Esta tesis explora la precisión de dos métodos de estimación de precipitación, Auto-Estimator y CRR (Convective Rainfall Rate), generados a partir de imágenes infrarrojas y visibles del Meteosat. Ambos métodos junto con una serie de correcciones de la intensidad de lluvia estimada se aplican y se verifican en dos casos de inundaciones acaecidas en zonas mediterráneas. El primer caso ocurrió en Albania del 21 al 23 de septiembre de 2002 y el segundo, conocido como caso Montserrat, ocurrió en Cataluña la noche del 9 al 10 se junio de 2000. Por otro lado se investiga la posibilidad de realizar calibraciones de ambos métodos directamente con datos de estaciones pluviométricas cuando lo común es calibrar con datos de radares meteorológicos. También se propone cambios en algunas de las correcciones ya que parecen mejorar los resultados y se propone una nueva corrección muy eficiente que utiliza las descargas eléctricas para determinar la zonas más convectivas y de mayor precipitación de los sistemas nubosos. / This Thesis work explores the precision of two methods to estimate rainfall called Auto-Estimator and CRR (Convective Rainfall Rate). They are obtained by using infrared and visible images from Meteosat. Both Algorithms within a set of correction factors are applied and verified in two severe flood cases that took place in Mediterranean regions. The first case has occurred in Albania from 21 to 23 September 2002 and the second, known as the Montserrat case, has occurred in Catalonia the night from the 9 to 10 of June 2000. On the other hand it is explored new methods to perform calibrations to both satellite algorithms using direct rain rates from rain gauges. These kinds of adjustments are usually done using rain rates from meteorological radars. In addition it is proposed changes on some correction factors that seem to improve the results on estimations and it is defined an efficient correction factor that employ electrical discharges to detect the most convective and rainy areas in cloud systems.

Cette thèse traite d'un des problèmes les plus stimulants dans le traitement de la parole concernant la prothèse auditive, où seulement un capteur est disponible avec de faibles coûts de calcul, de faible utilisation d'énergie et l'absence de bases de données. Basée sur les récents résultats dans les deux estimations statistiques paramétriques et non-paramétriques, ainsi que la représentation parcimonieuse. Cette étude propose quelques techniques non seulement pour améliorer la qualité et l'intelligibilité de la parole, mais aussi pour s'attaquer au débruitage du signal audio en général.La thèse est divisée en deux parties ; Dans la première partie, on aborde le problème d'estimation de la densité spectrale de puissance du bruit, particulièrement pour le bruit non-stationnaire. Ce problème est une des parties principales du traitement de la parole du mono-capteur. La méthode proposée prend en compte le modèle parcimonieux de la parole dans le domaine transféré. Lorsque la densité spectrale de puissance du bruit est estimée, une approche sémantique est exploitée pour tenir compte de la présence ou de l'absence de la parole dans la deuxième partie. En combinant l'estimation Bayésienne et la détection Neyman-Pearson, quelques estimateurs paramétriques sont développés et testés dans le domaine Fourier. Pour approfondir la performance et la robustesse de débruitage du signal audio, une approche semi-paramétrique est considérée. La conjointe détection et estimation peut être interprétée par Smoothed Sigmoid-Based Shrinkage (SSBS). Ainsi, la méthode Bloc-SSBS est proposée afin de prendre en compte les atomes voisinages dans le domaine temporel-fréquentiel. De plus, pour améliorer fructueusement la qualité de la parole et du signal audio, un estimateur Bayésien est aussi dérivé et combiné avec la méthode Bloc-SSBS. L'efficacité et la pertinence de la stratégie dans le domaine transformée cosinus pour les débruitages de la parole et de l'audio sont confirmées par les résultats expérimentaux. / This PhD thesis deals with one of the most challenging problem in speech enhancement for assisted listening where only one micro is available with the low computational cost, the low power usage and the lack out of the database. Based on the novel and recent results both in non-parametric and parametric statistical estimation and sparse representation, this thesis work proposes several techniques for not only improving speech quality and intelligibility and but also tackling the denoising problem of the other audio signal. In the first major part, our work addresses the problem of the noise power spectrum estimation, especially for non-stationary noise, that is the key part in the single channel speech enhancement. The proposed approach takes into account the weak-sparseness model of speech in the transformed model. Once the noise power spectrum has been estimated, a semantic road is exploited to take into consideration the presence or absence of speech in the second major part. By applying the joint of the Bayesian estimator and the Neyman-Pearson detection, some parametric estimators were developed and tested in the discrete Fourier transform domain. For further improve performance and robustness in audio denoising, a semi-parametric approach is considered. The joint detection and estimation can be interpreted by Smoothed Sigmoid-Based Shrinkage (SSBS). Thus, Block-SSBS is proposed to take into additionally account the neighborhood bins in the time-frequency domain. Moreover, in order to enhance fruitfully speech and audio, a Bayesian estimator is also derived and combined with Block-SSBS. The effectiveness and relevance of this strategy in the discrete Cosine transform for both speech and audio denoising are confirmed by experimental results.

Made available in DSpace on 2016-06-02T20:06:05Z (GMT). No. of bitstreams: 1 4180.pdf: 1039417 bytes, checksum: d09a61a4895fb47d1c2456468800fc2f (MD5) Previous issue date: 2012-02-29 / Financiadora de Estudos e Projetos / In this work we consider a new estimator proposed by Ferrari & Yang (2010), called the maximum Lq-likelihood estimator (MLqE), to estimate the parameters of the measurement error models, in particular, the structural model. The new estimator extends the classical maximum likelihood estimator (MLE) and its based on the minimization, by means of the Kullback-Leibler (KL) divergence, of the discrepancy between a distribuiton in a family and one that modifies the true distribution by the degree of distortion q. Depending on the choice of q, the transformed distribution can diminish or emphasize the role of extreme observations, unlike the ML method that equally weights each observation. For small and moderate sample sizes, the MLqE can trade bias for precision, causing a reduction of the mean square error (MSE). The structural model has the characteristic of non-identifiability. For this reason, we must make assumptions on the parameters to overcome the non-identifiability. We perform a analytical study and a simulation study to compare MLqE and MLE. To gauge performance of the estimators, we compute measures of overall performance, bias, standard deviation, standard error, MSE, probability of coverage and length of confidence intervals. / Neste trabalho utilizaremos um novo estimador proposto por Ferrari & Yang (2010), denominado de estimador de máxima Lq-verossimilhança (EMLqV), na estimação dos parâmetros de modelos com erros de medição estruturais normais. O novo estimador é uma generalização do estimador de máxima verossimilhança (EMV) usual e sua construção baseia-se na comparação, utilizando divergência de Kullback-Leibler (KL), entre duas distribuições, a distribuição inalterada e a distribuição modificada pelo grau de distorção da função de verossimilhança (q). Conforme a escolha para q, a distribuição modificada poderá atenuar ou exaltar o papel das observações extremas, diferentemente do EMV usual que atribui os mesmos pesos a todas as observações. Na comparação entre as duas distribuições pela divergência de KL é inserida certa quantidade de viés no estimador resultante, que é controlada pelo parâmetro q. O aumento do viés do estimador MLqV pode ser compensado com a redução de sua variância, pela escolha apropriada de q. O modelo estrutural possui a característica de ser inidentificável. Para torná-lo identificável faremos suposições sobre os parâmetros do modelo, analisando cinco casos de identificabilidade do modelo. A comparação entre os métodos MLqV e MV na estimação dos parâmetros do modelo será baseada em resultados analíticos e em simulações, sendo calculadas medidas de desempenho global, viés, desvio padrão (DP), erro padrão estimado (EP), erro quadrático médio (EQM), probabilidade de cobertura e amplitude dos intervalos de confiança.

Master of Science / Department of Statistics / Paul I. Nelson / Quantiles are useful in describing distributions of component lifetimes. Data, consisting of the lifetimes of sample units, used to estimate quantiles are often censored. Right censoring, the setting investigated here, occurs, for example, when some test units may still be functioning when the experiment is terminated. This study investigated and compared the performance of parametric and nonparametric estimators of quantiles from right censored data generated from Weibull and Lognormal distributions, models which are commonly used in analyzing lifetime data. Parametric quantile estimators based on these assumed models were compared via simulation to each other and to quantile estimators obtained from the nonparametric Kaplan- Meier Estimator of the survival function. Various combinations of quantiles, censoring proportion, sample size, and distributions were considered. Our simulation show that the larger the sample size and the lower the censoring rate the better the performance of the estimates of the 5th percentile of Weibull data. The lognormal data are very sensitive to the censoring rate and we observed that for higher censoring rates the incorrect parametric estimates perform the best. If you do not know the underlying distribution of the data, it is risky to use parametric estimates of quantiles close to one. A limitation in using the nonparametric estimator of large quantiles is their instability when the censoring rate is high and the largest observations are censored. Key Words: Quantiles, Right Censoring, Kaplan-Meier estimator

Thesis (MScEng)--Stellenbosch University, 2001. / ENGLISH ABSTRACT: This thesis describes a module that links AutoCAD and CeDeas (cost estimation software which was developed by Department of Mechanical Engineering, University of Stellenbosch). CeDeas is intended for estimating the direct manufacturing cost of simple welded assemblies in a batch production environment. It is aimed at use during late concept design or early detail design. The link module was developed in Borland C++ Builder. By using COM (Component Object Model) technology, the link module employs the methods and the properties of the AutoCAD automation interface to extract manufacturing information that is required by CeDeas. The link module prompts the user to pick objects in an AutoCAD drawing and then determines the values required by CeDeas to estimate the manufacturing cost. The user can choose between a "direct select method" (which uses the properties of geometric entities already in the drawing) and a "user define method" (whereby the user defmes temporary entities or combines aspects of existing entities in the AutoCAD drawing). With these results and some non-geometric inputs, the user can get a cost estimate of components and assemblies. After design changes, the link module can provide CeDeas with updated values with minimal user interaction in situations where the "direct select method" was used. The designer can therefore easily use the cost estimates to compare design alternatives to optimise the design. Validation studies demonstrated the numerical accuracy of the use of the link module. The link module can be regarded as an extension of CeDeas. At present it only supports AutoCAD R14, but can be extended to support AutoCAD 2000 and Mechanical Desktop. / AFRIKAANSE OPSOMMING: 'n Module wat dien as skakel tussen AutoCAD and CeDeas (kosteberamingsagteware ontwikkel deur die Departement van Meganiese Ingenieurswese, Universiteit van Stellenbosch) word in hierdie tesis beskryf. Die doel van CeDeas is om die direkte vervaardiginskoste van eenvoudige, gesweisde samestellings, in 'n lot-produksie omgewing, te beraam. Dit is gemik op gebruik tydens laat konsepontwerp en vroeë detailontwerp. Die skakelmodule is ontwikkel in Borland C++ Builder. Deur van COM (Component Object Model) tegnologie gebruik te maak, kry die skakelmodule toegang tot die funksies en eienskappe van AutoCAD se outomatisasie koppelvlak en kan sodoende die vervaardigingsinligting onttrek wat deur CeDeas benodig word. Die skakelmodule vra die gebruiker om voorwerpe in 'n AutoCAD tekening te kies en bepaal dan die waardes wat deur CeDeas benodig word om die vervaardigingskoste te skat. Die gebruiker kan kies tussen 'n "direkte keuse metode" (wat die eienskappe van geometriese entiteite wat reeds in die tekening is, gebruik) en 'n "gebruiker definieer metode" (waarin die gebruiker tydelike entiteite defmieer of kombinasies van aspekte van bestaande entiteite in die AutoCAD tekening gebruik). 'n Koste beraming van komponente of samestellings kan verkry word met hierdie inligting tesame met ander nie-geometriese inligting. Na ontwerpsveranderings, kan die skakelmodule hersiene waardes vir CeDeas voorsien met minimale gebruikers-interaksie in gevalle waar die "direkte keuse metode" gebruik is. Die gebruiker kan daarom maklik die kosteskattings gebruik om ontwerpsaltematiewe te vergelyk om die ontwerp te optimeer. Evalueringstudies het die numeriese akkuraatheid van die skakelmodule bevesting. Hierdie module kan as 'n uitbreiding van CeDeas beskou word. Tans werk die module slegs met AutoCAD R14, maar dit kan uitgebrei word om met AutoCAD 2000 en Mechanical Desktop te werk.

139 leaves printed single pages, preliminary pages i-xi and numbered pages 1-127. Includes bibliography and a list of figures and tables. Digitized at 600 dpi grayscale to pdf format (OCR),using a Bizhub 250 Konica Minolta Scanner. / Thesis (MScEng (Electrical and Electronic Engineering))--University of Stellenbosch, 2003. / ENGLISH ABSTRACT: Probability density functions (PDFs) and cumulative distribution functions (CDFs) play a central role in statistical pattern recognition and verification systems. They allow observations that do not occur according to deterministic rules to be quantified and modelled. An example of such observations would be the voice patterns of a person that is used as input to a biometric security device. In order to model such non-deterministic observations, a density function estimator is employed to estimate a PDF or CDF from sample data. Although numerous density function estimation techniques exist, all the techniques can be classified into one of two groups, parametric and non-parametric, each with its own characteristic advantages and disadvantages. In this research, we introduce a novel approach to density function estimation that attempts to combine some of the advantages of both the parametric and non-parametric estimators. This is done by considering density estimation using an abstract approach in which the density function is modelled entirely in terms of its moments or characteristic function. New density function estimation techniques are first developed in theory, after which a number of practical density function estimators are presented. Experiments are performed in which the performance of the new estimators are compared to two established estimators, namely the Parzen estimator and the Gaussian mixture model (GMM). The comparison is performed in terms of the accuracy, computational requirements and ease of use of the estimators and it is found that the new estimators does combine some of the advantages of the established estimators without the corresponding disadvantages. / AFRIKAANSE OPSOMMING: Waarskynlikheids digtheidsfunksies (WDFs) en Kumulatiewe distribusiefunksies (KDFs) speel 'n sentrale rol in statistiese patroonherkenning en verifikasie stelsels. Hulle maak dit moontlik om nie-deterministiese observasies te kwantifiseer en te modelleer. Die stempatrone van 'n spreker wat as intree tot 'n biometriese sekuriteits stelsel gegee word, is 'n voorbeeld van so 'n observasie. Ten einde sulke observasies te modelleer, word 'n digtheidsfunksie afskatter gebruik om die WDF of KDF vanaf data monsters af te skat. Alhoewel daar talryke digtheidsfunksie afskatters bestaan, kan almal in een van twee katagoriee geplaas word, parametries en nie-parametries, elk met hul eie kenmerkende voordele en nadele. Hierdie werk Ie 'n nuwe benadering tot digtheidsfunksie afskatting voor wat die voordele van beide die parametriese sowel as die nie-parametriese tegnieke probeer kombineer. Dit word gedoen deur digtheidsfunksie afskatting vanuit 'n abstrakte oogpunt te benader waar die digtheidsfunksie uitsluitlik in terme van sy momente en karakteristieke funksie gemodelleer word. Nuwe metodes word eers in teorie ondersoek en ontwikkel waarna praktiese tegnieke voorgele word. Hierdie afskatters het die vermoe om 'n wye verskeidenheid digtheidsfunksies af te skat en is nie net ontwerp om slegs sekere families van digtheidsfunksies optimaal voor te stel nie. Eksperimente is uitgevoer wat die werkverrigting van die nuwe tegnieke met twee gevestigde tegnieke, naamlik die Parzen afskatter en die Gaussiese mengsel model (GMM), te vergelyk. Die werkverrigting word gemeet in terme van akkuraatheid, vereiste numeriese verwerkingsvermoe en die gemak van gebruik. Daar word bevind dat die nuwe afskatters weI voordele van die gevestigde afskatters kombineer sonder die gepaardgaande nadele.

本文共一冊，約一萬餘字，分五章。 本文內容乃探討抽樣問題之關於非線性統計量之估計方法，主要方法有三：線性化法 、摺刀法，均衡半樣本法，並比較三者之變異數估計式之大小。 本文共分五章，第一章說明研究動機與目的，及本文結構。第二章說明非線性統計量 應用三種方法時所得之估計式及偏誤。第三章討論該非線性統計量之變異數估計式並 討論並偏誤大小。第四章為實例分析，以討論比率參數在三種方法下之絕對偏誤之大 小。第五章結論。

Cette thèse porte sur les modèles génératifs en apprentissage automatique. Deux nouveaux modèles basés sur les réseaux de neurones y sont proposés. Le premier modèle possède une représentation interne où une certaine structure a été imposée afin d’ordonner les caractéristiques apprises. Le deuxième modèle parvient à exploiter la structure topologique des données observées, et d’en tenir compte lors de la phase générative. Cette thèse présente également une des premières applications de l’apprentissage automatique au problème de la tractographie du cerveau. Pour ce faire, un réseau de neurones récurrent est appliqué à des données de diffusion afin d’obtenir une représentation des fibres de la matière blanche sous forme de séquences de points en trois dimensions.

Die wissenschaftliche Auseinandersetzung mit schnellwachsenden Baumarten in Form der Kurzumtriebswirtschaft beschränkte sich lange Zeit fast ausschließlich auf die Arten der Sektion Pappel. Ertragswerte bzw. waldwachstumskundliche Untersuchungen an KUP aus Weide konnten im größeren Maßstab erst im vergangenen Jahrzehnt erhoben werden. Die Ertragswerte der Weiden auf den in dieser Arbeit untersuchten KUP bewegen sich zwischen 1,7 und 9,2 tatro∙ha-1∙a-1. In Folgerotationen kann sogar mit einem wesentlich höheren Biomasseertrag gerechnet werden. Eine eigene Untersuchung an einer bereits achtjährigen KUP in Krummenhennersdorf (Sachsen) auf einem mittleren Standort erbrachte eine durchschnittliche Gesamtwuchsleistung von ca. 14 tatro∙ha-1∙a- 1 in der dritten Rotation. Damit ist die Weide neben der Pappel eine sehr ertragreiche, schnellwachsende Baumart. Ihr Anbau kann demzufolge in Form von KUP uneingeschränkt empfohlen werden. Um den Biomasseertrag auf bestehenden KUP aus Weide zu ermitteln, wurden bisher vor allem destruktive Verfahren angewendet, die mit hohem Zeit- und Arbeitsaufwand verbunden sind. Das Hauptziel dieser Arbeit stellte daher die Entwicklung eines überregional einsetzbaren, praxistauglichen Ertragsermittlers zur schnellen und einfachen Ertragsbestimmung von KUP aus Weide der ersten Rotation dar. Die Modellentwicklung erfolgte auf zwei Wegen: Modellierung über eine Standardisierung von Biomassefunktionen (Modell I) und Modellierung über Biomasseverteilungsfunktionen (Modell II). Unterschiede in der Genauigkeit im Hinblick auf die Validierung der Modelle konnten explizit anhand des Bias und der Präzision nachgewiesen werden. Dabei ermittelt das Modell II den Biomasseertrag wesentlich genauer (Bias: 4.43%, Präzision: 12.11%), weshalb die Methode bereits für den Praktiker in Form eines einfach anzuwendenden Computerprogrammes (Ertragsschätzer für KUP aus Pappel und Weide) zur Verfügung gestellt wurde. Für die Anwendung des Ertragsschätzers sind von dem Nutzer die Eingangsgrößen Durchmesser des Grundflächenmittelstammes (dg), Alter und Pflanzverband erforderlich.

(in English): The theme of this thesis project is the "Functioning of proper names in phraseological units". In the study researched phraseological units with proper names (anthroponymes) which are taken from the phraseological dictionary "Russian phraseology: historical and etymological dictionary", ed. V.M Mokienko. The subject of our study is to determine the characteristics of the functioning of proper names in phraseological units, in order to explain and show the translation of the function of proper names, to identify the role of proper names. In the thesis project were performed the following tasks: 1) identify and classify phraseological units with anthroponymes for reasons of their occurrence, and 3) explain by psycholinguistic experiment how the proper name loses its primary function - nominative, getting an evaluation function, and 4) check an acquisition of estimator function by anthroponymes on materials Russian National Corpus. In the study were analyzed 80 phraseological units with anthroponimes, revealed changes in the scope and content of the concept of a proper name in the phraseological unit. Finally, conclusions are made about acquisition the proper name in the frezeologicheskoy unit properties that can be applied to unlimited number of denotations.