|
|
DEEP LEARNING FOR SINGLE-VOXEL AND MULTIDIMENSIONAL MR-SPECTROSCOPIC SIGNAL QUANTIFICATION, AND ITS COMPARISON WITH NONLINEAR LEAST-SQUARES FITTING
Shamaei, Amirmohammad ; Latta,, Peter (oponent) ; Kozubek, Michal (oponent) ; Jiřík, Radovan (vedoucí práce)
Preprocessing, analysis, and quantification of Magnetic resonance spectroscopy (MRS) signals are required for obtaining the metabolite concentrations of the tissue under investigation. However, a fast, accurate, and efficient post-acquisition workflow (preprocessing, analysis, and quantification) of MRS is challenging. This thesis introduces novel deep learning (DL)-based approaches for preprocessing, analysis, and quantification of MRS data. The proposed methods achieved the objectives of robust data preprocessing, fast and efficient MR spectra quantification, in-vivo concentration quantification, and the uncertainty estimation of quantification. The results showed that the proposed approaches significantly improved the speed of MRS signal preprocessing and quantification in a self-supervised manner. Our proposed methods showed comparable results with the traditional methods in terms of accuracy. Furthermore, a standard data format was introduced to facilitate data sharing among research groups for artificial intelligence applications. The findings of this study suggest that the proposed DL-based approaches have the potential to improve the accuracy and efficiency of MRS for medical diagnosis. The dissertation is structured into four parts: an introduction, a review of state-of-the-art research, a summary of the aims and objectives, and a collection of publications that showcase the author's contribution to the field of DL applications in MRS.
|
|
|
Interpretace učebního procesu klasifikátoru fibrilace síně
Lichtblauová, Anna ; Ředina, Richard (oponent) ; Novotná, Petra (vedoucí práce)
V rámci teoretické části bakalářské práce byla nastudována problematika detekce fibrilace síní (FS) a práce s konvolučními neuronovými sítěmi (CNN). Následně byly v praktické části vytvořeny dva klasifikátory. První byl určen pro klasifikaci sinusového rytmu, fibrilace síní a ostatních patologií, druhý dále rozlišoval kategorii "fibrilace síní" podle skutečnosti, zda se nacházela v celém záznamu nebo jen v jeho části. Výsledná přesnost byla 82.12~\% v případě prvního, resp. 85.14 \% v případě druhého klasifikátoru.
|
|
|
Device for the traffic situation evaluating
Gábel, Matej ; Honec, Peter (oponent) ; Janáková, Ilona (vedoucí práce)
The bachelor's thesis deals with the implementation of a device for evaluating the traffic situation, specifically by traffic signs detection. In this work, I tried different methods of traffic sign recognition, where the resulting implementation on hardware is done using convolutional neural networks. More precisely, it is the YOLOv5 architecture, which is suitable for recognizing traffic signs in real time.
|
|
|
Detekce QR kódů pomocí hlubokého učení
Černohous, Matěj ; Kříž, Petr (oponent) ; Přinosil, Jiří (vedoucí práce)
Tato bakalářská práce se zabývá návrhem algoritmu pro detekci a dekódování QR kódů v obrazech využitím technik hlubokého učení. V rámci práce byly zhotoveny 2 datové sady, model neuronové sítě YOLOv7 pro detekci QR kódu v obraze, model neuronové sítě YOLOv4-tiny pro detekci pozičních bodů QR kódu a program v jazyce Python využívající těchto modelů pro čtení QR kódů v obrazech. Pro vyhodnocení byl algoritmus porovnán s jinými řešeními čtení QR kódů.
|
|
|
Konvoluční neuronová síť pro segmentaci obrazu
Mitrenga, Michal ; Petyovský, Petr (oponent) ; Jirsík, Václav (vedoucí práce)
Cílem bakalářské práce je seznámit se s problematikou konvolučních neuronových sítí a realizovat segmentaci obrazu. Toto téma v sobě zahrnuje obor počítačového vidění, který je používán v systémech s umělou inteligencí. Zvláštní pozornost je věnována procesu segmentace obrazu. Dále se práce věnuje základním principům umělých neuronových sítí, struktuře konvolučních neuronových sítí a zejména pak popisu jednotlivých architektur sémantické segmentace. Vybraná architektura SegNet je použiti v praktické aplikaci spolu s před-učenou sítí. Součástí práce je databáze obrazů CamVid, která je použita pro trénování. Pro testování je vytvořena databáze vlastních snímků. Praktická část je zaměřená na trénování CNN a hledání nevhodnějších parametrů pro učení sítě za pomocí SW Matlab.
|
|
|
Development of a universal cell for optical part inspection in a robotic workplace
Cahlík, Radim ; Rajchl, Matej (oponent) ; Adámek, Roman (vedoucí práce)
The aim of this master's thesis is to develop a universal cell for optical part inspection including defects detection software that utilises artificial intelligence methods. Another goal is to integrate the cell into a robotic workplace and to test it on distinct parts with different defects. The first part of the thesis describes the development of the optical cell concept and its design. After that, the optical inspection system is covered, starting with a description of the lens and camera, followed by the analysis of the inspection software, which uses the convolutional neural network for defect detection. The next part deals with the development of a robotic workplace including the design of a two degrees of freedom robot and the communication between the devices. In the end, testing on two distinct parts to verify its functioning is discussed.
|
|
|
Defect detection on fiber materials using machine learning
Lang, Matěj ; Richter, Miloslav (oponent) ; Honec, Peter (vedoucí práce)
Following Master's Thesis is presenting the creation of scanning unit, that will automate the quality check process in company SILON, for their fibered material. The process of manual detection is discussed and the automated solution is proposed. Several test are shown, that demonstrate effects of different lights on fibres dyed in Rhodamine. Optimal filter for camera is chosen, to achieve images with highest resolution possible and with enough definition. Next, the software tools for hardware control are presented and tools for building neural networks. Also, some basic info on current state of the art is provided, to explain some of the tools used. The network itself is shown and also its learning process and capabilities of defect detection.
|
|
|
Image annotation using deep learning
Zarapina, Natalya ; Rajnoha, Martin (oponent) ; Burget, Radim (vedoucí práce)
This semester thesis describes the design and implementation of the client-server program for classification and localization of certain elements which are present in provided images. This program loads a set of images and use deep learning, especially deep convolution neural network perform a classification. First part describes the architecture, basic principles of operations in convolution network and chosen machine learning algorithms for classification. Second part contains a description of created program.
|
|
|
Depth-Based Determination of a 3D Hand Position
Ondris, Ladislav ; Tinka, Jan (oponent) ; Drahanský, Martin (vedoucí práce)
This work aims to offer a real-time, depth-based gesture recognition system using a hand's skeletal information. The Tiny YOLOv3 neural network detects the hand in the depth image. The detected hand is rid of the background and used by the JGR-P2O neural network, which estimates the hand's skeleton represented by 21 key points. Furthermore, a novel technique for gesture recognition from hand key points that compares the input skeleton with user-defined gestures has been proposed. A dataset consisting of four thousand images was captured to evaluate the system.
|
|
|
Named Entity Recognition Exploiting Sub Word Information
Dobrovodský, Patrik ; Egorova, Ekaterina (oponent) ; Kesiraju, Santosh (vedoucí práce)
The aim of this thesis is the creation of a Named Entity Recognition system based on an older state-of-the-art model and studying how subword information can improve the recognition of out-of-vocabulary words. This proposed system besides English has to support two additional Indo-European languages: German and Hungarian. This work features a named entity tagger based on deep learning using pretrained and custom-trained word embeddings, sparse features, and character embeddings extracted by a Convolutional Neural Network. All these features are then processed by sequence-based (bidirectional Long Short-Term Memory) and feature-based (Conditional Random Field) approaches with the goal of achieving a F1-score similar to the work it is based on, and to compare how far present time state-of-the-art systems have evolved. The result is a system that achieves a 90.98% F1-score on the CoNLL 2003 English test dataset using pretrained word embeddings, not far behind the original work's 91.26%. For the other two languages, the model scores 89.34% on the WikiAnn German test dataset and 93.04% on the WikiAnn Hungarian test dataset with the usage of custom-trained embeddings.
|
host ::
RSS.