No description available.

La première partie de ce mémoire présente une nouvelle vision de l'estimation de mouvement, et donc de la compression, dans les séquences<br />vidéo. D'une part, nous avons choisi d'aborder l'estimation de mouvement à partir de familles d'ondelettes redondantes adaptées à différentes<br />transformations, dont, plus particulièrement, la vitesse. Ces familles, très peu connues, ont déjà été étudiées dans le cadre de la poursuite de<br />cibles. D'autre part, les standards de compression actuels comme MPEG4 prennent en compte une compression objet mais ne calculent toujours que de<br />simples vecteurs de mouvements de ``blocs''. Il nous a paru intéressant de chercher à mettre en oeuvre ces familles d'ondelettes car 1)<br />elle sont construites pour le calcul de paramètres sur plusieurs types de mouvement (rotation, vitesse, accélération) et 2) nous<br />pensons qu'une approche de l'estimation basée sur l'identification de trajectoires d'objets dans une scène est une solution intéressante pour les<br />méthodes futures de compression. En effet nous pensons que l'analyse et la compréhension des mouvements dans une scène est une voie pour des méthodes<br />de compression ``contextuelles'' performantes.<br /><br /><br /><br />La seconde partie présente deux développements concernant la segmentation non-supervisée dans une approche bayésienne. Le premier, destiné à réduire<br />les temps de calcul dans la segmentation de séquences vidéo, est basé sur une mise en oeuvre itérative, simple, de la segmentation. Il nous a aussi<br />permis de mettre une estimation de mouvement basée sur une segmentation ``région'' (voire objet). Le second est destiné à diminuer les temps de<br />segmentation d'images fixes en réalisant la segmentation dans le domaine des ondelettes. Ces deux développements sont basés sur une approche par<br />estimation bayésienne utilisant un modèle de champ aléatoire de Potts-Markov (PMRF) pour les étiquettes des pixels, dans le domaine direct, et pour<br />les coefficients d'ondelettes. Il utilise aussi un algorithme itératif de type MCMC (Markov Chain Monte Carlo) avec échantillonneur de Gibbs.<br />L'approche initiale, directe, utilise un modèle de Potts avec voisinage d'ordre un. Nous avons développé le modèle de Potts pour l'adapter à des<br />voisinages convenant aux orientations privilégiées des sous-bandes d'ondelettes. Ces réalisations apportent, à notre connaissance, des approches<br />nouvelles dans les méthodes de segmentation<br />non-supervisées.

Work is aimed at the time-frequency decomposition of a signal application for monitoring the EKG trend progression. Goal is to create algorithm which would watch changes in the ST segment in EKG recording and its realization in the Matlab program. Analyzed is substance of the origin of EKG and its measuring. For trend calculations after reading the signal is necessary to preprocess the signal, it consists of filtration and detection of necessary points of EKG signal. For taking apart, also filtration and measuring the signal is used wavelet transformation. Source of the data is biomedicine database Physionet. As an outcome of the algorithm are drawn ST segment trends for three recordings from three different patients and its comparison with reference method of ST qualification. For qualification of the heart stability, as a system, where designed methods watching differences in position of the maximal value in two-zone spectrum and the Poincare mapping method. Realized method is attached to this thesis.

Muscle fatigue is a severe problem for elite athletes, and this is due to the long resting times, which can vary. Various mechanisms can cause muscle fatigue which signifies that the specific muscle has reached its maximum force and cannot continue the task. This thesis was about surveying and exploring state-of-the-art methods and systematically, theoretically, and practically testing the applicability and performance of more recent machine learning methods on an existing EMG to muscle fatigue pipeline. Several challenges within the EMG domain exist, such as inadequate data, finding the most suitable model, and how they should be addressed to achieve reliable prediction. This required approaches for addressing these problems by combining and comparing various state-of-the-art methodologies, such as data augmentation techniques for upsampling, spectrogram methods for signal processing, and transfer learning to gain a reliable prediction by various pre-trained CNN models. The approach during this study was to conduct seven experiments consisting of a classification task that aims to predict muscle fatigue in various stages. These stages are divided into 7 classes from 0-6, and higher classes represent a fatigued muscle. In the tabular part of the experiments, the Decision Tree, Random Forest, and Support Vector Machine (SVM) were trained, and the accuracy was determined. A similar approach was made for the spectrogram part, where the signals were converted to spectrogram images, and with a combination of traditional- and intelligent data augmentation techniques, such as noise and DCGAN, the limited dataset was increased. A comparison between the performance of AlexNet, VGG16, DenseNet, and InceptionV3 pre-trained CNN models was made to predict differences in jump heights. The result was evaluated by implementing baseline classifiers on tabular data and pre-trained CNN model classifiers for CWT and STFT spectrograms with and without data augmentation. The evaluation of various state-of-the-art methodologies for a classification problem showed that DenseNet and VGG16 gave a reliable accuracy of 89.8 % on intelligent data augmented CWT images. The intelligent data augmentation applied on CWT images allows the pre-trained CNN models to learn features that can generalize unseen data. Proving that the combination of state-of-the-art methods can be introduced and address the challenges within the EMG domain.

Vibrations are everywhere around us all the time and we often recognise them as sounds that we can hear and analyse with our brain. In this thesis, data that has been gathered from a diverter switch (DS) in a controlled environment, is analysed. This data consists of vibroacoustic measurements and information to indicate what is happening inside the DS as the vibroacoustic data is gathered. The frequency properties of vibroacoustic data from the DS gathered before this thesis are displayed using a wavelet transformation model. This means the frequency properties of the signal can be approximated for all times in the operation with a certain accuracy. As the DS is built from many different components, the frequency properties of these components are compared to the time-frequency picture of the full DS operation. This sort of comparison ends up not being feasible as the complexity of the DS frequency pattern is much more than that of a sum of its component’s frequency pattern. A second approach of analysing the gathered vibroacoustic data is by using a classification model. The information about what is happening inside of the DS is used to train a logistic regression model on different defined regions of the vibroacoustic data. Before the training is preformed though, the different defined regions are transformed into frequency space with help of the fast Fourier transform. With this, a classification model is produced, where vibroacoustic data of any time region can be fed into the model and the model will classify which defined region this vibroacoustic data belongs to. The results are promising, and the model can be used both for classification of the defined regions and potentially used to determine if the vibroacoustic properties of the DS has changed due to wear of the mechanical components or transformer oil.

In the production of paper, the quality of the pulp is an important factor both for the productivity and for the final quality. Reliable real-time measurements of pulp quality are therefore needed. One way is to use acoustic or vibration sensors that give information-rich signals and place the sensors at suitable locations in a pulp production line. However, these sensors are not selective for the pulp properties of interest. Therefore, advanced signal processing and multivariate calibration are essential tools. The current work has been focused on the development of calibration routes for extraction of information from acoustic sensors and on signal processing algorithms for enhancing the information-selectivity for a specific pulp property or class of properties. Multivariate analysis methods like Principal Components Analysis (PCA), Partial Least Squares (PLS) and Orthogonal Signal Correction (OSC) have been used for visualization and calibration. Signal processing methods like Fast Fourier Transform (FFT), Fast Wavelet Transform (FWT) and Continuous Wavelet Transform (CWT) have been used in the development of novel signal processing algorithms for extraction of information from vibrationacoustic sensors. It is shown that use of OSC combined with PLS for prediction of Canadian Standard Freeness (CSF) using FFT-spectra produced from vibration data on a Thermo Mechanical Pulping (TMP) process gives lower prediction errors and a more parsimonious model than PLS alone. The combination of FFT and PLS was also used for monitoring of beating of kraft pulp and for screen monitoring. When using regular FFT-spectra on process acoustic data the obtained information tend to overlap. To circumvent this two new signal processing methods were developed: Wavelet Transform Multi Resolution Spectra (WT-MRS) and Continuous Wavelet Transform Fibre Length Extraction (CWT-FLE). Applying WT-MRS gave PLS-models that were more parsimonious with lower prediction error for CSF than using regular FFT-Spectra. For a Medium Consistency (MC) pulp stream WT-MRS gave predictions errors comparable to the reference methods for CSF and Brightness. The CWT-FLE method was validated against a commercial fibre length analyzer and good agreement was obtained. The CWT-FLE-curves could therefore be used instead of other fibre distribution curves for process control. Further, the CWT-FLE curves were used for PLS modelling of tensile strength and optical parameters with good results. In addition to the mentioned results a comprehensive overview of technologies used with acoustic sensors and related applications has been performed. / Vid framställning av pappersprodukter är kvaliteten på massan en viktig faktor för produktiviteten och kvalitén på slutresultatet. Det är därför viktigt att ha tillgång till tillförlitliga mätningar av massakvalitet i realtid. En möjlighet är att använda akustik- eller vibrationssensorer i lämpliga positioner vid enhetsoperationer i massaprocessen. Selektiviteten hos dessa mätningar är emellertid relativt låg i synnerhet om mätningarna är passiva. Därför krävs avancerad signalbehandling och multivariat kalibrering. Det nu presenterade arbetet har varit fokuserat på kalibreringsmetoder för extraktion av information ur akustiska mätningar samt på algoritmer för signalbehandling som kan ge förbättrad informationsselektivitet. Multivariata metoder som Principal Component Analysis (PCA), Partial Least Squares (PLS) and Orthogonal Signal Correction (OSC) har använts för visualisering och kalibrering. Signalbehandlingsmetoderna Fast Fourier Transform (FFT), Fast Wavelet Transform (FWT) och Continuous Wavelet Transform (CWT) har använts i utvecklingen av nydanande metoder för signalbehandling anpassade till att extrahera information ur signaler från vibrations/akustiska sensorer. En kombination av OSC och PLS applicerade på FFT-spektra från raffineringen i en Termo Mechnaical Pulping (TMP) process ger lägre prediktionsfel för Canadian Standard Freeness (CSF) än enbart PLS. Kombinationen av FFT och PLS har vidare använts för monitorering av malning av sulfatmassa och monitorering av silning. Ordinära FFT-spektra av t.ex. vibrationssignaler är delvis överlappande. För att komma runt detta har två signalbehandlingsmetoder utvecklats, Wavelet Transform Multi Resolution Spectra (WT-MRS) baserat på kombinationen av FWT och FFT samt Continuous Wavelet Transform Fibre Length Extraction (CWT-FLE) baserat på CWT. Tillämpning av WT-MRS gav enklare PLS-modeller med lägre prediktionsfel för CSF jämfört med att använda normala FFT-spektra. I en annan tillämpning på en massaström med relativt hög koncentration (Medium Consistency, MC) kunde prediktioner för CSF samt ljushet erhållas med prediktionsfel jämförbart med referensmetodernas fel. Metoden CWT-FLE validerades mot en kommersiell fiberlängdsmätare med god överensstämmelse. CWT-FLE-kurvorna skulle därför kunna användas i stället för andra fiberdistributionskurvor för processtyrning. Vidare användes CWT-FLE kurvor för PLS modellering av dragstyrka samt optiska egenskaper med goda resultat. Utöver de nämnda resultaten har en omfattande litteratursammanställning gjorts över området och relaterade applikationer. / QC 20100629

The idea of time series forecasting techniques is that the past has certain information about future. So, the question of how the information is encoded in the past can be interpreted and later used to extrapolate events of future constitute the crux of time series analysis and forecasting. Several methods such as qualitative techniques (e.g., Delphi method), causal techniques (e.g., least squares regression), quantitative techniques (e.g., smoothing method, time series models) have been developed in the past in which the concept lies in establishing a model either theoretically or mathematically from past observations and estimate future from it. Of all the models, time series methods such as autoregressive moving average (ARMA) process have gained popularity because of their simplicity in implementation and accuracy in obtaining forecasts. But, these models were formulated based on certain properties that a time series is assumed to possess. Classical decomposition techniques were developed to supplement the requirements of time series models. These methods try to define a time series in terms of simple patterns called trend, cyclical and seasonal patterns along with noise. So, the idea of decomposing a time series into component patterns, later modeling each component using forecasting processes and finally combining the component forecasts to obtain actual time series predictions yielded superior performance over standard forecasting techniques. All these methods involve basic principle of moving average computation. But, the developed classical decomposition methods are disadvantageous in terms of containing fixed number of components for any time series, data independent decompositions. During moving average computation, edges of time series might not get modeled properly which affects long range forecasting. So, these issues are to be addressed by more efficient and advanced decomposition techniques such as Wavelets and Empirical Mode Decomposition (EMD). Wavelets and EMD are some of the most innovative concepts considered in time series analysis and are focused on processing nonlinear and nonstationary time series. Hence, this research has been undertaken to ascertain the predictability of nonstationary time series using wavelet and Empirical Mode Decomposition (EMD) based ARMA models. The development of wavelets has been made based on concepts of Fourier analysis and Window Fourier Transform. In accordance with this, initially, the necessity of involving the advent of wavelets has been presented. This is followed by the discussion regarding the advantages that are provided by wavelets. Primarily, the wavelets were defined in the sense of continuous time series. Later, in order to match the real world requirements, wavelets analysis has been defined in discrete scenario which is called as Discrete Wavelet Transform (DWT). The current thesis utilized DWT for performing time series decomposition. The detailed discussion regarding the theory behind time series decomposition is presented in the thesis. This is followed by description regarding mathematical viewpoint of time series decomposition using DWT, which involves decomposition algorithm. EMD also comes under same class as wavelets in the consequence of time series decomposition. EMD is developed out of the fact that most of the time series in nature contain multiple frequencies leading to existence of different scales simultaneously. This method, when compared to standard Fourier analysis and wavelet algorithms, has greater scope of adaptation in processing various nonstationary time series. The method involves decomposing any complicated time series into a very small number of finite empirical modes (IMFs-Intrinsic Mode Functions), where each mode contains information of the original time series. The algorithm of time series decomposition using EMD is presented post conceptual elucidation in the current thesis. Later, the proposed time series forecasting algorithm that couples EMD and ARMA model is presented that even considers the number of time steps ahead of which forecasting needs to be performed. In order to test the methodologies of wavelet and EMD based algorithms for prediction of time series with non stationarity, series of streamflow data from USA and rainfall data from India are used in the study. Four non-stationary streamflow sites (USGS data resources) of monthly total volumes and two non-stationary gridded rainfall sites (IMD) of monthly total rainfall are considered for the study. The predictability by the proposed algorithm is checked in two scenarios, first being six months ahead forecast and the second being twelve months ahead forecast. Normalized Root Mean Square Error (NRMSE) and Nash Sutcliffe Efficiency Index (Ef) are considered to evaluate the performance of the proposed techniques. Based on the performance measures, the results indicate that wavelet based analyses generate good variations in the case of six months ahead forecast maintaining harmony with the observed values at most of the sites. Although the methods are observed to capture the minima of the time series effectively both in the case of six and twelve months ahead predictions, better forecasts are obtained with wavelet based method over EMD based method in the case of twelve months ahead predictions. It is therefore inferred that wavelet based method has better prediction capabilities over EMD based method despite some of the limitations of time series methods and the manner in which decomposition takes place. Finally, the study concludes that the wavelet based time series algorithm could be used to model events such as droughts with reasonable accuracy. Also, some modifications that could be made in the model have been suggested which can extend the scope of applicability to other areas in the field of hydrology.

Epilepsy affects more than 50 million people and is one of the most prevalent neurological disorders and has a high impact on the quality of life of those suffering from it. However, 70% of epilepsy patients can live seizure free with proper diagnosis and treatment. Patients are evaluated using scalp EEG recordings which is cheap and non-invasive. Diagnostic yield is however low and qualified personnel need to process large amounts of data in order to accurately assess patients. MindReader is an unsupervised classifier which detects spectral anomalies and generates a hypothesis of the underlying patient state over time. The aim is to highlight abnormal, potentially epileptiform states, which could expedite analysis of patients and let qualified personnel attest the results. It was used to evaluate 95 scalp EEG recordings from healthy adults and adult patients with epilepsy. Interictal Epileptiform discharges (IED) occurring in the samples had been retroactively annotated, along with the patient state and maneuvers performed by personnel, to enable characterization of the classifier’s detection performance. The performance was slightly worse than previous benchmarks on pediatric scalp EEG recordings, with a 7% and 33% drop in specificity and sensitivity, respectively. Electrode positioning and partial spatial extent of events saw notable impact on performance. However, no correlation between annotated disturbances and reduction in performance could be found. Additional explorative analysis was performed on serialized intermediate data to evaluate the analysis design. Hyperparameters and electrode montage options were exposed to optimize for the average Mathew’s correlation coefficient (MCC) per electrode per patient, on a subset of the patients with epilepsy. An increased window length and lowered amount of training along with an common average montage proved most successful. The Euclidean distance of cumulative spectra (ECS), a metric suitable for spectral analysis, and homologous L2 and L1 loss function were implemented, of which the ECS further improved the average performance for all samples. Four additional analyses, featuring new time-frequency transforms and multichannel convolutional autoencoders were evaluated and an analysis using the continuous wavelet transform (CWT) and a convolutional autoencoder (CNN) performed the best, with an average MCC score of 0.19 and 56.9% sensitivity with approximately 13.9 false positives per minute.