The foreign exchange market is one of the largest and most active financial markets with enormous daily trading volumes. Exchange rates are influenced by the interactions of a large number of agents, each operating with different intentions and on different time scales. This gives rise to nonlinear and non-stationary behaviour which complicates modelling. This research proposes a neural network based model trained on data filtered with a novel Empirical Mode Decomposition (EMD) filtering method for the forecasting of exchange rates. One minor and two major exchange rates are evaluated in this study. Firstly the ideal prediction horizons for trading are calculated for each of the exchange rates. The data is filtered according to this ideal prediction horizon using the EMD-filter. This EMD-filter dynamically filters the data based on the apparent number of intrinsic modes in the signal that can contribute towards prediction over the selected horizon. The filter is employed to filter out high frequency noise and components that would not contribute to the prediction of the exchange rate at the chosen timescale. This results in a clearer signal that still includes nonlinear behaviour. An artificial neural network predictor is trained on the filtered data using different sampling rates that are compatible with the cut-off frequency. The neural network is able to capture the nonlinear relationships between historic and future filtered data with greater certainty compared to a neural network trained on unfiltered data. Results show that the neural network trained on EMD-filtered data is significantly more accurate at prediction of exchange rates compared to the benchmark models of a neural network trained on unfiltered data and a random walk model for all the exchange rates. The EMD-filtered neural network’s predicted returns for the higher sample rates show higher correlations with the actual returns, and significant profits can be made when applying a trading strategy based on the predictions. Lower sample rates that just marginally satisfy the Nyquist criterion perform comparably with the neural network trained on unfiltered data; this may indicate that some aliasing occurs for these sampling rates as the EMD low-pass filter has a gradual cut-off, leaving some high frequency noise within the signal. The proposed model of the neural network trained on EMD-filtered data was able to uncover systematic relationships between the filtered inputs and actual outputs. The model is able to deliver profitable average monthly returns for most of the tested sampling rates and forecast horizons of the different exchange rates. This provides evidence that systematic predictable behaviour is present within exchange rates, and that this systematic behaviour can be modelled if it is properly separated from high frequency noise. / MIng (Computer and Electronic Engineering), North-West University, Potchefstroom Campus, 2015

Weighted graph matching is a good way to align a pair of shapes represented by a set of descriptive local features; the set of correspondences produced by the minimum cost of matching features from one shape to the features of the other often reveals how similar the two shapes are. However, due to the complexity of computing the exact minimum cost matching, previous algorithms could only run efficiently when using a limited number of features per shape, and could not scale to perform retrievals from large databases. We present a contour matching algorithm that quickly computes the minimum weight matching between sets of descriptive local features using a recently introduced low-distortion embedding of the Earth Mover's Distance (EMD) into a normed space. Given a novel embedded contour, the nearest neighbors in a database of embedded contours are retrieved in sublinear time via approximate nearest neighbors search. We demonstrate our shape matching method on databases of 10,000 images of human figures and 60,000 images of handwritten digits.

Time-frequency analysis for non-linear and non-stationary signals is extraordinarily challenging. To capture the changes in these types of signals, it is necessary for the analysis methods to be local, adaptive and stable. In recent years, decomposition based analysis methods were developed by different researchers to deal with non-linear and non-stationary signals. These methods share the feature that a signal is decomposed into finite number of components on which the time-frequency analysis can be applied. Differences lie in the strategies to extract these components: by iteration or by optimization. However, considering the requirements of being local, adaptive and stable, neither of these decompositions are perfectly satisfactory. Motivated to find a local, adaptive and stable decomposition of a signal, this thesis presents Adaptive Local Iterative Filtering (ALIF) algorithm. The adaptivity is obtained having the filter lengths being determined by the signal itself. The locality is ensured by the filter we designed based on a PDE model. The stability of this algorithm is shown and the convergence is proved. Moreover, we also propose a local definition for the instantaneous frequency in order to achieve a completely local analysis for non-linear and non-stationary signals. Examples show that this decomposition really helps in both simulated data analysis and real world application.

Weighted graph matching is a good way to align a pair of shapesrepresented by a set of descriptive local features; the set ofcorrespondences produced by the minimum cost of matching features fromone shape to the features of the other often reveals how similar thetwo shapes are. However, due to the complexity of computing the exactminimum cost matching, previous algorithms could only run efficientlywhen using a limited number of features per shape, and could not scaleto perform retrievals from large databases. We present a contourmatching algorithm that quickly computes the minimum weight matchingbetween sets of descriptive local features using a recently introducedlow-distortion embedding of the Earth Mover's Distance (EMD) into anormed space. Given a novel embedded contour, the nearest neighborsin a database of embedded contours are retrieved in sublinear time viaapproximate nearest neighbors search. We demonstrate our shapematching method on databases of 10,000 images of human figures and60,000 images of handwritten digits.

When it comes to monitoring human health, accuracy is not a choice. Accuracy in blood pressure (BP) estimation is essential for proper diagnosis and management of hypertension. An error of 5 mmHg is so serious, it can be responsible for doubling or halving number of patients diagnosed with hypertension. Motion artifacts are external sources of inaccuracy and can be due to sudden arm motion, muscle tremor, shivering, and transport vehicle vibration. Medium term drift, due to changing environmental factors, such as ambient temperature, can also contribute to the inaccuracy. Long term drift (ageing), can reach 9 mmHg during the first three months of usage. In this thesis, a new stage is added to current cuff based BP devices. This stage is responsible for adjusting the pressure reading before displaying it to end users. The proposed stage is provided with a 3-axis accelerometer, which makes the detection of motion artifacts during measurement possible. Moreover, it monitors changes in the ambient temperature and sensor ageing, so that it will adaptively compensate for these inaccuracies. These sources of inaccuracy are suppressed using algorithms based on Empirical Mode Decomposition (EMD), which has the feature of removing unwanted noise components little effect on the phase or the frequency distribution of the measured signal. With motion artifacts, measurements show that the proposed algorithms considerably improved the accuracy of the blood pressure estimates in comparison with the commonly-used conventional oscillometric algorithm that does not include a stage for artifact suppression, and allowed the estimates to consistent with the international ANSI/AAMI/ISO standard. Moreover, simulations based on experimental results show that the system is able to compensate for drift due to temperature changes and ageing with excellent performance. Results show promise towards building a robust BP monitor, with very low errors due to motion artifacts, environmental changes, and ageing.

Cette étude introduit une nouvelle approche appelée Bivariate 2D-EMD pour séparer le mouvement organisé à grande échelle, soit la composante basse fréquence de l’écoulement des fluctuations turbulentes, soit la composante haute fréquence dans un champ de vitesse instantané bidimensionnel.Cette séparation nécessite un seul champ de vitesse instantané contrairement aux autres méthodes plus couramment utilisées en mécanique des fluides, comme le POD. La méthode proposée durant cette thèse est tout à fait appropriée à l’analyse des écoulements qui sont intrinsèquement instationnaires et non linéaires comme l'écoulement dans le cylindre lorsque le moteur fonctionne dans des conditions transitoires. Bivariate 2D-EMD est validé à travers différents cas test, sur un écoulement turbulent homogène et isotrope (THI) expérimental, qui a été perturbé par un tourbillon synthétique de type Lamb-Ossen, qui simule le mouvement organisé dans le cylindre. Enfin, Il est appliqué sur un écoulement expérimental obtenu dans un cylindre et les résultats de la séparation d'écoulement sont comparés à ceux basés sur l'analyse POD. L’évolution d’écoulement dans le cylindre pendant le fonctionnement du moteur transitoire, c’est à dire une accélération du régime moteur de 1000 à 2000tr/min en différentes rampes, sont mesurée en utilisant de PIV 2D-2C haute cadence. Les champs de vitesse sont obtenus dans le plan de tumble dans un moteur un moteur GDI mono-cylindre transparent et forment une base de données nécessaire pour valider les résultats de simulation numérique. / This study introduces a new approach called Bivariate 2D-EMD to separate large-scale organizedmotion i.e., flow low frequency component from random turbulent fluctuations i.e., high frequency onein a given in-cylinder instantaneous 2D velocity field. This signal processing method needs only oneinstantaneous velocity field contrary to the other methods commonly used in fluid mechanics, as POD.The proposed method is quite appropriate to analyze the flows intrinsically both unsteady and nonlinearflows as in in-cylinder. The Bivariate 2D-EMD is validated through different test cases, by optimize itand apply it on an experimental homogeneous and isotropic turbulent flow (HIT), perturbed by asynthetic Lamb-Ossen vortex, to simulate the feature of in-cylinder flows. Furthermore, it applies onexperimental in-cylinder flows. The results obtained by EMD and POD analysis are compared. Theevolution of in-cylinder flow during transient engine working mode, i.e., engine speed acceleration from1000 to 2000 rpm with different time periods, was obtained by High speed PIV 2D-2C. The velocityfields are obtained within tumble plane in a transparent mono-cylinder DISI engine and provide a database to validate CFD.

Les superpositions d'ondes modulées en amplitude et en fréquence (modes AM--FM) sont couramment utilisées pour modéliser de nombreux signaux du monde réel : cela inclut des signaux audio (musique, parole), médicaux (ECG), ou diverses séries temporelles (températures, consommation électrique). L'objectif de ce travail est l'analyse et la compréhension fine de tels signaux, dits "multicomposantes" car ils contiennent plusieurs modes. Les méthodes mises en oeuvre vont permettre de les représenter efficacement, d'identifier les différents modes puis de les démoduler (c'est-à-dire déterminer leur amplitude et fréquence instantanée), et enfin de les reconstruire. On se place pour cela dans le cadre bien établi de l'analyse temps-fréquence (avec la transformée de Fourier à court terme) ou temps-échelle (transformée en ondelettes continue). On s'intéressera également à une méthode plus algorithmique et moins fondée mathématiquement, basée sur la notion de symétrie des enveloppes des modes : la décomposition modale empirique. La première contribution de la thèse propose une alternative au processus dit ``de tamisage'' dans la décomposition modale empirique, dont la convergence et la stabilité ne sont pas garanties. \`A la place, une étape d'optimisation sous contraintes ainsi qu'une meilleure détection des extrema locaux du mode haute fréquence garantissent l'existence mathématique du mode, tout en donnant de bons résultats empiriques. La deuxième contribution concerne l'analyse des signaux multicomposantes par la transformée de Fourier à court terme et à la transformée en ondelettes continues, en exploitant leur structure particulière ``en ridge'' dans le plan temps-fréquence. Plus précisément, nous proposons une nouvelle méthode de reconstruction des modes par intégration locale, adaptée à la modulation fréquentielle, avec des garanties théoriques. Cette technique donne lieu à une nouvelle méthode de débruitage des signaux multicomposantes. La troisième contribution concerne l'amélioration de la qualité de la représentation au moyen de la ``réallocation'' et du ``synchrosqueezing''. Nous prolongeons le synchrosqueezing à la transformée de Fourier à court terme, et en proposons deux extensions inversibles et adaptées à des modulations fréquentielles importantes, que nous comparons aux méthodes originelles. Une généralisation du synchrosqueezing à la dimension 2 est enfin proposée, qui utilise le cadre de la transformée en ondelettes monogène. / Many signals from the physical world can be modeled accurately as a superposition of amplitude- and frequency-modulated waves. This includes audio signals (speech, music), medical data (ECG) as well as temporal series (temperature or electric consumption). This thesis deals with the analysis of such signals, called multicomponent because they contain several modes. The techniques involved allow for the detection of the different modes, their demodulation (ie, determination of their instantaneous amplitude and frequency) and reconstruction. The thesis uses the well-known framework of time-frequency and time-scale analysis through the use of the short-time Fourier and the continuous wavelet transforms. We will also consider a more recent algorithmic method based on the symmetry of the enveloppes : the empirical mode decomposition. The first contribution proposes a new way to avoid the iterative ``Sifting Process'' in the empirical mode decomposition, whose convergence and stability are not guaranteed. Instead, one uses a constrained optimization step together with an enhanced detection of the local extrema of the high-frequency mode. The second contribution analyses multicomponent signals through the short-time Fourier transform and the continuous wavelet transform, taking advantage of the ``ridge'' structure of such signals in the time-frequency or time-scale planes. More precisely, we propose a new reconstruction method based on local integration, adapted to the local frequency modulation. Some theoretical guarantees for this reconstruction are provided, as well as an application to multicomponent signal denoising. The third contribution deals with the quality of the time-frequency representation, using the reassignment method and the synchrosqueezing transform: we propose two extensions of the synchrosqueezing, that enable mode reconstruction while remaining efficient for strongly modulated waves. A generalization of the synchrosqueezing in dimension 2 is also proposed, based on the so-called monogenic wavelet transform.

Nan Wan is a half-closed embayment in the most southern part of Taiwan. While facing the Luzon Strait, it also connects to the Pacific Ocean in its southeast, and is adjacent the Taiwan Strait and the South China Sea . In view of general oceanic circulation, Nan Wan Bay happens to lie to the rim of South China Sea circumfluence and Kuroshio where a variety of water mass exchange has taken place, causing saline intrusion and mixed of water. Seasonal variation and tidal fluctuations also contribute to the exchange of water masses. The Third Nuclear Power Station of Taiwan Power Company is located in Nan Wan with its thermal discharge outlet adjacent to Maobitou to the west of the bay in order to minimize the effect of warm water discharge on the local marine ecology and coral . A long-term monitoring program on water temperature and other environmental factors has been set up implemented .this research report will first describe the archives regarding the hydrology in Nan Wan in support of monitoring the process in temperature variation . Previous research efforts are found somehow unable reveal precisely the physical mechanism leading to water temperature variations in the bay, due to limited facilities, short of information or poor analytical tools. This report adopts 14 records of water temperature at the thermal outlet of the Third Nuclear Power Station for signal analysis. As to non-linear and unstable data analysis, it is based on the Hilbert-Huang Transform. HHT includes Empirical Mode Decomposition, EMD which could decompose the raw data into numerous Intrinsic Mode Function, IMF. It is allowed to comprehend the main causes for the rising and dropping of water temperature based on the variation of spectroscopy by transferring through Hilbert and analyzing via IMF. Furthermore, the characteristic of each quantity could be developed according to the quantities acquired from the former method of HHT. The analytical report of water temperature covers 14 records dating from 1999 to 2003. In light of the analytical report, tide and wind account for the main cause of the temperature variation in waters while demanding information to ensure whether it is influenced by other factors like internal waves, water masses or landforms, etc. In addition, the report compares the difference in the same of data between FFT and HHT and moreover concludes the advantages and disadvantages as reference for researches.

Flies are highly visually guided animals. In this thesis, I have used hoverflies as a model for studying motion vision. Flies process motion vision in three visual ganglia: the lamina, the medulla, and the lobula complex. In the posterior part of lobula complex, there are around 60 lobula plate tangential cells (LPTCs). Most of LPTCs have large receptive fields where the local direction sensitivity suggests that they function as matched filters to specific types of optic flow. LPTCs connect to descending or neck motor neurons that control wing and head movements, respectively. Therefore, in this thesis I have focused on the electrophysiological responses of LPTCs to gain understanding of visual behaviors in flies. The elementary motion detector (EMD) is a model that can explain the formation of local motion sensitivity. However, responses to higher order motion, where the direction of luminance change is uncorrelated with the direction of movement, cannot be predicted by classic EMDs. Nevertheless, behavior shows that flies can see and track bars with higher order motion cues. I showed (Paper I) that several LPTCs also respond to higher order motion. Many insects, including flies, release octopamine during flight. Therefore, adding octopamine receptor agonists can mimic physical activity. Our study (Paper II) investigated the effect of octopamine on three adaptation components. We found that the contrast gain reduction showed a frequency dependent increase after octopamine stimulation. Since the contrast gain is non-directional, it is likely presynaptic to the LPTC. We therefore believe that octopamine acts on the delay filter in the EMD. In the third paper we describe a novel LPTC. The centrifugal stationary inhibited flicker excited (cSIFE) is excited by flicker and inhibited by stationary patterns. Neither of these responses can be predicted by EMD models. Therefore, we provide a new type of motion detector that can explain cSIFE’s responses (Paper III). During bar tracking, self-generated optic flow may counteract the steering effect by inducing a contradictory optomotor response. Behavior shows that during bar fixation, flies ignore background optic flow. Our study (Paper IV) focus on the different receptive fields of two LPTCs, and relate these to the bar fixation behavior. In the neuron with a small and fronto-dorsal receptive field, we find a higher correlation with bar motion than with background motion. In contrast, the neuron with a larger receptive field shows a higher correlation with background motion.

An essential step to achieving human-machine speech communication with the naturalness of communication between humans is developing a machine that is capable of recognising emotions based on speech. This thesis presents research addressing this problem, by making use of acoustic and prosodic information. At a feature level, novel group delay and weighted frequency features are proposed. The group delay features are shown to emphasise information pertaining to formant bandwidths and are shown to be indicative of emotions. The weighted frequency feature, based on the recently introduced empirical mode decomposition, is proposed as a compact representation of the spectral energy distribution and is shown to outperform other estimates of energy distribution. Feature level comparisons suggest that detailed spectral measures are very indicative of emotions while exhibiting greater speaker specificity. Moreover, it is shown that all features are characteristic of the speaker and require some of sort of normalisation prior to use in a multi-speaker situation. A novel technique for normalising speaker-specific variability in features is proposed, which leads to significant improvements in the performances of systems trained and tested on data from different speakers. This technique is also used to investigate the amount of speaker-specific variability in different features. A preliminary study of phonetic variability suggests that phoneme specific traits are not modelled by the emotion models and that speaker variability is a more significant problem in the investigated setup. Finally, a novel approach to emotion modelling that takes into account temporal variations of speech parameters is analysed. An explicit model of the glottal spectrum is incorporated into the framework of the traditional source-filter model, and the parameters of this combined model are used to characterise speech signals. An automatic emotion recognition system that takes into account the shape of the contours of these parameters as they vary with time is shown to outperform a system that models only the parameter distributions. The novel approach is also empirically shown to be on par with human emotion classification performance.