. Nous proposons dans ce travail d'étendre les méthodes classiques de détection pour lesquelles la matrice de covariance et le vecteur de moyenne sont tous deux inconnus.Cependant, la distribution statistique multivariée des pixels de l'environnement peut s'éloigner de l'hypothèse gaussienne classiquement utilisée. La classe des distributions elliptiques a été déjà popularisée pour la caractérisation de fond pour l’HSI. Bien que ces modèles non gaussiens aient déjà été exploités dans la modélisation du fond et dans la conception de détecteurs, l'estimation des paramètres (matrice de covariance, vecteur moyenne) est encore généralement effectuée en utilisant des estimateurs conventionnels gaussiens. Dans ce contexte, nous analysons de méthodes d’estimation robuste plus appropriées à ces distributions non-gaussiennes : les M-estimateurs. Ces méthodes de détection couplées à ces nouveaux estimateurs permettent d'une part, d'améliorer les performances de détection dans un environment non-gaussien mais d'autre part de garder les mêmes performances que celles des détecteurs conventionnels dans un environnement gaussien. Elles fournissent ainsi un cadre unifié pour la détection de cibles et la détection d'anomalies pour la HSI. / Hyperspectral imaging (HSI) extends from the fact that for any given material, the amount of emitted radiation varies with wavelength. HSI sensors measure the radiance of the materials within each pixel area at a very large number of contiguous spectral bands and provide image data containing both spatial and spectral information. Classical adaptive detection schemes assume that the background is zero-mean Gaussian or with known mean vector that can be exploited. However, when the mean vector is unknown, as it is the case for hyperspectral imaging, it has to be included in the detection process. We propose in this work an extension of classical detection methods when both covariance matrix and mean vector are unknown.However, the actual multivariate distribution of the background pixels may differ from the generally used Gaussian hypothesis. The class of elliptical distributions has already been popularized for background characterization in HSI. Although these non-Gaussian models have been exploited for background modeling and detection schemes, the parameters estimation (covariance matrix, mean vector) is usually performed using classical Gaussian-based estimators. We analyze here some robust estimation procedures (M-estimators of location and scale) more suitable when non-Gaussian distributions are assumed. Jointly used with M-estimators, these new detectors allow to enhance the target detection performance in non-Gaussian environment while keeping the same performance than the classical detectors in Gaussian environment. Therefore, they provide a unified framework for target detection and anomaly detection in HSI.

Le contenu et l’état de conservation exceptionnel du Codex Borbonicus en font un témoin précieux de la culture du bassin de Mexico-Tenochtitlan à l’époque de l’arrivée des conquistadors. Pour certains historiens il s’agit d’un document assurément précolombien, tandis que pour d’autres, une partie de son contenu graphique dénote une influence culturelle Européenne. Afin d’explorer les savoir-faire mis en œuvre lors de sa production et d’apporter de nouvelles données à ce débat, la nature et le mode de préparation des constituants du manuscrit sont étudiés, dans les limites offertes par les instrumentations transportables et non-invasives (spectroscopies de fluorescence de rayons X, de réflexion, d’émission et Raman). Une première interprétation des données analytiques enregistrées sur le manuscrit se base sur les connaissances issues des sources historiques et du corpus de manuscrits mésoaméricains déjà étudiés. Une analyse plus fine des données est apportée par des calculs de combinaisons spectrales et par l’étude expérimentale de certains colorants, qui permettent une compréhension plus avancée des techniques de production picturale employées. Afin de généraliser les conclusions tirées des mesures localisées, la distribution des constituants sur la totalité du document est également abordée. L’imagerie hyperspectrale, par l’application d’outils statistiques et le développement de cartographies de motifs spectraux spécifiques, apporte ainsi une nouvelle perspective aux résultats des analyses. L’utilisation exclusive de colorants organiques d’origine animale (Dactylopius coccus) ou végétale (Indigofera suffruticosa, Comellina coelestis, Justicia spicigera) dans le Codex Borbonicus, seuls ou en mélanges, correspond aux traditions précolombiennes. L’hypothèse d’une influence européenne ne peut donc s’appuyer sur la nature des constituants du document. Les données présentées viennent par ailleurs enrichir les connaissances sur les techniques de production de manuscrits Mésoaméricains. / The Codex Borbonicus is a great source of knowledge regarding different aspects of the culture of the basin of Mexico-Tenochtitlan at the time of the Spanish conquest. For some historians, the manuscript is definitely Precolumbian while for others, parts of its graphical contents reveal a European cultural influence. In order to investigate the technological knowledge that was involved for its creation, and to bring fresh data to the debate, the manuscript material constituents are studied with transportable, non-invasive analytical techniques (X-Ray fluorescence, reflexion, emission and Raman spectroscopies). A first interpretation of the analytical data recorded on the document is based on the historical records and the corpus of Mesoamerican manuscripts that have already been studied. A finer analysis of the data is done through the calculation of spectral combinations as well as the experimental studies of certain coloring materials, that allow a better understanding of the paint preparation techniques. In order to generalize the conclusion based on localized analyses, the spatial distribution of the constituent on the whole document is also explored. Hyperspectral imaging, with the aid of statistical tools and the mapping of specific spectral features, brings new insights to the first results. The exclusive use of organic colorants extracted from animal (Dactylopius coccus) or vegetable sources (Indigofera suffruticosa, Comellina coelestis, Justicia spicigera) in the Codex Borbonicus, alone or in mixtures, fits what is known of Precolumbian traditions. Therefore, the hypothesis of a European influence can’t be supported by the nature of the manuscript constituents. Moreover, these new data enrich the current knowledge on Mesoamerican manuscript production techniques.

Compressive sensing (CS) is a novel sampling methodology representing a paradigm shift from conventional data acquisition schemes. The theory of compressive sensing ensures that under suitable conditions compressible signals or images can be reconstructed from far fewer samples or measurements than what are required by the Nyquist rate. So far in the literature, most works on CS concentrate on one-dimensional or two-dimensional data. However, besides involving far more data, three-dimensional (3D) data processing does have particularities that require the development of new techniques in order to make successful transitions from theoretical feasibilities to practical capacities. This thesis studies several issues arising from the applications of the CS methodology to some 3D image processing tasks. Two specific applications are hyperspectral imaging and video compression where 3D images are either directly unmixed or recovered as a whole from CS samples. The main issues include CS decoding models, preprocessing techniques and reconstruction algorithms, as well as CS encoding matrices in the case of video compression. Our investigation involves three major parts. (1) Total variation (TV) regularization plays a central role in the decoding models studied in this thesis. To solve such models, we propose an efficient scheme to implement the classic augmented Lagrangian multiplier method and study its convergence properties. The resulting Matlab package TVAL3 is used to solve several models. Computational results show that, thanks to its low per-iteration complexity, the proposed algorithm is capable of handling realistic 3D image processing tasks. (2) Hyperspectral image processing typically demands heavy computational resources due to an enormous amount of data involved. We investigate low-complexity procedures to unmix, sometimes blindly, CS compressed hyperspectral data to directly obtain material signatures and their abundance fractions, bypassing the high-complexity task of reconstructing the image cube itself. (3) To overcome the "cliff effect" suffered by current video coding schemes, we explore a compressive video sampling framework to improve scalability with respect to channel capacities. We propose and study a novel multi-resolution CS encoding matrix, and a decoding model with a TV-DCT regularization function. Extensive numerical results are presented, obtained from experiments that use not only synthetic data, but also real data measured by hardware. The results establish feasibility and robustness, to various extent, of the proposed 3D data processing schemes, models and algorithms. There still remain many challenges to be further resolved in each area, but hopefully the progress made in this thesis will represent a useful first step towards meeting these challenges in the future.

In this thesis, two different approaches are used to extract the relevant features for classifying the aflatoxin contaminated and uncontaminated scaled chili pepper samples: Statistical approach and Local Discriminant Bases (LDB) approach. In the statistical approach, First Order Statistical (FOS) features and Gray Level Cooccurrence Matrix (GLCM) features are extracted. In the LDB approach, the original LDB algorithm is modified to perform 2D searches to extract the most discriminative features from the hyperspectral images by removing irrelevant features and/or combining the features that do not provide sufficient discriminative information on their own. The classification is performed by using Linear Discriminant Analysis (LDA) classifier. Hyperspectral images of scaled chili peppers purchased from various locations in Turkey are used in this study. Correct classification accuracy about 80% is obtained by using the extracted features.