1 - 4 of 4
rss atomLink til resultatlisten
Permanent link
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Annet format
Fler format
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Annet språk
Fler språk
  • html
  • text
  • asciidoc
  • rtf
  • Disputas: 2019-02-21 13:15 Zeta, Västerås
    Barua, Shaibal
    Mälardalens högskola, Akademin för innovation, design och teknik, Inbyggda system.
    Multivariate Data Analytics to Identify Driver’s Sleepiness, Cognitive load, and Stress2019Doktoravhandling, med artikler (Annet vitenskapelig)
    Abstract [en]

    Driving a vehicle in a dynamic traffic environment requires continuous adaptation of a complex manifold of physiological and cognitive activities. Impaired driving due to, for example, sleepiness, inattention, cognitive load or stress, affects one’s ability to adapt, predict and react to upcoming traffic events. In fact, human error has been found to be a contributing factor in more than 90% of traffic crashes. Unfortunately, there is no robust, objective ground truth for determining a driver’s state, and researchers often revert to using subjective self-rating scales when assessing level of sleepiness, cognitive load or stress. Thus, the development of better tools to understand, measure and monitor human behaviour across diverse scenarios and states is crucial. The main objective of this thesis is to develop objective measures of sleepiness, cognitive load and stress, which can later be used as research tools, either to benchmark unobtrusive sensor solutions or when investigating the influence of other factors on sleepiness, cognitive load, and stress.

    This thesis employs multivariate data analysis using machine learning to detect and classify different driver states based on physiological data. The reason for using rather intrusive sensor data, such as electroencephalography (EEG), electrooculography (EOG), electrocardiography (ECG), skin conductance, finger temperature, and respiration is that these methods can be used to analyse how the brain and body respond to internal and external changes, including those that do not generate overt behaviour. Moreover, the use of physiological data is expected to grow in importance when investigating human behaviour in partially automated vehicles, where active driving is replaced by passive supervision.

    Physiological data, especially the EEG is sensitive to motion artifacts and noise, and when recorded in naturalistic environments such as driving, artifacts are unavoidable. An automatic EEG artifact handling method ARTE (Automated aRTifacts handling in EEG) was therefore developed. When used as a pre-processing step in the classification of driver sleepiness, ARTE increased classification performance by 5%. ARTE is data-driven and does not rely on additional reference signals or manually defined thresholds, making it well suited for use in dynamic settings where unforeseen and rare artifacts are commonly encountered. In addition, several machine-learning algorithms have been developed for sleepiness, cognitive load, and stress classification. Regarding sleepiness classification, the best achieved accuracy was achieved using a Support Vector Machine (SVM) classifier. For multiclass, the obtained accuracy was 79% and for binary class it was 93%. A subject-dependent classification exhibited a 10% improvement in performance compared to the subject-independent classification, suggesting that much can be gained by using personalized classifiers. Moreover, by embedding contextual information, classification performance improves by approximately 5%. In regard to cognitive load classification, a 72% accuracy rate was achieved using a random forest classifier. Combining features from several data sources may improve performance, and indeed, we observed classification performance improvement by 10%-20% compared to using features from a single data source. To classify drivers’ stress, using the Case-based reasoning (CBR) and data fusion approach, the system achieved an 83.33% classification accuracy rate.

    This thesis work encourages the use of multivariate data for detecting and classifying driver states, including sleepiness, cognitive load, and stress. A univariate data source often presents challenges, since features from a single source or one just aspect of the feature are not entirely reliable; Therefore, multivariate information requires accurate driver state detection. Often, driver states are a subjective experience, in which other contextual data plays a vital role. Thus, the implication of incorporating contextual information in the classification scheme is presented in this thesis work. Although there are several commonalities, physiological signals are modulated differently in different driver states; Hence, multivariate data could help detect multiple driver states simultaneously – for example, cognitive load detection when a person is under the influence of different levels of stress.

  • Disputas: 2019-02-22 09:30 Delta, Västerås
    Abbaspour, Sara
    Mälardalens högskola, Akademin för innovation, design och teknik, Inbyggda system.
    Electromyogram Signal Enhancement and Upper-Limb Myoelectric Pattern Recognition2019Doktoravhandling, med artikler (Annet vitenskapelig)
    Abstract [en]

    Losing a limb causes difficulties in our daily life. To regain the ability to live an independent life, artificial limbs have been developed. Hand prostheses belong to a group of artificial limbs that can be controlled by the user through the activity of the remnant muscles above the amputation. Electromyogram (EMG) is one of the sources that can be used for control methods for hand prostheses. Surface EMGs are powerful, non-invasive tools that provide information about neuromuscular activity of the subjected muscle, which has been essential to its use as a source of control for prosthetic limbs. However, the complexity of this signal introduces a big challenge to its applications. EMG pattern recognition to decode different limb movements is an important advancement regarding the control of powered prostheses. It has the potential to enable the control of powered prostheses using the generated EMG by muscular contractions as an input. However, its use has yet to be transitioned into wide clinical use. Different algorithms have been developed in state of the art to decode different movements; however, the challenge still lies in different stages of a successful hand gesture recognition and improvements in these areas could potentially increase the functionality of powered prostheses. This thesis firstly focuses on improving the EMG signal’s quality by proposing novel and advanced filtering techniques. Four efficient approaches (adaptive neuro-fuzzy inference system-wavelet, artificial neural network-wavelet, adaptive subtraction and automated independent component analysis-wavelet) are proposed to improve the filtering process of surface EMG signals and effectively eliminate ECG interferences. Then, the offline performance of different EMG-based recognition algorithms for classifying different hand movements are evaluated with the aim of obtaining new myoelectric control configurations that improves the recognition stage. Afterwards, to gain proper insight on the implementation of myoelectric pattern recognition, a wide range of myoelectric pattern recognition algorithms are investigated in real time. The experimental result on 15 healthy volunteers suggests that linear discriminant analysis (LDA) and maximum likelihood estimation (MLE) outperform other classifiers. The real-time investigation illustrates that in addition to the LDA and MLE, multilayer perceptron also outperforms the other algorithms when compared using classification accuracy and completion rate.

  • Disputas: 2019-03-08 13:15 Gamma, Västerås
    Weishaupt, Hrafn Holger
    Mälardalens högskola, Akademin för utbildning, kultur och kommunikation, Utbildningsvetenskap och Matematik.
    Graph theory based approaches for gene prioritization in biological networks: Application to cancer gene detection in medulloblastoma2019Doktoravhandling, med artikler (Annet vitenskapelig)
    Abstract [en]

    Networks provide an intuitive and highly adaptable means to model relationships between objects. When translated to mathematical graphs, they become further amenable to a plethora of mathematical operations that allow a detailed study of the underlying relational data. Thus, it is not surprising that networks have evolved to a predominant method for analyzing such data in a vast variety of research fields. However, with increasing complexity of the studied problems, application of network modeling also becomes more challenging. Specifically, given a process to be studied, (i) which interactions are important and how can they be modeled, (ii) how can relationships be inferred from complex and potentially noisy data, and (iii) which methods should be used to test hypotheses or answer the relevant questions? This thesis explores the concept and challenges of network analysis in the context of a well-defined application area, i.e. the prediction of cancer genes from biological networks, with an application to medulloblastoma research.

    Medulloblastoma represents the most common malignant brain tumor in children. Currently about 70% of treated patients survive, but they often suffer from permanent cognitive sequelae. Medulloblastoma has previously been shown to harbor at least four distinct molecular subgroups. Related studies have also greatly advanced our understanding of the genetic aberrations associated with MB subgroups. However, to translate such findings to novel and improved therapy options, further insights are required into how the dysregulated genes interact with the rest of the cellular system, how such a cross-talk can drive tumor development, and how the arising tumorigenic processes can be targeted by drugs. Establishing such understanding requires investigations that can address biological processes at a more system-wide level, a task that can be approached through the study of cellular systems as mathematical networks of molecular interactions.

    This thesis discusses the identification of cancer genes from a network perspective, where specific focus is placed on one particular type of network, i.e. so called gene regulatory networks that model relationships between genes at the expression level. The thesis outlines the bridge between biological and mathematical network concepts. Specifically, the computational challenge of inferring such networks from molecular data is presented. Mathematical approaches for analyzing these networks are outlined and it is explored how such methods might be affected by network inference. Further focus is placed on dealing with the challenges of establishing a suitable gene expression dataset for network inference in MB. Finally, the thesis is concluded with an application of various network approaches in a hypothesis-driven study in MB, in which various novel candidate genes were prioritized.  

  • Disputas: 2019-03-19 09:30 Gamma, Västerås
    Du, Jiaying
    Mälardalens högskola, Akademin för innovation, design och teknik, Inbyggda system. -.
    Real-time signal processing in MEMS sensor-based motion analysis systems2019Doktoravhandling, med artikler (Annet vitenskapelig)
    Abstract [en]

    This PhD thesis focuses on real-time signal processing for hardware-limited micro-electro-mechanical system (MEMS) sensor-based human motion analysis systems. The aim of the thesis is to improve the signal quality of MEMS gyroscopes and accelerometers by minimizing the effects of signal errors, considering the hardware limitations and the users' perception.

    MEMS sensors such as MEMS gyroscopes and MEMS accelerometers are important components in motion analysis systems. They are known for their small size, light weight, low power consumption, low cost, and high sensitivity. This makes them suitable for wearable systems for measuring body movements. The data can further be used as input for advanced human motion analyses. However, MEMS sensors are usually sensitive to environmental disturbances such as shock, vibration, and temperature change. A large portion of the MEMS sensor signals actually originate from error sources such as noise, offset, null drift and temperature drift, as well as integration drift. Signal processing is regarded as the major key solution to reduce these errors. For real-time signal processing, the algorithms need to be executed within a certain specified time limit. Two crucial factors have to be considered when designing real-time signal processing algorithms for wearable embedded sensor systems. One is the hardware limitations leading to a limited calculation capacity, and the other is the user perception of the delay caused by the signal processing.

    Within this thesis, a systematic review of different signal error reduction algorithms for MEMS gyroscope-based motion analysis systems for human motion analysis is presented. The users’ perceptions of the delay when using different computer input devices were investigated. 50 ms was found as an acceptable delay for the signal processing execution in a real-time motion analysis system. Real-time algorithms for noise reduction, offset/drift estimation and reduction, improvement of position accuracy and system stability considering the above mentioned requirements, are presented in this thesis. The algorithms include a simplified high-pass filter and low-pass filter, a LMS algorithm, a Kalman filter, a WFLC algorithm, two simple novel algorithms (a TWD method and a velocity drift estimation method), and a novel combination method KWT.  Kalman filtering was found to be efficient to reduce the problem of temperature drift and the WFLC algorithm was found the most suitable method to reduce human physiological tremor and electrical noise. The TWD method resulted in a signal level around zero without interrupting the continuous movement signal. The combination method improved the static stability and the position accuracy considerably.  The computational time for the execution of the algorithms were all perceived as acceptable by users and kept within the specified time limit for real-time performance.  Implementations and experiments showed that these algorithms are feasible for establishing high signal quality and good system performance in previously developed systems, and also have the potential to be used in similar systems.

    Fulltekst tilgjengelig fra 2019-02-26 08:00