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.

Submitted by Fabio Sobreira Campos da Costa (fabio.sobreira@ufpe.br) on 2016-06-29T11:48:02Z No. of bitstreams: 2 license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) Dissertação de Mestrado - Lívia Rodrigues e Brito.pdf: 11880513 bytes, checksum: cdf56fe284940b9c31e62271753b913f (MD5) / Made available in DSpace on 2016-06-29T11:48:02Z (GMT). No. of bitstreams: 2 license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) Dissertação de Mestrado - Lívia Rodrigues e Brito.pdf: 11880513 bytes, checksum: cdf56fe284940b9c31e62271753b913f (MD5) Previous issue date: 2014-08-25 / CNPq / A adição de aditivos detergentes dispersantes nas gasolinas brasileiras será obrigatória a partir de julho de 2015. É necessário, portanto, desenvolver uma metodologia que permita quantificar esses aditivos para verificar o cumprimento da lei. Neste trabalho, é proposto um método que associa a técnica do ring-oven com as imagens hiperespectrais no infravermelho próximo (NIR-HI). Como os aditivos são adicionados em baixas concentrações, a técnica do ring-oven foi empregada para concentrá-los previamente à análise por NIR-HI. Anéis foram produzidos a partir de amostras de gasolinas comum adicionadas dos aditivos (denominados G, T, W e Y) fornecidos pela Agência Nacional do Petróleo, Gás Natural e Biocombustíveis (ANP) e as imagens adquiridas utilizando uma câmera hiperespectral (SisuCHEMA). Três estratégias de extração dos espectros do anel foram testadas a fim de se escolher a mais rápida e objetiva. A estratégia escolhida se baseia nos histogramas dos escores da primeira componente principal das imagens analisadas individualmente. Modelos de calibração individuais para cada aditivo foram construídos empregando a regressão por mínimos quadrados parciais (PLS), por isso, fez-se necessária uma etapa prévia de classificação. O melhor resultado para classificação foi obtido empregando a análise discriminante linear (LDA) associada ao algoritmo genético (GA) para seleção de variáveis, o qual apresentou uma taxa de classificações corretas de 92,31 %. Observou-se que a maioria dos erros de classificação envolveram amostras dos aditivos G e W. Um único modelo de regressão para esses dois aditivos foi, então, construído e seu erro foi equivalente aos dos modelos individuais. Os modelos de regressão apresentaram erros médios de predição entre 2 e 15 %. Esses resultados mostram que a metodologia proposta pode ser utilizada para determinar as concentrações dos aditivos com confiabilidade e garantir que eles estão sendo adicionados conforme a lei. / The addition of detergent dispersant additives to Brazilian gasoline will be mandatory from July 2015. It is necessary, therefore, to develop a methodology that allows quantifying these additives to verify their compliance with the law. In this work, a method that associates the ring-oven technique with near infrared hyperspectral images (NIR-HI) is proposed. Because the additives are added in low concentrations, the ring-oven technique was employed to concentrate them prior to the NIR-HI analysis. Rings were produced from samples of gasolines without additives spiked with additives (called G, T, W and Y) provided by the National Agency of Petroleum, Natural Gas and Biofuels (ANP) and the images were acquired using a hyperspectral camera (SisuCHEMA). Three strategies for extraction of the ring spectra were tested in order to select the faster and most objective. The chosen strategy is based on the histograms of the first principal component scores of the images analyzed individually. Regression models were built for each additive using partial least squares (PLS) regression, so it was necessary to have a previous classification stage. The best classification result was obtained using the linear discriminant analysis (LDA) associated with the genetic algorithm (GA) for variable selection, which showed a correct classification rate of 92.31 %. It was observed that most of the misclassification errors involved the samples of the G and W additives. A single regression model was then built for these two additives and its error was equivalent to the errors of the individual models. The regression models showed average prediction errors between 2 and 15 %. These results show that the proposed methodology can be used to determine the additive concentrations with reliability and to ensure that they are been added according to the law.

Made available in DSpace on 2016-12-23T14:37:32Z (GMT). No. of bitstreams: 1 DISSERTACAO QUARTO JUNIOR.pdf: 3333993 bytes, checksum: 71f8ffd0a33ee359bda776da65bc6e40 (MD5) Previous issue date: 2007-02-27 / Este trabalho teve como objetivos: i) analisar a inter-relação da reflectância das bandas do vermelho (Ver) e do infravermelho-próximo (IVP) e do Índice de Vegetação da Diferença Normalizada (NDVI) dos sensores multiespectrais MODIS (Moderate Resolution Imaging Spectroradiometer), ETM+ (Enhanced Thematic Mapper Plus) e HRV (High Resolution Visible); e ii) avaliar a importância das características espectrais das regiões do pico do verde (~550 nm), da borda do vermelho (~680 nm a ~780 nm) e da absorção de água na folha (~940 nm) nas bandas Ver, IVP e no NDVI. Uma imagem hiperespectral do Hyperion foi adquirida sobre a região de Campos dos Goytacazes, Estado do Rio de Janeiro, tendo os efeitos atmosféricos corrigidos. A simulação das bandas foi realizada a partir de 210 e 50 amostras de dados de reflectância hiperespectral, coletadas da imagem Hyperion, em áreas de cana-de-açúcar e de floresta natural, respectivamente. Como resultados têm: a) as relações entre as bandas individuais e o NDVI dos diferentes sensores têm comportamento variado, por exemplo, foi observado que as diferenças de Ver entre os diferentes sensores foram significativas, o mesmo não ocorrendo para IVP; b) as translações do NDVIETM+ para NDVIMODIS e do NDVIETM+ para NDVIHRV, têm boa relação, ambas com R2=0,71, já a translação do NDVIHRV para NDVIMODIS apresenta uma relação inferior (R2=0,31), e c) quanto à influência das características espectrais frente às bandas e o NDVI, verifica-se que à medida que as características são mais incorporadas, maiores são as diferenças quando comparados às bandas e o NDVI do MODIS. / The objectives of this paper were: i) to analyze cross-sensor relation of the normalìzed difference vegetation index (NDVI) and red/near-infrared reflectance of the multispectral sensors MODIS ( Moderate Resolution Imaging Spectroradiometer ), ETM+ ( Enhanced Thematic Mapper Plus ) and HRV ( High Resolution Visible ); and ii) to evaluate the importance of the spectral features green peak (~550 nm), red edge (~680 nm to ~780 nm) and leaf liquid water absorption region (~940 nm) in red/near-infrared reflectance and in NDVI. A Hyperion hyperspectral image acquired over Campos dos Goytacazes, State of Rio de Janeiro in Brazil, having the atmospherical effects corrected. The simulated bandpasses were done for 210 and 50 samples of hyperspectral reflectance data over sugar cane and natural forest, respectively, collected from Hyperion reflectance data. The results are: a) the relationship between the individual bands and NDVI of the different sensors have diverse behavior, for example, it was observed that the differences of red among sensors are statistical significant and the same is not observed for NIR; b) Translation of NDVIETM+ to NDVIMODIS and of NDVIETM+ to NDVIHRV, present good relationship, both with R2=0.71, nevertheless the translation of NDVIHRV to NDVIMODIS is not good (R2=0.31); c) it is verified that when the spectral features are incorporated in the reflectance bands and NDVI, the differences are bigger when compared to the bands and NDVI of MODIS.

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.

La thèse propose une méthodologie d’analyse rapide basée sur l’étude de l’image hyperspectrale Raman d’un produit pharmaceutique falsifié sous forme solide afin, d’une part d’identifier les substances présentes, et d’autre part estimer la teneur du principe actif dans l’échantillon sans étalonnage préalable.La présence de produits pharmaceutiques falsifiés est un véritable enjeu de santé publique. Ce type de produits de santé est facilement disponible sur internet, et beaucoup d’exemples montrent leur dangerosité. De nombreuses techniques sont disponibles pour analyser ces produits et ainsi participer à la lutte contre la falsification de médicament. La combinaison de ces techniques analytiques permet une caractérisation approfondie de l’échantillon. Cependant, peu de techniques analytiques procurent l’ensemble des informations chimiques.L’imagerie chimique Raman est une technique qui répond aux exigences requises pour l’analyse de produits falsifiés sous forme solide. En effet, cette technique peu destructive permet de réutiliser l’échantillon pour des analyses complémentaires. L’imagerie chimique Raman combine les trois disciplines de spectroscopie Raman, microscopie et chimiométrie. Cette technique réalise des mesures successives de spectres Raman sur des zones adjacentes couvrant la surface de l’échantillon. Elle intègre donc des informations spatiales et spectrales. Les méthodes chimiométriques dites de résolution (MCR-ALS et DCLS) analysent le jeu de spectres pour extraire des informations qualitatives (détection des spectres purs du mélange) et des informations quantitatives (estimation de la concentration de la substance active). La méthodologie a été optimisée et validée avec des échantillons préparés en laboratoire, puis appliquée à des échantillons réels authentiques et falsifiés. La sensibilité de la méthode qualitative a été démontrée par la détection d’un principe actif antibiotique à la teneur de 0,3% m/m dans un comprimé à visée anabolisante. De plus la méthode a été capable de détecter les substances utilisées pour le traitement de dysfonctions sexuelles (sildénafil, tadalafil, vardénafil, dapoxétine). Malgré une forte émission de fluorescence, la méthode a réussi à discriminer les 3 sels de clopidogrel (hydrogénosulfate, bésilate et chlorhydrate). L’analyse quantitative directe sur des échantillons de Viagra® et de Plavix® a été jugée convenable avec une déviation de la teneur entre -15% et +24%. Cette déviation est considérée acceptable pour évaluer le risque sanitaire pour le patient et alerter les autorités de santé.Dans le cadre de l’analyse des produits falsifiés, il a été démontré que la micro-spectroscopie Raman associée aux méthodes chimiométriques permet de réaliser un « screening spectroscopique » des composants de l’échantillon, d’identifier les substances chimiques, de visualiser leur distribution sur la surface de l’échantillon et d’estimer leur teneur par « quantification directe ». / The thesis proposes a rapid methodology of analysis based on the Raman hyperspectral image study of a solid form falsified pharmaceutical product in order to identify the substances in the sample and to estimate the content of the active ingredient in the sample without prior calibration.The presence of falsified pharmaceuticals is a real public health issue. This type of products is easily available on the internet, and many examples show their dangerousness. Many techniques are available for the analysis of these products and thus participate in the fight against drug falsification. The combination of these analytical techniques allows a comprehensive characterization of the sample. However few analytical techniques provide all the chemical information. Raman chemical imaging is a technique that meets the requirements for the analysis of falsified products in solid form. Indeed, this non-destructive technique makes it possible the reuse of the sample for additional testing. Raman chemical imaging combines the three disciplines of Raman spectroscopy, microscopy and chemometrics. This technique performs successive measurements of Raman spectra on adjacent location covering the surface of the sample. It therefore collects spatial and spectral information. The so-called resolution chemometric methods analyse the set of spectra in order to extract qualitative information (detection of pure spectra in the mixture) and quantitative information (estimate of the concentration of the chemical substance). The methodology was optimized and validated with samples prepared in the laboratory, and then applied to genuine and falsified real samples. The sensitivity of the qualitative method was demonstrated with the detection of an antibiotic active ingredient at a content of 0.3% m/m in an anabolic tablet. Moreover, the method was able to distinguish substances used for the treatment of sexual dysfunctions (sildenafil, tadalafil, vardenafil, dapoxetine). Despite a high fluorescence emission, the method successfully discriminated the 3 salts of clopidogrel (hydrogen sulfate, besylate and hydrochloride). Direct quantitative analysis of samples of Viagra® and Plavix® was found to be appropriate with a deviation between -15% and +24%. This deviation is considered acceptable to assess the health risk to the patient and to alert health authorities.For the analysis of falsified products, it has been demonstrated that Raman micro-spectroscopy combined with chemometric methods allows to perform a "spectroscopic screening" of the components in the sample, to identify chemical substances, to visualize their distribution on the sample surface and to estimate their content by "direct quantification".

L'objectif de ce travail de recherche est le développement d'un système fibré d'imagerie point par point d'auto fluorescence multi-excitation, de tissus biologiques en utilisant la technique de fluorescence hyperspectrale et l'étude d'un système de tomographie optique cohérente comme possible modalité supplémentaire. La première partie de ce rapport présente les propriétés optique des tissus biologiques et les fluorophores pertinents pour la détection de tumeurs cancéreuses. La deuxième partie présente l'instrumentation du système d'imagerie de fluorescence et l'analyse hyperspectrale des résultats obtenus in vitro.Il est démontré la pertinence de ce type d'analyse qui permet de déterminer la concentration de certains fluorophores. La troisième partie présente le système de tomographie optique cohérente appelé "scan free" OCT car il permet de réaliser des images sans déplacement d'éléments optiques. Ce système est caractérisé et présente des fonctionnalités intéressantes comme la compensation de la dispersion dépendante de la profondeur. Les divers résultats obtenus montrent que ces deux techniques sont complémentaires car elles apportent des informations de nature différentes. La première technique donne de se informations sur la composition biochimique des tissus, la seconde donne des information sur la structure. / The aim of this activity is the development of a mono point imaging fiber system which uses hyperspectral multi-excitation auto fluorescence technique for biological tissues and the study of an Optical Coherence Tomography system like another modality. At first, this report presents the optical properties of biological tissues and the relevant fluorophores for cancerous tumors detection. Secondly, the fluorescence imaging system instrumentation and hyperspectral analysis are presented with in vitro results. The third part presents the "scan free" optical coherence tomography system which is able to image without optical displacement. It's characterized and have interesting functionality like depth dependant dispersion compensation. These both techniques are complementary because they get different kind of information. The information of the first one is about biochemical composition of the tissues and the information of the second one is about the stucture.

Upland peatlands provide vital ecosystem services, especially carbon storage and biodiversity. However, large areas of peatland are heavily degraded in the UK. When peat becomes exposed the potential for it to actively sequester carbon is greatly reduced and carbon stores are rapidly lost through erosion. Peatland restoration is a tool that addresses the government public service agreement targets for biodiversity, and soil and water protection in uplands. Blanket bogs are a UK Biodiversity Action Plan priority habitat. Many areas fall under designations for sites of protection under the EU habitats directive which is aimed at bringing the areas into ‘favourable condition’.The Moors for the Future Partnership is restoring large areas of badly eroded peat in the Peak District National Park to stabilise the surface and re-establish ecosystem functions. Monitoring is of pivotal importance to judge the success of the restoration work. This project assesses the suitability of high resolution remote sensing as an alternative monitoring tool to traditional field based plot surveys which are both time consuming and expensive. Remote sensing has been seen as a potential tool for mapping and monitoring peatlands, but to date the application of high spatial and spectral resolution remote sensing to monitoring peatland restoration has not been fully investigated. A floristic restoration trajectory has been established using a statistical classification (TWINSPAN) of vegetation cover data combined with expert knowledge of previous restoration, and autecology of the moorland species. Hyperspectral classification techniques were applied, including: Spectral Angle Mapping (SAM); Support Vector Machines (SVM); and maximum likelihood classification using both Minimum Noise Fraction (MNF), and narrow band vegetation indices. A successful classification of the restoration succession has been achieved. A predictive model for vegetation cover of plant functional types has been produced using a Partial Least Squares Regression and applied to the whole restoration site at the landscape-scale. RMSEs of between 10 and 16% indicate that the models can be used as a useful operational tool. A spectral library of key moorland species and their phenological response has been established using field spectroscopy in parallel to the image analysis. This has enabled the suggestion that the species are most separable from one another in July and it is recommended that this is the optimal month for remote sensing monitoring. This has facilitated the development of a set of recommendations for the most appropriate vegetation indices to use throughout the year depending species to be differentiated. High spatial and spectral resolution remote sensing data is needed to successfully characterise the vegetation response to restoration management in the upland peatland environment.

El diagnóstico final de la mayoría de tipos de cáncer lo realiza un médico experto en anatomía patológica que examina muestras tisulares o celulares sospechosas extraídas del paciente. Actualmente, esta evaluación depende en gran medida de la experiencia del médico y se lleva a cabo de forma cualitativa mediante técnicas de imagen tradicionales como la microscopía óptica. Esta tarea tediosa está sujeta a altos grados de subjetividad y da lugar a niveles de discordancia inadecuados entre diferentes patólogos, especialmente en las primeras etapas de desarrollo del cáncer. La espectroscopía infrarroja por Transformada de Fourier (siglas FTIR en inglés) es una tecnología ampliamente utilizada en la industria que recientemente ha demostrado una capacidad creciente para mejorar el diagnóstico de diferentes tipos de cáncer. Esta técnica aprovecha las propiedades del infrarrojo medio para excitar los modos vibratorios de los enlaces químicos que forman las muestras biológicas. La principal señal generada consiste en un espectro de absorción que informa sobre la composición química de la muestra iluminada. Los microespectrómetros FTIR modernos, compuestos por complejos componentes ópticos y detectores matriciales de alta sensibilidad, permiten capturar en un laboratorio de investigación común imágenes hiperespectrales de alta calidad que aúnan información química y espacial. Las imágenes FTIR son estructuras de datos ricas en información que se pueden analizar individualmente o junto con otras modalidades de imagen para realizar diagnósticos patológicos objetivos. Por lo tanto, esta técnica de imagen emergente alberga un alto potencial para mejorar la detección y la graduación del riesgo del paciente en el cribado y vigilancia de cáncer. Esta tesis estudia e implementa diferentes metodologías y algoritmos de los campos interrelacionados de procesamiento de imagen, visión por ordenador, aprendizaje automático, reconocimiento de patrones, análisis multivariante y quimiometría para el procesamiento y análisis de imágenes hiperespectrales FTIR. Estas imágenes se capturaron con un moderno microscopio FTIR de laboratorio a partir de muestras de tejidos y células afectadas por cáncer colorrectal y de piel, las cuales se prepararon siguiendo protocolos alineados con la práctica clínica actual. Los conceptos más relevantes de la espectroscopía FTIR se investigan profundamente, ya que deben ser comprendidos y tenidos en cuenta para llevar a cabo una correcta interpretación y tratamiento de sus señales especiales. En particular, se revisan y analizan diferentes factores fisicoquímicos que influyen en las mediciones espectroscópicas en el caso particular de muestras biológicas y pueden afectar críticamente su análisis posterior. Todos estos conceptos y estudios preliminares entran en juego en dos aplicaciones principales. La primera aplicación aborda el problema del registro o alineación de imágenes hiperespectrales FTIR con imágenes en color adquiridas con microscopios tradicionales. El objetivo es fusionar la información espacial de distintas muestras de tejido medidas con esas dos modalidades de imagen y centrar la discriminación en las regiones seleccionadas por los patólogos, las cuales se consideran más relevantes para el diagnóstico de cáncer colorrectal. En la segunda aplicación, la espectroscopía FTIR se lleva a sus límites de detección para el estudio de las entidades biomédicas más pequeñas. El objetivo es evaluar las capacidades de las señales FTIR para discriminar de manera fiable diferentes tipos de células de piel que contienen fenotipos malignos. Los estudios desarrollados contribuyen a la mejora de métodos de decisión objetivos que ayuden al patólogo en el diagnóstico final del cáncer. Además, revelan las limitaciones de los protocolos actuales y los problemas intrínsecos de la tecnología FTIR moderna, que deberían abordarse para permit / The final diagnosis of most types of cancers is performed by an expert clinician in anatomical pathology who examines suspicious tissue or cell samples extracted from the patient. Currently, this assessment largely relies on the experience of the clinician and is accomplished in a qualitative manner by means of traditional imaging techniques, such as optical microscopy. This tedious task is subject to high degrees of subjectivity and gives rise to suboptimal levels of discordance between different pathologists, especially in early stages of cancer development. Fourier Transform infrared (FTIR) spectroscopy is a technology widely used in industry that has recently shown an increasing capability to improve the diagnosis of different types of cancer. This technique takes advantage of the ability of mid-infrared light to excite the vibrational modes of the chemical bonds that form the biological samples. The main generated signal consists of an absorption spectrum that informs of the chemical composition of the illuminated specimen. Modern FTIR microspectrometers, composed of complex optical components and high-sensitive array detectors, allow the acquisition of high-quality hyperspectral images with spatially-resolved chemical information in a common research laboratory. FTIR images are information-rich data structures that can be analysed alone or together with other imaging modalities to provide objective pathological diagnoses. Hence, this emerging imaging technique presents a high potential to improve the detection and risk stratification in cancer screening and surveillance. This thesis studies and implements different methodologies and algorithms from the related fields of image processing, computer vision, machine learning, pattern recognition, multivariate analysis and chemometrics for the processing and analysis of FTIR hyperspectral images. Those images were acquired with a modern benchtop FTIR microspectrometer from tissue and cell samples affected by colorectal and skin cancer, which were prepared by following protocols close to the current clinical practise. The most relevant concepts of FTIR spectroscopy are thoroughly investigated, which ought to be understood and considered to perform a correct interpretation and treatment of its special signals. In particular, different physicochemical factors are reviewed and analysed, which influence the spectroscopic measurements for the particular case of biological samples and can critically affect their later analysis. All these knowledge and preliminary studies come into play in two main applications. The first application tackles the problem of registration or alignment of FTIR hyperspectral images with colour images acquired with traditional microscopes. The aim is to fuse the spatial information of distinct tissue samples measured by those two imaging modalities and focus the discrimination on regions selected by the pathologists, which are meant to be the most relevant areas for the diagnosis of colorectal cancer. In the second application, FTIR spectroscopy is pushed to their limits of detection for the study of the smallest biomedical entities. The aim is to assess the capabilities of FTIR signals to reliably discriminate different types of skin cells containing malignant phenotypes. The developed studies contribute to the improvement of objective decision methods to support the pathologist in the final diagnosis of cancer. In addition, they reveal the limitations of current protocols and intrinsic problems of modern FTIR technology, which should be tackled in order to enable its transference to anatomical pathology laboratories in the future. / El diagnòstic final de la majoria de tipus de càncer ho realitza un metge expert en anatomia patològica que examina mostres tissulars o cel¿lulars sospitoses extretes del pacient. Actualment, aquesta avaluació depèn en gran part de l'experiència del metge i es porta a terme de forma qualitativa mitjançant tècniques d'imatge tradicionals com la microscòpia òptica. Aquesta tasca tediosa està subjecta a alts graus de subjectivitat i dóna lloc a nivells de discordança inadequats entre diferents patòlegs, especialment en les primeres etapes de desenvolupament del càncer. L'espectroscòpia infraroja per Transformada de Fourier (sigles FTIR en anglès) és una tecnologia àmpliament utilitzada en la indústria que recentment ha demostrat una capacitat creixent per millorar el diagnòstic de diferents tipus de càncer. Aquesta tècnica aprofita les propietats de l'infraroig mitjà per excitar els modes vibratoris dels enllaços químics que formen les mostres biològiques. El principal senyal generat consisteix en un espectre d'absorció que informa sobre la composició química de la mostra il¿luminada. Els microespectrómetres FTIR moderns, compostos per complexos components òptics i detectors matricials d'alta sensibilitat, permeten capturar en un laboratori d'investigació comú imatges hiperespectrals d'alta qualitat que uneixen informació química i espacial. Les imatges FTIR són estructures de dades riques en informació que es poden analitzar individualment o juntament amb altres modalitats d'imatge per a realitzar diagnòstics patològics objectius. Per tant, aquesta tècnica d'imatge emergent té un alt potencial per a millorar la detecció i la graduació del risc del pacient en el cribratge i vigilància de càncer. Aquesta tesi estudia i implementa diferents metodologies i algoritmes dels camps interrelacionats de processament d'imatge, visió per ordinador, aprenentatge automàtic, reconeixement de patrons, anàlisi multivariant i quimiometria per al processament i anàlisi d'imatges hiperespectrals FTIR. Aquestes imatges es van capturar amb un modern microscopi FTIR de laboratori a partir de mostres de teixits i cèl¿lules afectades per càncer colorectal i de pell, les quals es van preparar seguint protocols alineats amb la pràctica clínica actual. Els conceptes més rellevants de l'espectroscòpia FTIR s'investiguen profundament, ja que han de ser compresos i tinguts en compte per dur a terme una correcta interpretació i tractament dels seus senyals especials. En particular, es revisen i analitzen diferents factors fisicoquímics que influeixen en els mesuraments espectroscòpiques en el cas particular de mostres biològiques i poden afectar críticament la seua anàlisi posterior. Tots aquests conceptes i estudis preliminars entren en joc en dues aplicacions principals. La primera aplicació aborda el problema del registre o alineació d'imatges hiperespectrals FTIR amb imatges en color adquirides amb microscopis tradicionals. L'objectiu és fusionar la informació espacial de diferents mostres de teixit mesurades amb aquestes dues modalitats d'imatge i centrar la discriminació en les regions seleccionades pels patòlegs, les quals es consideren més rellevants per al diagnòstic de càncer colorectal. En la segona aplicació, l'espectroscòpia FTIR es porta als seus límits de detecció per a l'estudi de les entitats biomèdiques més xicotetes. L'objectiu és avaluar les capacitats dels senyals FTIR per discriminar de manera fiable diferents tipus de cèl¿lules de pell que contenen fenotips malignes. Els estudis desenvolupats contribueixen a la millora de mètodes de decisió objectius que ajuden el patòleg en el diagnòstic final del càncer. A més, revelen les limitacions dels protocols actuals i els problemes intrínsecs de la tecnologia FTIR moderna, que haurien d'abordar per permetre la seva transferència als laboratoris d'anatomia patològica en el futur. / Peñaranda Gómez, FJ. (2018). Application of artificial vision algorithms to images of microscopy and spectroscopy for the improvement of cancer diagnosis [Tesis doctoral no publicada]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/99748 / TESIS

Hyperspectral signature classification is a kind of quantitative analysis approach for hyperspectral imagery which performs detection and classification of the constituent materials at pixel level in the scene. The classification procedure can be operated directly on hyperspectral data or performed by using some features extracted from corresponding hyperspectral signatures containing information like signature energy or shape. In this paper, we describe a technique that applies non-homogeneous hidden Markov chain (NHMC) models to hyperspectral signature classification. The basic idea is to use statistical models (NHMC models) to characterize wavelet coefficients which capture the spectrum structural information at multiple levels. Experimental results show that the approach based on NHMC models outperforms existing approaches relevant in classification tasks.

This thesis investigates the possibility of utilizing data from multiple modalities to enable an automated recycling system to separate ferrous from non-ferrous debris. The two methods sensor fusion and hallucinogenic sensor fusion were implemented in a four-step approach of deep CNNs. Sensor fusion implies that multiple modalities are run simultaneously during the operation of the system.The individual outputs are further fused, and the joint performance expects to be superior to having only one of the sensors. In hallucinogenic sensor fusion, the goal is to achieve the benefits of sensor fusion in respect to cost and complexity even when one of the modalities is reduced from the system. This is achieved by leveraging data from a more complex modality onto a simpler one in a student/teacher approach. As a result, the teacher modality will train the student sensor to hallucinate features beyond its visual spectra. Based on the results of a performed prestudy involving multiple types of modalities, a hyperspectral sensor was deployed as the teacher to complement a simple RGB camera. Three studies involving differently composed datasets were further conducted to evaluate the effectiveness of the methods. The results show that the joint performance of a hyperspectral sensor and an RGB camera is superior to both individual dispatches. It can also be concluded that training a network with hyperspectral images can improve the classification accuracy when operating with only RGB data. However, the addition of a hyperspectral sensor might be considered as superfluous as this report shows that the standardized shapes of industrial debris enable a single RGB to achieve an accuracy above 90%. The material used in this thesis can also be concluded to be suboptimal for hyperspectral analysis. Compared to the vegetation scenes, only a limited amount of additional data could be obtained by including wavelengths besides the ones representing red, green and blue.