In this contributon we evaluate on-line and off-line techniques to train a single hidden layer neural network classifier with logistic hidden and softmax output transfer functions on a multispectral pixel-by-pixel classification problem. In contrast to current practice a multiple class cross-entropy error function has been chosen as the function to be minimized. The non-linear diffierential equations cannot be solved in closed form. To solve for a set of locally minimizing parameters we use the gradient descent technique for parameter updating based upon the backpropagation technique for evaluating the partial derivatives of the error function with respect to the parameter weights. Empirical evidence shows that on-line and epoch-based gradient descent backpropagation fail to converge within 100,000 iterations, due to the fixed step size. Batch gradient descent backpropagation training is superior in terms of learning speed and convergence behaviour. Stochastic epoch-based training tends to be slightly more effective than on-line and batch training in terms of generalization performance, especially when the number of training examples is larger. Moreover, it is less prone to fall into local minima than on-line and batch modes of operation. (authors' abstract) / Series: Discussion Papers of the Institute for Economic Geography and GIScience

Nowadays, the industrial standard for infrared imaging systems is to interconnect an infrared photodetector array with a silicon-based read-out-integrated circuit pixel by pixel through existing indium bumping technology for infrared scene detection and then the signal is output optically through a LCD or other imaging devices. Motivated by the high-cost and low-resolution of such configurations, technology that up-converts infrared light to visible light and in particular, an inorganic/organic hybrid imaging upconverter has been developed. The end goal was to provide a high-efficiency and high-resolution alternative for infrared imaging. The inorganic/organic hybrid architecture takes advantage of both the high quantum efficiency of photo-detection for inorganic semiconductors, and the low-cost processing and the topologically perfect structure of organic semiconductors that does not require lattice matching for materials. Based on previous single-element hybrid infrared upconverter designs, both pixel-less and pixel-lated hybrid infrared imaging devices are presented, with experimental results, in this thesis. The pixel-less hybrid infrared imaging upconverter suppresses the lateral carrier diffusion by using a hybrid Schottky junction with an intrinsic interconnection layer between the inorganic and organic parts. The device was fabricated in one large-area mesa and proved that the emitting light spatially correlated with the infrared imaging shone at its back. This device is the first-ever hybrid pixel-less infrared upconverter to successfully demonstrate the imaging of infrared patterns. In contrast, the pixel-lated device consisted of 128 by 128 pixels, and each pixel was an individually working infrared upconverter that integrated a heterojunction phototransistor (HPT) and an organic light emitting diode (OLED). The HPT provides not only the photoresponse upon incoming infrared light but also an amplification of the photocurrent. The pixel-lated device also successfully demonstrated the first-ever upconversion of infrared light, up-converting a light with a wavelength of 1.5 μm to 520 nm.

Nowadays, the industrial standard for infrared imaging systems is to interconnect an infrared photodetector array with a silicon-based read-out-integrated circuit pixel by pixel through existing indium bumping technology for infrared scene detection and then the signal is output optically through a LCD or other imaging devices. Motivated by the high-cost and low-resolution of such configurations, technology that up-converts infrared light to visible light and in particular, an inorganic/organic hybrid imaging upconverter has been developed. The end goal was to provide a high-efficiency and high-resolution alternative for infrared imaging. The inorganic/organic hybrid architecture takes advantage of both the high quantum efficiency of photo-detection for inorganic semiconductors, and the low-cost processing and the topologically perfect structure of organic semiconductors that does not require lattice matching for materials. Based on previous single-element hybrid infrared upconverter designs, both pixel-less and pixel-lated hybrid infrared imaging devices are presented, with experimental results, in this thesis. The pixel-less hybrid infrared imaging upconverter suppresses the lateral carrier diffusion by using a hybrid Schottky junction with an intrinsic interconnection layer between the inorganic and organic parts. The device was fabricated in one large-area mesa and proved that the emitting light spatially correlated with the infrared imaging shone at its back. This device is the first-ever hybrid pixel-less infrared upconverter to successfully demonstrate the imaging of infrared patterns. In contrast, the pixel-lated device consisted of 128 by 128 pixels, and each pixel was an individually working infrared upconverter that integrated a heterojunction phototransistor (HPT) and an organic light emitting diode (OLED). The HPT provides not only the photoresponse upon incoming infrared light but also an amplification of the photocurrent. The pixel-lated device also successfully demonstrated the first-ever upconversion of infrared light, up-converting a light with a wavelength of 1.5 μm to 520 nm.

This paper evaluates the classification accuracy of three neural network classifiers on a satellite image-based pattern classification problem. The neural network classifiers used include two types of the Multi-Layer-Perceptron (MLP) and the Radial Basis Function Network. A normal (conventional) classifier is used as a benchmark to evaluate the performance of neural network classifiers. The satellite image consists of 2,460 pixels selected from a section (270 x 360) of a Landsat-5 TM scene from the city of Vienna and its northern surroundings. In addition to evaluation of classification accuracy, the neural classifiers are analysed for generalization capability and stability of results. Best overall results (in terms of accuracy and convergence time) are provided by the MLP-1 classifier with weight elimination. It has a small number of parameters and requires no problem-specific system of initial weight values. Its in-sample classification error is 7.87% and its out-of-sample classification error is 10.24% for the problem at hand. Four classes of simulations serve to illustrate the properties of the classifier in general and the stability of the result with respect to control parameters, and on the training time, the gradient descent control term, initial parameter conditions, and different training and testing sets. (authors' abstract) / Series: Discussion Papers of the Institute for Economic Geography and GIScience

Cette thèse est consacrée au démixage en imagerie hyperspectrale en particulier dans le cas où des objets de petite taille sont présents dans la scène. Les images hyperspectrales contiennent une grande quantité d’information à la fois spectrale et spatiale, et chaque pixel peut être vu comme le spectre de réflexion de la zone imagée. Du fait de la faible résolution spatiale des capteurs le spectre de réflexion observé au niveau de chaque pixel est un mélange des spectres de réflexion de l’ensemble des composants imagés dans le pixel. Une problématique de ces images hyperspectrales est le démixage, qui consiste à décomposer l’image en une liste de spectres sources, appelés endmembers, correspondants aux spectres de réflexions des composants de la scène d’une part, et d’autre part la proportion de chacun de ces spectres source dans chaque pixel de l’image. De nombreuses méthodes de démixage existent mais leur efficacité reste amoindrie en présence de spectres sources dits rares (c’est-à-dire des spectres présents dans très peu de pixels, et souvent à un niveau subpixelique). Ces spectres rares correspondent à des composants présents en faibles quantités dans la scène et peuvent être vus comme des anomalies dont la détection est souvent cruciale pour certaines applications.Nous présentons dans un premier temps deux méthodes de détection des pixels rares dans une image, la première basée sur un seuillage de l’erreur de reconstruction après estimation des endmembers abondants, la seconde basée sur les coefficients de détails obtenus par la décomposition en ondelettes. Nous proposons ensuite une méthode de démixage adaptée au cas où une partie des endmembers sont connus a priori et montrons que cette méthode utilisée avec les méthodes de détection proposées permet le démixage des endmembers des pixels rares. Enfin nous étudions une méthode de rééchantillonnage basée sur la méthode du bootstrap pour amplifier le rôle de ces pixels rares et proposer des méthodes de démixage en présence d’objets de petite taille. / This thesis is devoted to the unmixing issue in hyperspectral images, especiallyin presence of small sized objects. Hyperspectral images contains an importantamount of both spectral and spatial information. Each pixel of the image canbe assimilated to the reflection spectra of the imaged scene. Due to sensors’ lowspatial resolution, the observed spectra are a mixture of the reflection spectraof the different materials present in the pixel. The unmixing issue consists inestimating those materials’ spectra, called endmembers, and their correspondingabundances in each pixel. Numerous unmixing methods have been proposed butthey fail when an endmembers is rare (that is to say an endmember present inonly a few of the pixels). We call rare pixels, pixels containing those endmembers.The presence of those rare endmembers can be seen as anomalies that we want todetect and unmix. In a first time, we present two detection methods to retrievethis anomalies. The first one use a thresholding criterion on the reconstructionerror from estimated dominant endmembers. The second one, is based on wavelettransform. Then we propose an unmixing method adapted when some endmembersare known a priori. This method is then used with the presented detectionmethod to propose an algorithm to unmix the rare pixels’ endmembers. Finally,we study the application of bootstrap resampling method to artificially upsamplerare pixels and propose unmixing methods in presence of small sized targets.

O sucesso do tratamento de implantes dentários depende da quantidade e qualidade óssea da área de interesse e o uso da Tomografia Computadorizada por Feixe Cônico (TCFC) pode ser correlacionado para a sua estimativa. O objetivo deste estudo foi avaliar e comparar os níveis de escala de cinza dos voxels de imagens obtidas por TCFC de mandíbulas atróficas com o auxílio de programa para diagnóstico e planejamento de implantes e correlacionar a Intensidade Pixel (IP) com a classificação subjetiva do Índice de Qualidade Óssea (IQO). Foram avaliadas imagens tomográficas de 10 mandíbulas atróficas obtidas do banco de dados do LAPI-FOUSP (Laboratório para Análise e Processamento de Imagens) que receberam 8 marcações em regiões normalmente utilizadas para colocação de implantes (4 em região anterior e 4 em região posterior), sobre estas foram posicionados fios ortodônticos para padronização dos cortes tomográficos e simular posições ideais de implantes. Cortes transaxiais das áreas demarcadas foram obtidas para visualização e análise de 4 regiões corticais e 3 regiões trabeculares do tecido ósseo de cada corte, para a seleção de área foi utilizada a ferramenta Região de Interesse (ROI) de cada programa. As imagens foram analisadas por 3 programas de planejamento Romexis (Planmeca,Finlandia), XoranCAT® (Xoran Technologies, EUA) e OsiriX® (Free Software Foundation, 32 bits, EUA). Os dados foram submetidos a análise estatística de Correlação de Spearman em um nível de significância de 5% (p<0,05), análise de correlação intraclasse intra-observador e inter-observador pelo teste Kappa ponderado e o teste qui-quadrado. Os resultados apresentaram que os programas Romexis e OsiriX apresentaram melhor acuidade para a avaliação subjetiva de qualidade óssea, em contrapartida o programa Xoran apresentou inconsistência de dados. Concluiu-se que há correlação entre Intensidade de Pixel e Qualidade óssea de acordo com os resultados encontrados para os programas OsiriX e Romexis / The success of dental implants treatment depends on the bone quantity and quality of the area of interest and the use of Cone Beam Computed Tomography (CBCT) can be correlated to an estimative. The aim of this study was to evaluate and compare the grey scale levels of voxels obtained by CBCT of atrophic mandibles with the aid of software for diagnosis and planning implants and correlates the Pixel Intensity (PI) with the subjective classification of Bone Quality Index (BQI). Tomographic images of 10 atrofic mandibles obtained from the data base of LAPI-FOUSP (Laboratory of analysis and imaging processing) were evaluated, received 8 marks in areas normally used for implant placement (4 in anterior and 4 in posterior region) on these orthodontic wires were positioned for standardization of tomographic images and simulate optimal implant positions. Transaxial slices of the demarcated areas were made for visualization and analysis of four regions of interest for cancellous bone and three regions of interest for cortical bone and for selection of each area a tool called Region of Interest (ROI) was used by each software. The images were analised by three diferent planning software, Romexis (Planmeca, Finland), XoranCAT® (Xoran Technologies, EUA) e OsiriX® (Free Software Foundation, 32 bits, EUA). Data were subjected to statistical analysis of Spearman correlation, intraclass intra-observer and inter-observer analysis correlation and weighted Kappa concordance coefficient with 5% significance level and chi-squared test; The results presented that Romexis and OsiriX software showed better accuracy for subjective assessment of bone quality, on the other hand Xoran software showed data inconsistency. This study concluded that a correlation was found between Pixel Intensity (PI) and Bone Quality (BQ) due to the results of OsiriX and Romexis software.

ATLAS est l'une des deux principales expériences du LHC dans le but d'étudier les propriétés microscopiques de la matière afin de répondre aux questions les plus fondamentales de la physique des particules. Après les réalisations accomplies lors de la première prise de données, le potentiel de nouvelles découvertes et de mesures précises au LHC est étendu en repoussant les limites en matière d’énergie dans le centre de masse et de luminosité grâce à trois mises à niveau de l’accélérateur aboutissant au LHC à haute luminosité (HL-LHC). Pour tirer pleinement parti de l'augmentation de la luminosité, deux mises à niveau principales du détecteur interne ATLAS sont prévues. La première mise à niveau était déjà achevée au début de l'année 2015 avec l'insertion de l'IBL, une quatrième couche de pixels située à seulement 3,2 cm de la ligne de faisceau. Dans la deuxième mise à niveau majeure, prévue pour 2024, le détecteur interne complet sera remplacé par un tout nouveau dispositif de suivi interne entièrement constitué de dispositifs en silicium pour faire face à la forte densité de particules et à l’environnement de rayonnement intense du HL-LHC, qui pendant son fonctionnement période fournira 3000 fb-1, près de dix fois la luminosité intégrée du programme complet du LHC. Cette thèse aborde l’étude de nouveaux détecteurs de pixels de bord actifs n + -in-p en développant deux nouvelles méthodes d'analyse du profil du dopage pour étudier les effets des dommages d’irradiation sur les performances des détecteurs de pixels. Ces méthodes sont la méthode d'imagerie 3D sims et la méthode TLM. La simulation TCAD a été utilisée pour simuler les profils de dopage, le comportement électrique et les dommages dus au rayonnement. La validation des modèles de simulation avec les données a été effectuée. De plus, la caractérisation de la salle blanche ainsi que la mesure sur un faisceau de test ont été effectuées pour tester les différentes conceptions de détecteurs. Dans la deuxième partie de la thèse, je discute de l'observation de la désintégration du boson de Higgs en une paire de quarks b à l’aide des données collectées par ATLAS lors du Run 2 du LHC à une énergie de 13 TeV dans le centre de masse et une luminosité intégrée de 79.8 $fb^{-1}$. J'ai contribué à l'analyse où le boson de Higgs est produit en association avec un boson de jauge W ou Z. L'analyse VH(bb) ne considérant pas les leptons tau, j'ai réalisé une étude estimant l'impact de leur utilisation sur l'analyse. De plus, pour l’analyse VH (bb), j’ai travaillé sur l’estimation de fond multi-jets dans le canal à 1 lepton en utilisant la méthode d’analyse dijet-masse. / ATLAS is one of the two main experiments at LHC with the purpose of investigating the microscopic properties of matter to address the most fundamental questions of particle physics. After the achievements of the first years of running, the potential reach for new discoveries and precise measurements at LHC is being extended by pushing further the energy and luminosity frontiers through three upgrades of the accelerator culminating in the High Luminosity LHC (HL-LHC). To fully profit from the increased luminosity, two main upgrades of the ATLAS inner detector are planned. The first upgrade was already completed at the beginning of 2015 with the insertion of the IBL, a fourth pixel layer located at just 3.2 cm from the beam line. In the second major upgrade, foreseen for 2024, the full inner detector will be replaced by a completely new inner tracker fully made of silicon devices to cope with the high particle density and the harsh radiation environment at the HL-LHC, which during its operational period will deliver 3000 fb-1, almost ten times the integrated luminosity of the full LHC program. This thesis addresses the study of new n+-in-p active edge pixel detectors by developing two novel doping profile analysis methods to study the radiation damage effects on the pixel detectors performance. These methods are the 3D sims imaging method and the TLM Method. TCAD simulation has been used to simulate the doping profiles, the electrical behavior and the radiation damage. Validating the simulation models with data have been done. Moreover, clean-room characterization, as well as testbeam measurement have been performed to test the different detector designs. In the second part of the thesis, I discuss the observation of the standard model Higgs boson bb decay mode using the data collected by ATLAS during the LHC Run2 at center-of-mass energy 13 TeV and an integrated luminosity 79.8 fb-1 of a proton-proton collision. I contributed specifically to the search of the standard model Higgs boson in VH(bb) production mode. In the VH(bb) analysis we don't have any channel that considers the tau leptons in the final state. I have performed a feasibility study to verify the gain of using the taus in the analysis. In addition, for the VH(bb) analysis I have worked on the multi-jet background estimation in the 1-lepton channel using the dijet-mass analysis method.

Image sensors play a vital role in many image sensing and capture applications. Among the various types of image sensors, complementary metal oxide semiconductor (CMOS) based active pixel sensors (APS), which are characterized by reduced pixel size, give fast readouts and reduced noise. APS are used in many applications such as mobile cameras, digital cameras, Webcams, and many consumer, commercial and scientific applications. With these developments and applications, CMOS APS designs are challenging the old and mature technology of charged couple device (CCD) sensors. With the continuous improvements of APS architecture, pixel designs, along with the development of nanometer CMOS fabrications technologies, APS are optimized for optical sensing. In addition, APS offers very low-power and low-voltage operations and is suitable for monolithic integration, thus allowing manufacturers to integrate more functionality on the array and building low-cost camera-on-a-chip. In this thesis, I explore the current state-of-the-art of CMOS APS by examining various types of APS. I show design and simulation results of one of the most commonly used APS in consumer applications, i.e. photodiode based APS. We also present an approach for technology scaling of the devices in photodiode APS to present CMOS technologies. Finally, I present the most modern CMOS APS technologies by reviewing different design models. The design of the photodiode APS is implemented using commercial CAD tools.

No description available.

Hybrid pixel detectors are widely used in many fields, including military, environment, industry and medical treatment. When integrating such a detector, a vertical connection technique called flip-chip bonding is almost the only way to realize the high-density interconnection between each pixel detector to the read-out chip. Such bonding can offer high-density I/O and a short interconnect distance, which can make the resulting device show excellent performance. Electro deposition is a promising approach to enable a low cost and high yield bump bonding process, compared with conventional sputtering or evaporation which is currently utilized for small-scale production. Due to that, Indium bumping process using electroplating is selected, as a result of which indium bump arrays with a pitch of 220 μm and a diameter of 30 μm have been fabricated using a standard silicon wafer processing. UBM (under bump metallization) for indium bumping was Ti/Ni (300 Å/ 2000 Å). It helps to increase adhesion between the wafer and the bumps and also serves as an excellent diffusion barrier both at room temperature and at 200°C. The indium is electroplated, using an indium sulfamate plating bath, and then formed into bumps through a reflow process. The reflow is made on a 200°C hot plate with a continuous flow of nitrogen over the wafer. During the reflow the indium is melted and forms into bumps due to surface tension. All the corresponding procedural processing steps and results are incorporated in this paper.