This dissertation develops new approaches for improving the performance of hyperspectral target detection. Different aspects of hyperspectral target detection are reviewed and studied to effectively distinguish target features from background interference. The contributions of this dissertation are detailed as follows. 1) Propose an adaptive background characterization method that integrates region segmentation with target detection. In the experiments, not only unstructured matched filter based detectors are considered, but also two hybrid detectors combining fully constrained least squared abundance estimation with statistic test (i.e., adaptive matched subspace detector and adaptive cosine/coherent detector) are investigated. The experimental results demonstrate that using local adaptive background characterization, background clutters can be better suppressed than the original algorithms with global characterization. 2) Propose a new approach to estimate abundance fractions based on the linear spectral mixture model for hybrid structured and unstructured detectors. The new approach utilizes the sparseness constraint to estimate abundance fractions, and achieves better performance than the popular non-negative and fully constrained methods in the situations when background endmember spectra are not accurately acquired or estimated, which is very common in practical applications. To improve the dictionary incoherence, the use of band selection is proposed to improve the sparseness constrained linear unmixing. 3) Propose random projection based dimensionality reduction and decision fusion approach for detection improvement. Such a data independent dimensionality reduction process has very low computational cost, and it is capable of preserving the original data structure. Target detection can be robustly improved by decision fusion of multiple runs of random projection. A graphics processing unit (GPU) parallel implementation scheme is developed to expedite the overall process. 4) Propose nonlinear dimensionality reduction approaches for target detection. Auto-associative neural network-based Nonlinear Principal Component Analysis (NLPCA) and Kernel Principal Component Analysis (KPCA) are applied to the original data to extract principal components as features for target detection. The results show that NLPCA and KPCA can efficiently suppress trivial spectral variations, and perform better than the traditional linear version of PCA in target detection. Their performance may be even better than the directly kernelized detectors.

<p>  </p> <p>The Leafspec portable imaging device had provided a reliable low-cost hyperspectral image acquisition solution to a wide range of users. The Leafspec implemented a enclosed imaging environment and a build-in halogen light source, which eliminated the influence of ambient light. However, a uniform light source was hard to achieve due to dimension and power restrains. White Reference, in many cases refer to an image of a uniform white material such as a Teflon board. White Referencing is a widely used calibration technique in the effort of minimizing noises in an image, including ones that are induced by the light source. However, an abnormal spatial distribution was found remained in images collected by Leafspec even after performing White Referencing technique. In hypothesis, the huge spectral difference between Teflon board and leaves caused this issue. Following this assumption, this article proposed to use a uniform section of soybean leaf to create a Green Reference for calibrating hyperspectral images. In our experiment, 20 green reference samples were collected by imaging a 15 𝑚𝑚 × 15 𝑚𝑚 most uniform section from soybean leaf along a 100 𝑚𝑚 imaging window with a 5 𝑚𝑚 increment. </p>

Le titre de cette thèse de doctorat est formé de trois mots clés: parcimonie, image hyperspectrale, et détection de cibles. La parcimonie signifie généralement « petit en nombre ou quantité, souvent répartie sur une grande zone ». Une image hyperspectrale est constituée d'une série d'images de la même scène spatiale, mais prises dans plusieurs dizaines de longueurs d'onde contiguës et très étroites, qui correspondent à autant de "couleurs". Lorsque la dimension spectrale est très grande, la détection de cibles devient délicate et caractérise une des applications les plus importantes pour les images hyperspectrales. Le but principal de cette thèse de doctorat est de répondre à la question « Comment et Pourquoi la parcimonie peut-elle être exploitée pour détecter de cibles dans les images hyperspectrales ? ». La réponse à cette question nous a permis de développer des méthodes de détection de cibles prenant en compte l'hétérogénéité de l'environnement, le fait que les objets d'intérêt sont situés dans des parties relativement réduites de l'image observée et enfin que l'estimation de la matrice de covariance d'un pixel d'une image hyperspectrale peut être compliquée car cette matrice appartient à un espace de grande dimension. Les méthodes proposées sont évaluées sur des données synthétiques ainsi que réelles, dont les résultats démontrent leur efficacité pour la détection de cibles dans les images hyperspectrales. / The title of this PhD thesis is formed by three keywords: sparsity, hyperspectral image, and target detection. Sparsity is a word that is used everywhere and in everyday life. It generally means « small in number or amount, often spread over a large area ». A hyperspectral image is a three dimensional data cube consisting of a series of images of the same spatial scene in a contiguous and multiple narrow spectral wavelength (color) bands. According to the high spectral dimensionality, target detection is not surprisingly one of the most important applications in hyperspectral imagery. The main objective of this PhD thesis is to answer the question « How and Why can sparsity be exploited for hyperspectral target detection? ». Answering this question has allowed us to develop different target detection methods that mainly take into consideration the heterogeneity of the environment, the fact that the total image area of all the targets is very small relative to the whole image, and the estimation challenge of the covariance matrix (surrounding the test pixel) in large dimensions. The proposed mehods are evaluated on both synthetic and real experiments, the results of which demonstrate their effectiveness for hyperspectral target detection.

<p>Unprecedented amount of data has been collected in diverse fields such as social network, infectious disease and political science in this information explosive era. The high dimensional, complex and heterogeneous data imposes tremendous challenges on traditional statistical models. Bayesian nonparametric methods address these challenges by providing models that can fit the data with growing complexity. In this thesis, we design novel Bayesian nonparametric models on dataset from three different fields, hyperspectral images analysis, infectious disease and voting behaviors. </p><p>First, we consider analysis of noisy and incomplete hyperspectral imagery, with the objective of removing the noise and inferring the missing data. The noise statistics may be wavelength-dependent, and the fraction of data missing (at random) may be substantial, including potentially entire bands, offering the potential to significantly reduce the quantity of data that need be measured. We achieve this objective by employing Bayesian dictionary learning model, considering two distinct means of imposing sparse dictionary usage and drawing the dictionary elements from a Gaussian process prior, imposing structure on the wavelength dependence of the dictionary elements.</p><p>Second, a Bayesian statistical model is developed for analysis of the time-evolving properties of infectious disease, with a particular focus on viruses. The model employs a latent semi-Markovian state process, and the state-transition statistics are driven by three terms: ($i$) a general time-evolving trend of the overall population, ($ii$) a semi-periodic term that accounts for effects caused by the days of the week, and ($iii$) a regression term that relates the probability of infection to covariates (here, specifically, to the Google Flu Trends data).</p><p>Third, extensive information on 3 million randomly sampled United States citizens is used to construct a statistical model of constituent preferences for each U.S. congressional district. This model is linked to the legislative voting record of the legislator from each district, yielding an integrated model for constituency data, legislative roll-call votes, and the text of the legislation. The model is used to examine the extent to which legislators' voting records are aligned with constituent preferences, and the implications of that alignment (or lack thereof) on subsequent election outcomes. The analysis is based on a Bayesian nonparametric formalism, with fast inference via a stochastic variational Bayesian analysis.</p> / Dissertation

Segundo a Agência Nacional de Transportes Terrestres (ANTT) em seu Anuário Estatístico dos Transportes Terrestres AETT (2008), o Brasil em todo o seu território possui 211.678 quilômetros de rodovias pavimentadas. O valor de serventia do pavimento diminui com o passar do tempo por dois fatores principais: o tráfego e as intempéries (BERNUCCI et al., 2008). Monitorar a condição de uso de toda a extensão das rodovias brasileiras é tarefa dispendiosa e demorada. A investigação de novas técnicas que permitam o levantamento da condição dos pavimentos de forma ágil e automática é parte da pesquisa deste trabalho. Nos últimos anos, um número crescente de imagens de alta resolução espacial tem surgido no mercado mundial com o aparecimento dos novos satélites e sensores aeroembarcados de sensoriamento remoto. Da mesma forma, imagens multiespectrais e até mesmo hiperespectrais estão sendo disponibilizadas comercialmente e para pesquisa científica. Neste trabalho são utilizadas imagens hiperespectrais de sensor digital aeroembarcado. Uma metodologia para identificação automática dos pavimentos asfaltados e classificação das principais ocorrências dos defeitos do asfalto foi desenvolvida. A primeira etapa da metodologia é a identificação do asfalto na imagem, utilizando uma classificação híbrida baseada inicialmente em pixel e depois refinada por objetos foi possível a extração da informação de asfalto das imagens disponíveis. A segunda etapa da metodologia é a identificação e classificação das ocorrências dos principais defeitos nos pavimentos flexíveis que são observáveis nas imagens de alta resolução espacial. Esta etapa faz uso intensivo das novas técnicas de classificação de imagens baseadas em objetos. O resultado final é a geração de índices da condição do pavimento, a partir das imagens, que possam ser comparados com os indicadores da qualidade da superfície do pavimento já normatizados pelos órgãos competentes no país. / According to Statistical Survey of Land Transportation AETT (2008) of National Agency of Land Transportation (ANTT), Brazil has in its territory 211,678 kilometers of paved roads. The pavement Present Serviceability Ratio (PSR) value decreases over time by two main factors: traffic and weather (BERNUCCI et al., 2008). Monitor the condition of use of all Brazilian roads is expensive and time consuming task. The investigation of new techniques that allow a quick and automatic survey of pavement condition is part of this research. In recent years, an increasing number of images with high spatial resolution has emerged on the world market with the advent of new remote sensing satellites and airborne sensors. Similarly, multispectral and even hyperspectral imagery are become available commercially and for scientific research nowadays. Hyperspectral images from digital airborne sensor have been used in this work. A new methodology for automatic identification of asphalted pavement and also for classification of the main defects of the asphalt has been developed. The first step of the methodology is the identification of the asphalt in the image, using hybrid classification based on pixel initially and after improved by objects. Using this approach was feasible to extract asphalt information from the available images. The second step of the methodology is the identification and classification of the main defects of flexible pavement surface that are observable in high spatial resolution imagery. This step makes intensive use of new techniques for classification of images based on objects. The goal, is the generation of pavement surface condition index from the images that can be compared with quality index of pavement surface that are already regulated by the regulatory agency in the country.

Les images hyperspectrales (HSI) fournissent des informations spectrales détaillées sur les objets analysés. Étant donné que différents matériaux ont des signatures spectrales distinctes, les objets ayant des couleurs et des formes similaires peuvent être distingués dans le domaine spectral. Toutefois, l’énorme quantité de données peut poser des problèmes en termes de stockage et de transmission des données. De plus, la haute dimensionnalité des images hyperspectrales peut entraîner un surajustement du classificateur en cas de données d'apprentissage insuffisantes. Une façon de résoudre de tels problèmes consiste à effectuer une sélection de bande (BS), car elle réduit la taille du jeu de données tout en conservant des informations utiles et originales. Dans cette thèse, nous proposons trois méthodes de sélection de bande différentes. La première est supervisée, conçu pour utiliser seulement 20% des données disponibles. Pour chaque classe du jeu de données, une classification binaire un contre tous utilisant un réseau de neurones est effectuée et les bandes liées aux poids le plus grand et le plus petit sont sélectionnées. Au cours de ce processus, les bandes les plus corrélées avec les bandes déjà sélectionnées sont rejetées. Par conséquent, la méthode proposée peut être considérée comme une approche de sélection de bande orientée par des classes. La deuxième méthode que nous proposons est une version non supervisée du premier framework. Au lieu d'utiliser les informations de classe, l'algorithme K-Means est utilisé pour effectuer une classification binaire successive de l'ensemble de données. Pour chaque paire de grappes, un réseau de neurones à une seule couche est utilisé pour rechercher l'hyperplan de séparation, puis la sélection des bandes est effectuée comme décrit précédemment. Pour la troisième méthode de BS proposée, nous tirons parti de la nature non supervisée des auto-encodeurs. Pendant la phase d'apprentissage, le vecteur d'entrée est soumis au bruit de masquage. Certaines positions de ce vecteur sont basculées de manière aléatoire sur zéro et l'erreur de reconstruction est calculée sur la base du vecteur d'entrée non corrompu. Plus l'erreur est importante, plus les fonctionnalités masquées sont importantes. Ainsi, à la fin, il est possible d'avoir un classement des bandes spectrales de l'ensemble de données. / Hyperspectral images (HSIs) are capable of providing a detailed spectral information about scenes or objects under analysis. It is possible thanks to both numerous and contiguous bands contained in such images. Given that di_erent materials have distinct spectral signatures, objects that have similar colors and shape can be distinguished in the spectral domain that goes beyond the visual range. However, in a pattern recognition system, the huge amount of data contained in HSIs may pose problems in terms of data storage and transmission. Also, the high dimensionality of hyperspectral images can cause the overfitting of the classifer in case of insufficient training data. One way to solve such problems is to perform band selection(BS) in HSIs, because it decreases the size of the dataset while keeping both useful and original information. In this thesis, we propose three different band selection frameworks. The first one is a supervised one, and it is designed to use only 20% of the available training data. For each class in the dataset, a binary one-versus-all classification using a single-layer neural network is performed, and the bands linked to the largest and smallest coefficients of the resulting hyperplane are selected. During this process, the most correlated bands with the bands already selected are automatically discarded, following a procedure also proposed in this thesis. Consequently, the proposed method may be seen as a classoriented band selection approach, allowing a BS criterion that meets the needs of each class. The second method we propose is an unsupervised version of the first framework. Instead of using the class information, the K-Means algorithm is used to perform successive binary clustering of the dataset. For each pair of clusters, a single-layer neural network is used to find the separating hyperplane, then the selection of bands is done as previously described. For the third proposed BS framework, we take advantage of the unsupervised nature of autoencoders. During the training phase, the input vector is subjected to masking Noise - some positions of this vector are randomly flipped to zero and the reconstruction error is calculated based on the uncorrupted input vector. The bigger the error, the more important the masked features are. Thus, at the end, it is possible to have a ranking of the spectral bands of the dataset.

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.

As well as the many benefits associated with the evolution of multispectral sensors into hyperspectral sensors there is also a considerable increase in storage space and the computational load to process the data. Consequently the remote sensing ommunity is investigating and developing statistical methods to alleviate these problems. / The research presented here investigates several approaches to reducing the dimensionality of hyperspectral remotely sensed data while maintaining the levels of accuracy achieved using the full dimensionality of the data. It was conducted with an emphasis on applications in maximum likelihood classification (MLC) of hyperspectral image data. An inherent characteristic of hyperspectral data is that adjacent bands are typically highly correlated and this results in a high level of redundancy in the data. The high correlations between adjacent bands can be exploited to realise significant reductions in the dimensionality of the data, for a negligible reduction in classification accuracy. / The high correlations between neighbouring bands is related to their response functions overlapping with each other by a large amount. The spectral band filter functions were modelled for the HyMap instrument that acquires hyperspectral data used in this study. The results were compared with measured filter function data from a similar, more recent HyMap instrument. The results indicated that on average HyMap spectral band filter functions exhibit overlaps with their neighbouring bands of approximately 60%. This is considerable and partly accounts for the high correlation between neighbouring spectral bands on hyperspectral instruments. / A hyperspectral HyMap image acquired over an agricultural region in the south west of Western Australia has been used for this research. The image is composed of 512 × 512 pixels, with each pixel having a spatial resolution of 3.5 m. The data was initially reduced from 128 spectral bands to 82 spectral bands by removing the highly overlapping spectral bands, those which exhibit high levels of noise and those bands located at strong atmospheric absorption wavelengths. The image was examined and found to contain 15 distinct spectral classes. Training data was selected for each of these classes and class spectral mean and covariance matrices were generated. / The discriminant function for MLC makes use of not only the measured pixel spectra but also the sample class covariance matrices. This thesis first examines reducing the parameterization of these covariance matrices for use by the MLC algorithm. The full dimensional spectra are still used for the classification but the number of parameters needed to describe the covariance information is significantly reduced. When a threshold of 0.04 was used in conjunction with the partial correlation matrices to identify low values in the inverse covariance matrices, the resulting classification accuracy was 96.42%. This was achieved using only 68% of the elements in the original covariance matrices. / Both wavelet techniques and cubic splines were investigated as a means of representing the measured pixel spectra with considerably fewer bands. Of the different mother wavelets used, it was found that the Daubechies-4 wavelet performed slightly better than the Haar and Daubechies-6 wavelets at generating accurate spectra with the least number of parameters. The wavelet techniques investigated produced more accurately modelled spectra compared with cubic splines with various knot selection approaches. A backward stepwise knot selection technique was identified to be more effective at approximating the spectra than using regularly spaced knots. A forward stepwise selection technique was investigated but was determined to be unsuited to this process. / All approaches were adapted to process an entire hyperspectral image and the subsequent images were classified using MLC. Wavelet approximation coefficients gave slightly better classification results than wavelet detail coefficients and the Haar wavelet proved to be a more superior wavelet for classification purposes. With 6 approximation coefficients, the Haar wavelet could be used to classify the data with an accuracy of 95.6%. For 11 approximation coefficients this figure increased to 96.1%. / First and second derivative spectra were also used in the classification of the image. The first and second derivatives were determined for each of the class spectral means and for each band the standard deviations were calculated of both the first and second derivatives. Bands were then ranked in order of decreasing standard deviation. Bands showing the highest standard deviations were identified and the derivatives were generated for the entire image at these wavelengths. The resulting first and second derivative images were then classified using MLC. Using 25 spectral bands classification accuracies of approximately 96% and 95% were achieved using the first and second derivative images respectively. These results are comparable with those from using wavelets although wavelets produced higher classification accuracies when fewer coefficients were used.

Thesis (Msc Food Sc (Food Science))--University of Stellenbosch, 2009. / The use of near infrared (NIR) hyperspectral imaging and hyperspectral image analysis for distinguishing between whole maize kernels of varying degrees of hardness and fungal infected and non-infected kernels have been investigated. Near infrared hyperspectral images of whole maize kernels of varying degrees of hardness were acquired using a Spectral Dimensions MatrixNIR camera with a spectral range of 960-1662 nm as well as a sisuChema SWIR (short wave infrared) hyperspectral pushbroom imaging system with a spectral range of 1000-2498 nm. Exploratory principal component analysis (PCA) on absorbance images was used to remove background, bad pixels and shading. On the cleaned images, PCA could be used effectively to find histological classes including glassy (hard) and floury (soft) endosperm. PCA illustrated a distinct difference between floury and glassy endosperm along principal component (PC) three. Interpreting the PC loading line plots important absorbance peaks responsible for the variation were 1215, 1395 and 1450 nm, associated with starch and moisture for both MatrixNIR images (12 and 24 kernels). The loading line plots for the sisuChema (24 kernels) illustrated peaks of importance at the aforementioned wavelengths as well as 1695, 1900 and 1940 nm, also associated with starch and moisture. Partial least squares-discriminant analysis (PLS-DA) was applied as a means to predict whether the different endosperm types observed, were glassy or floury. For the MatrixNIR image (12 kernels), the PLS-DA model exhibited a classification rate of up to 99% for the discrimination of both floury and glassy endosperm. The PLS-DA model for the second MatrixNIR image (24 kernels) yielded a classification rate of 82% for the discrimination of glassy and 73% for floury endosperm. The sisuChema image (24 kernels) yielded a classification rate of 95% for the discrimination of floury and 92% for glassy endosperm. The fungal infected and sound whole maize kernels were imaged using the same instruments. Background, bad pixels and shading were removed by applying PCA on absorbance images. On the cleaned images, PCA could be used effectively to find the infected regions, pedicle as well as non-infected regions. A distinct difference between infected and sound kernels was illustrated along PC1. Interpreting the PC loading line plots showed important absorbance peaks responsible for the variation and predominantly associated with starch and moisture: 1215, 1450, 1480, 1690, 1940 and 2136 nm for both MatrixNIR images (15 and 21 kernels). The MatrixNIR image (15 kernels) exhibited a PLS-DA classification rate of up to 96.1% for the discrimination of infected kernels and the sisuChema had a classification rate of 99% for the same region of interest. The The iv sisuChema image (21-kernels) had a classification rate for infected kernels of 97.6% without pre-processing, 97.7% with multiplicative scatter correction (MSC) and 97.4% with standard normal variate (SNV). Near infrared hyperspectral imaging is a promising technique, capable of distinguishing between maize kernels of varying hardness and between fungal infected and sound kernels. While there are still limitations with hardware and software, these results provide the platform which would greatly assist with the determination of maize kernel hardness in breeding programmes without having to destroy the kernel. Further, NIR hyperspectral imaging could serve as an objective, rapid tool for identification of fungal infected kernels.

Nessa dissertação é investigada e testada uma metodologia para otimizar os parâmetros do kernel do classificador Support Vector Machines (SVM). Experimentos são realizados utilizando dados de imagens em alta dimensão. Imagens em alta dimensão abrem novas possibilidades para a classificação de imagens de sensoriamento remoto que capturam cenas naturais. É sabido que classes que são espectralmente muito similares, i.e, classes que possuem vetores de média muito próximos podem não obstante serem separadas com alto grau de acurácia em espaço de alta dimensão, desde que a matriz de covariância apresente diferenças significativas. O uso de dados de imagens em alta dimensão pode apresentar, no entanto, alguns desafios metodológicos quando aplicado um classificador paramétrico como o classificador de Máxima Verossimilhança Gaussiana. Conforme aumenta a dimensionalidade dos dados, o número de parâmetros a serem estimados a partir de um número geralmente limitado de amostras de treinamento também aumenta. Esse fato pode ocasionar estimativas pouco confiáveis, que por sua vez resultam em baixa acurácia na imagem classificada. Existem diversos abordagens propostas na literatura para minimizar esse problema. Os classificadores não paramétricos podem ser uma boa alternativa para mitigar esse problema. O SVM atualmente tem sido investigado na classificação de dados de imagens em alta-dimensão com número limitado de amostras de treinamento. Para que o classificador SVM seja utilizado com sucesso é necessário escolher uma função de kernel adequada, bem como os parâmetros dessa função. O kernel RBF tem sido frequentemente mencionado na literatura por obter bons resultados na classificação de imagens de sensoriamento remoto. Neste caso, dois parâmetro devem ser escolhidos para o classificador SVM: (1) O parâmetro de margem (C) que determina um ponto de equilíbrio razoável entre a maximização da margem e a minimização do erro de classificação, e (2) o parâmetro que controla o raio do kernel RBF. Estes dois parâmetros podem ser vistos como definindo um espaço de busca. O problema nesse caso consiste em procurar o ponto ótimo que maximize a acurácia do classificador SVM. O método de Busca em Grade é baseado na exploração exaustiva deste espaço de busca. Esse método é proibitivo do ponto de vista do tempo de processamento, sendo utilizado apenas com propósitos comparativos. Na prática os métodos heurísticos são a abordagem mais utilizada, proporcionado níveis aceitáveis de acurácia e tempo de processamento. Na literatura diversos métodos heurísticos são aplicados ao problema de classificação de forma global, i.e, os valores selecionados são aplicados durante todo processo de classificação. Esse processo, no entanto, não considera a diversidade das classes presentes nos dados. Nessa dissertação investigamos a aplicação da heurística Simulated Annealing (Recozimento Simulado) para um problema de múltiplas classes usando o classificador SVM estruturado como uma arvore binária. Seguindo essa abordagem, os parâmetros são estimados em cada nó da arvore binária, resultado em uma melhora na acurácia e tempo razoável de processamento. Experimentos são realizados utilizando dados de uma imagem hiperespectral disponível, cobrindo uma área de teste com controle terrestre bastante confiável. / In this dissertation we investigate and test a methodology to optimize the kernel parameters in a Support Vector Machines classifier. Experiments were carried out using remote sensing high-dimensional image data. High dimensional image data opens new possibilities in the classification of remote sensing image data covering natural scenes. It is well known that classes that are spectrally very similar, i.e., classes that show very similar mean vectors can notwithstanding be separated with an high degree of accuracy in high dimensional spaces, provided that their covariance matrices differ significantly. The use of high-dimensional image data may present, however, some drawbacks when applied in parametric classifiers such as the Gaussian Maximum Likelihood classifier. As the data dimensionality increases, so does the number of parameters to be estimated from a generally limited number of training samples. This fact results in unreliable estimates for the parameters, which in turn results in low accuracy in the classified image. There are several approaches proposed in the literature to minimize this problem. Non-parametric classifiers may provide a sensible way to overcome this problem. Support Vector Machines (SVM) have been more recently investigated in the classification of high-dimensional image data with a limited number of training samples. To achieve this end, a proper kernel function has to be implemented in the SVM classifier and the respective parameters selected properly. The RBF kernel has been frequently mentioned in the literature as providing good results in the classification of remotely sensed data. In this case, two parameters must be chosen in the SVM classification: (1) the margin parameter (C) that determines the trade-off between the maximization of the margin in the SVM and minimization of the classification error, and (2) the parameter that controls the radius in the RBF kernel. These two parameters can be seen as defining a search space, The problem here consists in finding an optimal point that maximizes the accuracy in the SVM classifier. The Grid Search approach is based on an exhaustive exploration in the search space. This approach results prohibitively time consuming and is used only for comparative purposes. In practice heuristic methods are the most commonly used approaches, providing acceptable levels of accuracy and computing time. In the literature several heuristic methods are applied to the classification problem in a global fashion, i.e., the selected values are applied to the entire classification process. This procedure, however, does not take into consideration the diversity of the classes present in the data. In this dissertation we investigate the application of Simulated Annealing to a multiclass problem using the SVM classifier structured as a binary tree. Following this proposed approach, the parameters are estimated at every level of the binary tree, resulting in better accuracy and a reasonable computing time. Experiments are done using a set of hyperspectral image data, covering a test area with very reliable ground control available.