Bartosz Binias

Real-time detection and filtering of eye movement and blink related artifacts in EEG

Eye movement related artifacts are the most significant source of noise in EEG signals. Thus, a special approach to reduction of their influence is required. However, most of currently used methods of detecting and filtering eye movement related artifacts require either an additional recording of noise signal, or are not suitable for real time applications, such as Brain-Computer Interfaces. In this paper it was proven that it is possible to detect and filter those artifacts in real time, without the need of providing an additional recording of noise signal.

Real-Time Detection and Filtering of Eye Blink Related Artifacts for Brain-Computer Interface Applications

Artifacts related with eye movements are the most significant source of noise in EEG signals. Although there are many methods of their filtering available, most of them are not suitable for real-time applications, such as Brain-Computer Interfaces. In addition, most of those methods require an additional recording of noise signal to be provided. Applying filtering to the recorded EEG signal may unintentionally distort its uncontaminated segments. To reduce that effect filtering should be applied only to those parts of signal that were marked as artifacts. In this paper it was proven that it is possible to detect and filter those artifacts in real-time, without the need of providing an additional recording of noise signal.

Pixel Classification for Skin Detection in Color Images

In this paper a direct, pixel-based skin detection method is proposed and evaluated. Proposed approach discards any spatial information that can be found in digital image and focuses entirely on data-oriented analysis. To ensure the best perfomance two classifiers (Regularized Logistic Regression and Artificial Neural Network with Regularization trained with Backpropagation) were deeply examined, evaluated and compared for this task. The best model achieved the almost perfect accuracy and quality of classification on the used ‘Skin Segmentation Dataset’ provided for the UCI Machine Learning Repository with over 99% accuracy, precision, recall and specificity.

A Machine Learning Approach to the Detection of Pilot’s Reaction to Unexpected Events Based on EEG Signals

This work considers the problem of utilizing electroencephalographic signals for use in systems designed for monitoring and enhancing the performance of aircraft pilots. Systems with such capabilities are generally referred to as cognitive cockpits. This article provides a description of the potential that is carried by such systems, especially in terms of increasing flight safety. Additionally, a neuropsychological background of the problem is presented. Conducted research was focused mainly on the problem of discrimination between states of brain activity related to idle but focused anticipation of visual cue and reaction to it. Especially, a problem of selecting a proper classification algorithm for such problems is being examined. For that purpose an experiment involving 10 subjects was planned and conducted. Experimental electroencephalographic data was acquired using an Emotiv EPOC+ headset. Proposed methodology involved use of a popular method in biomedical signal processing, the Common Spatial Pattern, extraction of bandpower features, and an extensive test of different classification algorithms, such as Linear Discriminant Analysis, k-nearest neighbors, and Support Vector Machines with linear and radial basis function kernels, Random Forests, and Artificial Neural Networks.

Adaptive Nonlinear Projective Filtering - Application to Filtering of Artifacts in EEG Signals.

In this work a novel approach to filtering of eyeblink related artifacts from EEG signals is presented. Proposed solution, the Adaptive Nonlinear Projective Filtering (ANPF) algorithm, combines the classic approach to adaptive filtering with algorithms from nonlinear state space projection family. Performance of described method is compared with adaptive filter based on Normalized Least Mean Squares algorithm in terms of median Normalized Mean Squared Error. Data used in conducted research was simulated according to described procedure. Such approach allowed for a reliable comparison and evaluation of algorithm’s signal correction properties. Additionally, a real time modification of ANPF algorithm is proposed and tested. The analysis of sensitivity to changes of parameter values was also performed. Achieved results were tested for statistical significance. According to obtained scores ANPF significantly outperforms referential method during offline processing.

Approximation of thrust generated by rotor in quadrotor unit with use of generalized polynomial

This paper presents the way of modeling thrust generated by a quadrotor flying unit. The highly non-linear equations describing rotor thrust are derived with use of momentum theory and approximated with use of generalized polynomial. The article includes the model formulation and the example of approximation. The polynomial approximation ensures the simplicity in solving during simulation with sufficient accuracy.

Normalization of feature distribution in motor imagery based brain-computer interfaces

Brain-Computer Interfaces (BCIs) are systems capable of capturing and interpreting the consent changes in the activity of brain (e.g. intention of limb movement, attention focus on specific frequency or symbol) and translating them into sets of instructions, which can be used for the control of a computer. The most popular hardware solutions in BCI are based on the signals recorded by the electroencephalograph (EEG). Such signals can be used to record and monitor the bioelectrical activity of the brain. However, raw EEG scalp potentials are characterized by a weak spatial resolution. Due to that reason, multichannel EEG recordings tend to provide an unclear image of the activity of brain and the use of special signal processing and analysis methods is needed. A typical approach towards modern BCIs requires an extensive use of Machine Learning methods. It is generally accepted that the performance of such systems is highly sensitive to the feature extraction step. One of the most effective and widely used descriptors of EEG data is the power of the signal calculated in a specific frequency range. In order to improve the performance of chosen classification algorithm, the distribution of the extracted bandpower features is often normalized with the use of natural logarithm function. In this study the step of normalization of feature distribution was taken into careful consideration. Commonly used logarithm function is not always the best choice for this process. Therefore, the influence on the skewness of features, as well as, on the general classification accuracy of different settings of Box-Cox transformation will be tested in this article and compared to classical approach that employs natural logarithm function. For the better evaluation of the performance of the proposed approach, its effectiveness is tested in the task of classification of the benchmark data provided for the “BCI Competition III” (dataset “IVa”) organized by the Berlin Brain-Computer Interface group.

Windowed local area average reference filter for increasing the spatial resolution of EEG signals

The phenomenon of transmitting the electric fields from their primary bioelectric sources through biological tissues towards measurement sensors is known as the volume conduction. It is a basis for the operation of many biosensors used for the measurement of bioelectromagnetical phenomena. An electroencephalograph (EEG) used for the recording and monitoring the bioelectrical activity of brain is an excellent example of such devices. Due to the volume conduction an overlapping of the contribution to each electrode channel from neighbouring bioelectrical sources is present. As a result raw EEG scalp potentials are characterized by weak spatial resolution. Owning to described reasons raw, multichannel EEG recordings tend to provide an unclear image of the brain activity. In this study the influence of applying 2D window function to the neighbourhood of each electrode before removal of common reference locally is examined. The Local Spatial Filters focus on removing the common signal of electrodes in a specific neighbourhood of the electrode of interest. The performance of algorithm is examined in the task of eliminating source overlapping from EEG-based BCI system recordings and compared with classic approaches such as Common Average Reference (CAR), Surface Laplacian (SL) and Common Spatial Pattern (CSP). For the better evaluation of performance of proposed approaches performance, their effectiveness is tested during the classification of the dataset IVa provided for the “BCI Competition III” organized by the Berlin Brain-Computer Interface group.