In this thesis, research by Crum and Mao [J. Acoust. Soc. Am. 99, 2898--2907 (1996)] and Houser, Howard, and Ridgway [J. Theor. Biol. 213, 183--195 (2001)] is extended by numerically investigating bubble growth (initial radius of 10 microns) during rectified diffusion for gas supersaturations up to 300% by using the Fyrillas-Szeri equation [J. Fluid Mech. 277, 381--407 (1994)]. Bubble growth is simulated for a range of frequencies (100 Hz to 10 kHz), sound pressure levels (205 dB to 215 dB re 1 micropascal), and gas supersaturations (150% to 300%). Simulations are presented for continuous and monofrequency excitation, repeated tone bursts, and pulsed frequency-modulated waveforms. The potential for bubble growth to occur in marine mammals is also considered. For the parameters considered, static diffusion becomes the dominant growth mechanism as supersaturation is increased, bubble growth is frequency independent away from bubble resonance, and bubble growth due to duty-cycle excitation can be modeled as an effective continuous source with a reduced sound pressure level. A more accurate model of in vivo marine mammal tissue is required to determine if rectified diffusion can trigger bubble growth at the levels predicted in this thesis. / text

À partir de fonds d’archives, il s’est agi tout d’abord de rechercher, de transcrire et d’éditer la correspondance maçonnique entre Jean-Baptiste Willermoz, fondateur du Régime Écossais Rectifié à Lyon et Claude-François Achard, Vénérable Maître de la Triple Union de Marseille. On s’est intéressé aux échanges d’idées entre Willermoz et Achard et au discours du maçon lyonnais. Le but a été de mettre en évidence les principaux axes de la correspondance et notamment les thèmes ésotériques. / The scope of this work has been to complete and transcribe, then edit the masonic correspondence, from various archives collections, between Jean-Baptiste Willermoz, founder of the Rectified Scottish Order, and Claude-François Achard, Master of the “Triple Union de Marseille” Lodge. A particular attention has been given to their exchange of ideas, while focusing on JB Willermoz’s way of reasoning, in order to enlighten the major streams of this correspondence, among which its esoterical themes.

Number of lanes is a basic roadway attribute that is widely used in many transportation applications. Traditionally, number of lanes is collected and updated through field surveys, which is expensive especially for large coverage areas with a high volume of road segments. One alternative is through manual data extraction from high-resolution aerial images. However, this is feasible only for smaller areas. For large areas that may involve tens of thousands of aerial images and millions of road segments, an automatic extraction is a more feasible approach. This dissertation aims to improve the existing process of extracting number of lanes from aerial images automatically by making improvements in three specific areas: (1) performance of lane model, (2) automatic acquisition of external knowledge, and (3) automatic lane location identification and reliability estimation. In this dissertation, a framework was developed to automatically recognize and extract number of lanes from geo-rectified aerial images. In order to address the external knowledge acquisition problem in this framework, a mapping technique was developed to automatically estimate the approximate pixel locations of road segments and the travel direction of the target roads in aerial images. A lane model was developed based on the typical appearance features of travel lanes in color aerial images. It provides more resistance to “noise” such as presence of vehicle occlusions and sidewalks. Multi-class classification test results based on the K-nearest neighbor, logistic regression, and Support Vector Machine (SVM) classification algorithms showed that the new model provides a high level of prediction accuracy. Two optimization algorithms based on fixed and flexible lane widths, respectively, were then developed to extract number of lanes from the lane model output. The flexible lane-width approach was recommended because it solved the problems of error-tolerant pixel mapping and reliability estimation. The approach was tested using a lane model with two SVM classifiers, i.e., the Polynomial kernel and the Radial Basis Function (RBF) kernel. The results showed that the framework yielded good performance in a general test scenario with mixed types of road segments and another test scenario with heavy plant occlusions.

In this thesis / analysis, design and implementation of a DC-DC converter with active clamp forward topology is presented. The main objective of this thesis is generating a rectified sinusoidal voltage at the output of the converter. This is accomplished by changing the reference signal of the converter. The converter output is applied to an inverter circuit in order to obtain sinusoidal waveform. The zero crossing points of the converter is detected and the inverter drive signals are generated in order to obtain sinusoidal waveform from the output of the converter. Next, the operation of the DC-DC converter and sinusoidal output inverter coupled performance is investigated with resistive and inductive loads to find out how the proposed topology performs. The design is implemented with an experimental set-up and steady state and dynamic performance of the designed power supply is tested. Finally an evaluation of how better performance can be obtained from this kind of arrangement to obtain a sinusoidal output inverted is thoroughly discussed

熱影像可獲取不同於可見光與近紅外光的溫度資訊，可運用於監測地表火山及斷層帶的溫度或災害防治上。以往於空載或衛載上的熱感測器解析度皆較低，判釋熱影像受到限制；如今，低成本、高機動性的無人飛行載具發展趨於成熟，可搭載熱感測器，並近空垂直拍攝近景熱影像，得到較高空間解析度之熱影像。 然而，熱影像上之地物內容與邊緣較一般可見光影像模糊，若要將熱影像應用於地理空間資訊系統上時，為使熱影像可與其他地面坐標資料結合，勢必需先幾何改正熱影像，並以相同區域之數值地表模型，正射化熱影像，同時三維展示熱影像與地表模型，提供研究者地形與熱分佈資訊；此外，對於火山地帶來說，高程資料也常是研究者判釋分析的重點資訊，此做法可看出區域之溫度分佈。 為正射糾正熱影像，利用共線式執行空中三角測量平差，本研究不僅率定熱像儀，求其內方外元素，更以空中三角測量平差，計算熱影像之外方位元素。此外，因熱影像紀錄地表輻射資訊，與可見光資訊大不相同，故熱影像經共線式空中三角測量平差後，建製之數值地表模型 (Digital Surface Model, DSM)，並非該拍攝地區之真實地表起伏模型，因此本研究利用一既有的DSM，正射糾正空中三角測量後之熱影像，並以誤差向量圖表示正射糾正之成果。 / Usually, thermal images contain abundant temperature information which can often be used to monitor the surface temperature or volcanic disaster prevention. Previously, thermal images acquired by satellite platform have low resolution. Today, low-cost, highly maneuverable unmanned aerial vehicle (UAV) can carry thermal sensors and obtain close-range thermal images with high spatial resolution. Due to the distortion of thermal sensor, geometric correction should be applied to the thermal images. In this study, a UAV-borne thermal sensor has been calibrated, and used for taking thermal images. The exterior orientation elements of the thermal images have been determined by using aerial triangulation. A digital surface model generated by LiDAR was then used to ortho-rectify the thermal images. Gray values of the rectified thermal images were also normalized for generating a thermal mosaic. The resultant rectified thermal mosaic has excellent appearance for showing the temperature distribution and elevation simultaneously.

In variable speed motor drive and uninterruptible power supply (UPS) applications, taditional method is to employ some kind of a modulation technique at a high frequency typically 6 kHz to 20 kHz range. In these modulation techniques, the switches are hard switched. The result is application of a series of pulses to the load, and if the load is inductive, sine wave current flows into the load. Hard and rapid switching causes a voltage waveform with a very high dv/dt (rate of change in voltage) causing high EMI problems, reduced life expectancy of the motor and additional losses. So a power supply generating pure sinusoidal voltage waveform is very desirable. In industry some low pass filters called sinusoidal filters, are used at the output of the inverters but this comes with additional cost and bulky filter elements. In this study, a novel approach for generating power level sinusoidal waveforms is proposed. The basic structure is a DC-DC converter that produces a rectified DC-link at its output and an H-bridge inverter that inverts the rectified sinusoids to form a sinusoidal voltage. Main advantages of the circuit are that the H-bridge inverter switches have no switching stresses, they are switched at low frequency so the reliability is increased. Throughout the study different circuit topologies have been investigated and the analysis of the chosen topologies is supported with computer simulations. The system is then set up in the laboratory. In order to prove of the concept, only a single phase inverter has been investigated at steady state conditions. Efficiency, distortion level, magnitude error and device stresses have been obtained. The results indicate that the proposed configuration is very promising.

En apprentissage automatique, domaine qui consiste à utiliser des données pour apprendre une solution aux problèmes que nous voulons confier à la machine, le modèle des Réseaux de Neurones Artificiels (ANN) est un outil précieux. Il a été inventé voilà maintenant près de soixante ans, et pourtant, il est encore de nos jours le sujet d'une recherche active. Récemment, avec l'apprentissage profond, il a en effet permis d'améliorer l'état de l'art dans de nombreux champs d'applications comme la vision par ordinateur, le traitement de la parole et le traitement des langues naturelles. La quantité toujours grandissante de données disponibles et les améliorations du matériel informatique ont permis de faciliter l'apprentissage de modèles à haute capacité comme les ANNs profonds. Cependant, des difficultés inhérentes à l'entraînement de tels modèles, comme les minima locaux, ont encore un impact important. L'apprentissage profond vise donc à trouver des solutions, en régularisant ou en facilitant l'optimisation. Le pré-entraînnement non-supervisé, ou la technique du ``Dropout'', en sont des exemples. Les deux premiers travaux présentés dans cette thèse suivent cette ligne de recherche. Le premier étudie les problèmes de gradients diminuants/explosants dans les architectures profondes. Il montre que des choix simples, comme la fonction d'activation ou l'initialisation des poids du réseaux, ont une grande influence. Nous proposons l'initialisation normalisée pour faciliter l'apprentissage. Le second se focalise sur le choix de la fonction d'activation et présente le rectifieur, ou unité rectificatrice linéaire. Cette étude a été la première à mettre l'accent sur les fonctions d'activations linéaires par morceaux pour les réseaux de neurones profonds en apprentissage supervisé. Aujourd'hui, ce type de fonction d'activation est une composante essentielle des réseaux de neurones profonds. Les deux derniers travaux présentés se concentrent sur les applications des ANNs en traitement des langues naturelles. Le premier aborde le sujet de l'adaptation de domaine pour l'analyse de sentiment, en utilisant des Auto-Encodeurs Débruitants. Celui-ci est encore l'état de l'art de nos jours. Le second traite de l'apprentissage de données multi-relationnelles avec un modèle à base d'énergie, pouvant être utilisé pour la tâche de désambiguation de sens. / Machine learning aims to leverage data in order for computers to solve problems of interest. Despite being invented close to sixty years ago, Artificial Neural Networks (ANN) remain an area of active research and a powerful tool. Their resurgence in the context of deep learning has led to dramatic improvements in various domains from computer vision and speech processing to natural language processing. The quantity of available data and the computing power are always increasing, which is desirable to train high capacity models such as deep ANNs. However, some intrinsic learning difficulties, such as local minima, remain problematic. Deep learning aims to find solutions to these problems, either by adding some regularisation or improving optimisation. Unsupervised pre-training or Dropout are examples of such solutions. The two first articles presented in this thesis follow this line of research. The first analyzes the problem of vanishing/exploding gradients in deep architectures. It shows that simple choices, like the activation function or the weights initialization, can have an important impact. We propose the normalized initialization scheme to improve learning. The second focuses on the activation function, where we propose the rectified linear unit. This work was the first to emphasise the use of linear by parts activation functions for deep supervised neural networks, which is now an essential component of such models. The last two papers show some applications of ANNs to Natural Language Processing. The first focuses on the specific subject of domain adaptation in the context of sentiment analysis, using Stacked Denoising Auto-encoders. It remains state of the art to this day. The second tackles learning with multi-relational data using an energy based model which can also be applied to the task of word-sense disambiguation.