Student EEICT 2019

Browse

Recent Submissions

Now showing 1 - 5 of 174
  • Thumbnail Image
    Item
    Language-Independent Text Classifier Based On Recurrent Neural Networks
    (Vysoké učení technické v Brně, Fakulta elektrotechniky a komunikačních technologií, 2019) Myska, Vojtech
    This paper deals with a proposal of language independent text classifiers based on recurrent neural networks. They work at a character level thus they do not require any text preprocessing. The classifiers have been trained and evaluated on a multilingual data set that is privately collected from film review databases. It contains Czech (Slovak), English, German and Spanish language subset. The resulting accuracy of the proposed language independent classifiers base on the recurrent neural networks in polarity sentiment analysis task is 78.55%.
  • Thumbnail Image
    Item
    Pseudo-Differential High-Order Frequency Filter
    (Vysoké učení technické v Brně, Fakulta elektrotechniky a komunikačních technologií, 2019) Sládok, Ondřej
    This article describes a pseudo-differential third-order frequency filter operating in voltage mode, using four differential difference current conveyors (DDCCs), and six passive elements. The circuit has a high input impedance, low active and passive sensitivity and high common-mode rejection ratio (CMRR). The proposed structure is able to realize one type of frequency responses low pass. Non-ideal analysis has been performed by considering the real parasitic parameters of the active elements. Optimization was made with a view to greatly influence low pass filter attenuation.
  • Thumbnail Image
    Item
    Cycling Of Vrla Lead-Acid Batteries For Use In Uninterruptible Power Supplies And Measurement Of Failed Batteries
    (Vysoké učení technické v Brně, Fakulta elektrotechniky a komunikačních technologií, 2019) Musil, Petr
    This article deals with study on the influence of cycling of batteries with regard to their parameters. Electro-chemical reactions which take part in battery cells are described. This paper also offers a measurement scheme for automated measuring workstation. Furthermore, results of measurements are being presented on measured parameters of chosen batteries. Comparison of the measurements results and datasheet values is included. Conclusion sums up parameters of chosen batteries and their feasibility for further usage in Uninterruptible Power Supply as a crucial part of critical infrastructure.
  • Thumbnail Image
    Item
    Common Carotid Artery Wall Localization In B-Mode Ultrasound Images
    (Vysoké učení technické v Brně, Fakulta elektrotechniky a komunikačních technologií, 2019) Dorazil, Jan
    Analysis of B-mode ultrasound images capturing the common carotid artery (CCA) provides significant indicators of the overall health of the cardiovascular system. In this paper we propose a novel method for automatic localization of the artery wall contour (approximated by circle) in utrasound images of the transverse section of the CCA. After detection of a region of interest (ROI) using a modified Viola-Jones detector we localize the best-fitting circle, which delimits the artery wall contour, by exhaustive search. Experimental results on a dataset of 145 ultrasound images show that the method outperforms a reference method based on the Hough transform and presents an excellent obustness against additive noise on different SNR levels.
  • Thumbnail Image
    Item
    Denoise Pre-Training For Segmentation Neural Networks
    (Vysoké učení technické v Brně, Fakulta elektrotechniky a komunikačních technologií, 2019) Kolarik, Martin
    This paper proposes a method for pre-training segmentation neural networks on small datasets using unlabelled training data with added noise. The pre-training process helps the network with initial better weights settings for the training itself and also augments the training dataset when dealing with small labelled datasets especially in medical imaging. The experiment comparing results of pre-trained and not pre-trained networks on MRI brain segmentation task has shown that the denoise pre-training helps the network with faster training convergence without overfitting and achieving better results in all compared metrics even on very small datasets.