In this thesis, 2D Local Discriminant Bases (LDB) algorithm is used to 2D search structure to classify remotely sensed data. 2D Linear Discriminant Analysis (LDA) method is converted into an M-ary classifier by combining majority voting principle and linear distance parameters. The feature extraction algorithm extracts the relevant features by removing the irrelevant ones and/or combining the ones which do not represent supplemental information on their own. The algorithm is implemented on a remotely sensed airborne data set from Tippecanoe County, Indiana to evaluate its performance. The spectral and spatial-frequency features are extracted from the multispectral data and used for classifying vegetative species like corn, soybeans, red clover, wheat and oat in the data set.

Tissue optical properties, absorption, scattering and fluorescence, reveal important information about health, and holds the potential for non-invasive diagnosis and therefore earlier treatment for many diseases. On the other hand, tissue structure determines its function. Studying tissue structural properties helps us better understand structure-function relationship. Optical imaging is an ideal tool to study these tissue properties. However, conventional optical imaging techniques have limitations, such as not being able to quantitatively evaluate tissue absorption and scattering properties and only providing volumetrically averaged quantities with no depth control capability. To better study tissue properties, we integrated spatial frequency domain imaging (SFDI) with conventional reflectance imaging modalities. SFDI is a non-invasive, non-contact wide-field imaging technique which utilizes structured illumination to probe tissues. SFDI imaging is able to accurately quantify tissue optical properties. By adjusting spatial frequency, the imaging depth can be tuned which allows for depth controlled imaging. Especially at high spatial frequency, SFDI reflectance image is more sensitive to tissue scattering property than absorption property. The imaging capability of SFDI allows for studying tissue properties from a whole new perspective. In our study, we developed both benchtop and handheld SFDI imaging systems to accommodate different applications. By evaluating tissue optical properties, we corrected attenuation in fluorescence imaging using an analytical model; and we quantified optical and physical properties of skin diseases. By imaging at high spatial frequency, we demonstrated that absorption in fluorescence imaging can also be reduced because of a reduced imaging depth. This correction can be performed in real-time at 19 frames/second. Furthermore, fibrous structures orientation from the superficial layer can be accurately quantified in a multi-layered sample by limiting imaging depth. Finally, we color rendered SFDI reflectance image at high spatial frequency to reveal structural changes in skin lesions.

Cereal grains are an important part of human diet; hence, there is a need to maintain high quality and these grains must be free of physical and biological contaminants. A procedure was developed to differentiate physical contaminants from wheat using NIR (1000-1600 nm) hyperspectral imaging. Three experiments were conducted to select the best combinations of spectral pre-processing technique and statistical classifier to classify physical contaminants: seven foreign material types (barley, canola, maize, flaxseed, oats, rye, and soybean); six dockage types (broken wheat kernels, buckwheat, chaff, wheat spikelets, stones, and wild oats); and two animal excreta types (deer and rabbit droppings) from Canada Western Red Spring (CWRS) wheat. These spectra were processed using five spectral pre-processing techniques (first derivative, second derivative, Savitzky-Golay (SG) smoothing and differentiation, multiplicative scatter correction (MSC), and standard normal variate (SNV)). The raw and pre-processed data were classified using Support Vector Machines (SVM), Naïve Bayes (NB), and k-nearest neighbors (k-NN) classifiers. In each experiment, two-way and multi-way classifications were conducted. Among all the contaminant types, stones, chaff, deer droppings and rabbit droppings were classified with 100% accuracy using the raw reflectance spectra and different statistical classifiers. The SNV technique with k-NN classifier gave the highest accuracy for the classification of foreign material types from wheat (98.3±0.2%) and dockage types from wheat (98.9±0.2%). The MSC and SNV techniques with SVM or k-NN classifier gave perfect classification (100.0±0.0%) for the classification of animal excreta types from wheat. Hence, the SNV technique with k-NN classifier was selected as the best model. Two separate model performance evaluation experiments were conducted to identify and quantify (by number) the amount of contaminant type present in wheat. The overall identification accuracy of the first degree of contamination (one contaminant type with wheat) and the highest degree of contamination (all the contaminant type with wheat) was 97.6±1.6% and 92.5±6.5%, for foreign material types; 98.0±1.8% and 94.3±6.2%r for dockage types; and 100.0±0.0% and 100.0±0.0%, respectively for animal excreta types. The canola, stones, deer, and rabbit droppings were perfectly quantified (100.0±0.0%) at all the levels of contaminations. / February 2016

The scientific applications and engineering aspects of multispectral and hyperspectral imaging systems have been studied extensively. The traditional geometric spectral imaging system model is specifically developed aiming at spectral sensors with spectrally non-overlapping bands. Spectral imaging systems with overlapping bands also exist. For example, the quantum-dot infrared photodetectors (QDIPs) for midwave- and longwave-infrared (IR) imaging systems exhibit highly overlapping spectral responses tunable through the bias voltages applied. This makes it possible to build spectrally tunable imaging system in IR range based on single QDIP. Furthermore, the QDIP based system can be operated as being adaptive to scenes. Other optical imaging systems like the human eye and some polarimetric sensing systems also have overlapping bands. To analyze such sensors, a functional analysis-based framework is provided in this dissertation. The framework starts from the mathematical description of the interaction between sensor and the radiation from scene reaching it. A geometric model of the spectral imaging process is provided based on the framework. The spectral response functions and the scene spectra are considered as vectors inside an 1-dimensional spectral space. The spectral imaging process is abstracted to represent a projection of scene spectrum onto sensor. The projected spectrum, which is the least-square error reconstruction of the scene vectors, contains the useful information for image processing. Spectral sensors with arbitrary spectral response functions are can be analyzed with this model. The framework leads directly to an image pre-processing algorithm to remove the data correlation between bands. Further discussion shows that this model can also serve the purpose of sensor evaluation, and thus facilitates comparison between different sensors. The spectral shapes and the Signal-to-Noise Ratios (SNR) of different bands are seen to influence the sensor's imaging ability in different manners, which are discussed in detail. With the newly defined SNR in spectral space, we can quantitatively characterize the photodetector noise of a spectral sensor with overlapping bands. The idea of adaptive imaging with QDIP based sensor is proposed and illustrated.

This thesis consists of three sections that provide detailed knowledge of nutrient estimation and management in wild blueberry production. The first section investigated the main and interactive effects of long term fertilizer (NPK) enrichments on soil mineral nitrogen, organic nitrogen and carbon, microbial biomass nitrogen and carbon, net mineralization and net nitrification in wild blueberry soils. The second section studied the optimization of wild blueberry growth, development, foliar nutrients and harvestable yields by using response surface methodology. The third section examined nutrient estimation technologies using field spectroscopy. The remote sensing data was analysed with a combination partial least squares regression and variable selection algorithms (Chemometric analysis). The results indicated elevated nitrification activity under nitrogen enrichments, mainly performed by heterotrophs, report unusually high levels of dissolved organic carbon (> 150 C ha-1), a fungal dominated soil system and high concentration of soluble organic nitrogen in the crop year of production. Nitrification and high dissolved organic carbon levels were observed in connection with possible nitrogen saturation and potential environmental hazards. The results imply a need for nitrification inhibition measures. Results from field studies examining the main and interactive effects of soil applied N, P and K suggested that applications of nitrogen (35 kg ha-1), phosphorus (40 kg ha-1) and potassium (30 kg ha-1) were required to optimize growth, development and harvestable yields of wild blueberry. Under these fertilizer rates, the corresponding predicted harvestable yield was 4,126 kg ha-1 that is as much as 13% higher than would be produced by commonly used fertilizer rate in the industry. This study presented new leaf nutrient ranges for sprout and crop years for wild blueberry fields in Atlantic Canada. Hyperspectral remote sensing technologies were used for estimating macro and micro nutrients. This study provides critical information on wavelengths important for nutrient estimation in reflectance spectra (400-2500 nm). The results and inferences from this thesis may be employed to improve crop production, increase economic returns and health of soil and sustainability of wild blueberry production in Nova Scotia. / This study was undertaken to examine the response of the wild blueberry plant to soil applied fertilizers and encompasses soil nitrogen and carbon pools, plant growth and development, leaf nutrient concentrations and harvestable yields. In addition, given the vast area in which wild blueberry fields are located, the study also examined the feasibility of assessing plant nutrient status through the use of remote sensing hyperspectral technologies. Our results emphasize the importance of monitoring for soil nitrogen and carbon pools in the context of accelerated nitrogen cycling, nitrogen saturation, the fine-tuning of current leaf nutrient ranges in Atlantic Canada in connection to fertilizer rates, the possibility of estimating leaf nutrient contents by remote sensing technologies all with the aim of optimizing wild blueberry yields. In terms of statistical techniques, this thesis used response surface methodologies with a central composite design as a means of discovering, the main and interactive effects of soil applied fertilizers to determine the most appropriate soil nitrogen levels and leaf nutrient ranges that correlate to the highest harvestable yields. The remote sensing data used to estimate leaf nutrients concentrations, various models that combined chemometrics and response surface methodologies for determining model efficiencies with aim of getting informative wavelengths in wild blueberry fields.

Nanoparticles, i.e. particles with at least one dimension smaller than 100 nm, are present in large quantities in ambient air and can also be found in an increasing amount of consumer products. It is known that many nanomaterials have physicochemical properties that differ from physicochemical properties of the same material in bulk size. It is therefore important to characterize nanoparticles and to evaluate their toxicity. To understand mechanisms behind nanotoxicity, it is important to study the uptake of nanoparticles, and how they are accumulated. For these purposes model studies of cellular uptake are useful. In this thesis metal oxide and carbon-based nanoparticles have been studied in living cells using Raman spectroscopy. Raman spectroscopy is a method that facilitates a non-destructive analysis without using any fluorescent labels, or any other specific sample preparation. It is possible to collect Raman images, i.e. images where each pixel corresponds to a Raman spectrum, and to use the spectral information to detect nanoparticles, and to identify organelles in cells. In this thesis the question whether or not nanoparticles can enter the cell nucleus of lung epithelial cells has been addressed using hyperspectral analysis. It is shown that titanium dioxide nanoparticles and iron oxide nanoparticles are taken up by cells, and also in the cell nucleus. In contrast, graphene oxide nanoparticles are mainly found attached on the outside of the cell membrane and very few nanoparticles are found in the cell, and none have been detected in the nucleus. It is concluded that graphene oxide nanoparticles are not cytotoxic. However, a comparison of Raman spectra of biomolecules in cells exposed to graphene oxide, unexposed cells and apoptotic cells, shows that the graphene oxide nanoparticles do affect lipid and protein structures. In this thesis, several multivariate data analysis methods have been used to analyze Raman spectra and Raman images. In addition, super-resolution algorithms, which originally have been developed to improve the resolution in photographic images, were optimized and applied to Raman images of cells exposed to submicron polystyrene particles in living cells.

Une image hyperspectrale est formée par un ensemble de bandes spectrales contenant les informations correspondantes à un intervalle du spectre électromagnétique. Le grand avantage de l'imagerie hyperspectrale par rapport l'imagerie traditionnelle est la capacité de mesurer le rayonnement électromagnétique dans le visible et dans d'autres longueurs d'onde. Cette caractéristique permet la détection des différences subtiles existantes parmi les plusieurs objets composant une image. Le traitement de ces images aussi volumineuses nécessite le développement d'algorithmes avancés qui permettent une exploitation optimale des données hyperspectrales. La représentation traditionnelle de ces images est un ensemble de mesures spectrales, ou spectres, une pour chaque pixel de l'image. Le principal inconvénient de cette représentation est que le pixel est l'unité la plus fondamentale des images numériques. Une analyse individuelle des spectres formant une image hyperspectrale fournit une information qui n'est pas optimale. Dans ce cadre, il est nécessaire d'établir des connexions entre les pixels d'une image hyperspectral afin de distinguer des formes dans l'image qui caractérisent leur contenu. Les représentations basées sur des régions fournissent un moyen de réaliser un premier niveau d'abstraction permettant une réduction du nombre d'éléments à traiter et une obtention des informations sémantiques du contenu de l'image. Ce type de représentations fournit une nette amélioration par rapport la représentation classique basée sur des pixels individuels. Sous le titre "La représentation et le traitement des images hyperspectrales en utilisant l'arbre binaire de partitions", cette thèse propose la construction d'une nouvelle représentation hiérarchique d'images hyperspectrales basée sur des régions : l'arbre binaire des partitions (ou BPT, sigles en anglais). Cette nouvelle représentation peut être interprétée comme un ensemble de régions de l'image dans une structure arborescente. L'arbre binaire de partitions peut être utilisé pour représenter : (i) la décomposition d'une image en plusieurs régions ayant un contenu sémantique et (ii) les différentes relations d'inclusion des régions dans la scène. L'arbre binaire de partitions est basée sur la construction d'un algorithme itératif de fusion de régions. La construction du BPT a été étudiée dans cette thèse par l'étude de différents modèles de représentation d'une région hyperspectrale et de différentes distances de similitude entre deux régions hyperspectrales. Cette recherche a été nécessaire en face la grande dimensionalité et complexité des données qui font nécessaire la définition d'un modèle de région et d'une distance de similarité spécifiques. Grâce à la structure en forme d'arbre, le BPT permet la définition d'un grand nombre de techniques pour un traitement avancé des images hyperspectrales. Ces techniques sont typiquement basées sur l'élagage de l'arbre grâce auquel les régions les plus intéressantes pour une application donnée sont extraites. Cette thèse se concentre sur trois applications particulières : la segmentation, la classification et la détection d'objets dans les images hyperspectrales. Les résultats expérimentaux obtenus sur différentes jeux de données montrent les qualités de la représentation BPT.

Assessment of leaf and canopy chlorophyll content provides information on plant physiological status; it is related to nitrogen content and hence, photosynthesis process, net primary productivity and carbon budget. In this study, a method is developed for the retrieval of total chlorophyll content (Chlorophyll a+b) per unit leaf and per unit ground area based on improved vegetation structural parameters which are derived using multispectral multi-angle remote sensing data. Structural characteristics such as clumping and gaps within a canopy affect its solar radiation absorption and distribution and impact its reflected radiance acquired by a sensor. One of the main challenges for the remote sensing community is to accurately estimate vegetation structural parameters, which inevitably influence the retrieval of leaf chlorophyll content. Multi-angle optical measurements provide a means to characterize the anisotropy of surface reflectance, which has been shown to contain information on vegetation structural characteristics. Hyperspectral optical measurements, on the other hand, provide a fine spectral resolution at the red-edge, a narrow spectral range between the red and near infra-red spectra, which is particularly useful for retrieving chlorophyll content. This study explores a new refined measurement concept of combining multi-angle and hyperspectral remote sensing that employs hyperspectral signals only in the vertical (nadir) direction and multispectral measurements in two additional (off-nadir) directions within two spectral bands, red and near infra-red (NIR). The refinement has been proposed in order to reduce the redundancy of hyperspectral data at more than one angle and to better retrieve the three-dimensional vegetation structural information by choosing the two most useful angles of measurements. To illustrate that hyperspectral data acquired at multiple angles exhibit redundancy, a radiative transfer model was used to generate off-nadir hyperspectral reflectances. It has been successfully demonstrated that the off-nadir hyperspectral simulations could be closely reconstructed based on the nadir hyperspectral reflectance and off-nadir multi-spectral reflectance in the red and NIR bands. This is shown using the Compact High-resolution Imaging Spectrometer (CHRIS) and Compact Airborne Spectrographic Imager (CASI) data acquired over a forested area in the Sudbury region (Ontario, Canada). Through intensive validation using field data, it is demonstrated that the combination of reflectances at two angles, the hotspot and darkspot, through the Normalized Difference between Hotspot and Darkspot (NDHD) index has the strongest response to changes in vegetation clumping, an important structural component of canopy. Clumping index (Ω) and Leaf Area Index (LAI) maps are generated based on previous algorithms as well as empirical relationships developed in this study. To retrieve chlorophyll content, inversion of the 5-Scale model is performed by developing Look-Up Tables (LUTs) that are based on the improved structural characteristics developed using multi-angle data. The generated clumping index and LAI maps are used in the LUTs to estimate leaf reflectance. Inversion of the leaf reflectance model, PROSPECT, is further employed to estimate chlorophyll content per unit leaf area. The estimated leaf chlorophyll contents are in good agreement with field-measured values. The refined measurement concept of combining hyperspectral with multispectral multi-angle data provides the opportunity for simultaneous retrieval of vegetation structural and biochemical parameters.

For many digital image/video processing applications increasing the spatial resolution is highly beneficial. At higher resolution, TV pictures look more natural and pleasing to the eye, computer vision tasks such as object detection and tracking can be performed with higher precision, medical diagnoses can be made with a higher confidence, security cameras can offer better identification, and satellite imagery can be interpreted with higher accuracy. As such, spatial resolution is an influential parameter in many mainstream imaging applications, and resolution enhancement task naturally arises as a means of increasing the effectiveness of any imaging system used in the mentioned applications. In this thesis, we concentrate on two enhancement problems of practical importance, namely, low-complexity resolution enhancement for customer grade flat panel televisions, and resolution enhancement of noisy high-dimensional hyperspectral imagery. For TV resolution enhancement our main concern is keeping computational complexity at a minimum. The hardware limitations of average customer grade televisions effectively rule out a multi-frame approach. Hence, we take a low-complexity single-frame approach based on exploiting natural image characteristics. For hyperspectral imagery we take advantage of multiple observations in a modified superresolution framework. Here the main challenges are the high dimensionality of hyperspectral data and the noise present in all spectral bands. We design a physical model of the hyperspectral image acquisition process, and based on this model we formulate an iterative resolution enhancement algorithm.

The desire to use the tools and concepts of chemometrics when studying problems in the life sciences, especially biology and medicine, has prompted chemometricians to shift their focus away from their field‘s traditional emphasis on model predictivity and towards the more contemporary objective of optimizing information exchange via model interpretation. The complex data structures that are captured by modern advanced analytical instruments open up new possibilities for extracting information from complex data sets. This in turn imposes higher demands on the quality of data and the modeling techniques used. The introduction of the concept of orthogonal variation in the late 1990‘s led to a shift of focus within chemometrics; the information gained from analysis of orthogonal structures complements that obtained from the predictive structures that were the discipline‘s previous focus. OPLS, which was introduced in the beginning of 2000‘s, refined this view by formalizing the model structure and the separation of orthogonal variations. Orthogonal variation stems from experimental/analytical issues such as time trends, process drift, storage, sample handling, and instrumental differences, or from inherent properties of the sample such as age, gender, genetics, and environmental influence. The usefulness and versatility of OPLS has been demonstrated in over 500 citations, mainly in the fields of metabolomics and transcriptomics but also in NIR, UV and FTIR spectroscopy. In all cases, the predictive precision of OPLS is identical to that of PLS, but OPLS is superior when it comes to the interpretation of both predictive and orthogonal variation. Thus, OPLS models the same data structures but provides increased scope for interpretation, making it more suitable for contemporary applications in the life sciences. This thesis discusses four different research projects, including analyses of NIR, FTIR and NMR spectroscopic data. The discussion includes comparisons of OPLS and PLS models of complex datasets in which experimental variation conceals and confounds relevant information. The PLS and OPLS methods are discussed in detail. In addition, the thesis describes new OPLS-based methods developed to accommodate hyperspectral images for supervised modeling. Proper handling of orthogonal structures revealed the weaknesses in the analytical chains examined. In all of the studies described, the orthogonal structures were used to validate the quality of the generated models as well as gaining new knowledge. These aspects are crucial in order to enhance the information exchange from both past and future studies.