|
|
Image Super-Resolution Using Deep Learning
Mojžiš, Tomáš ; Beran, Vítězslav (referee) ; Španěl, Michal (advisor)
The aim of this thesis is to create a deep neural net capable of super-resolution on images acquired by electron microscopes. The thesis consists of two parts - finding appropriate data and creating a dataset for the super-resolution task and designing a neural net architecture capable of solving the super-resolution task. Within the thesis, two datasets comprised of images acquired by electron microscopes were created. The datasets differ in the approach to data augmentation. They allow to train a neural network which fulfills the super-resolution task. To solve this task, two U-Net based and one GAN based architecture were trained. The resolution of images was upscaled by a factor of two and four. The best artificially upscaled images were created by neural network Real-ESRGAN. The values of metrics were not higher than the tested interpolation method, but the images seem more visually pleasing especially when they were upscaled four times. Thanks to this thesis, two datasets were created allowing to train other possible neural network architectures to improve the quality of the artificially upscaled images. The neural networks trained in this thesis can be utilized in the process of acquiring higher quality data from low resolution electron microscope images.
|
|
|
Recognition of Driving Lane Borders in Video from On-Board Camera
Fridrich, David ; Kohút, Jan (referee) ; Herout, Adam (advisor)
This paper talks about lane detection. Specifically custom generator of synthetic images, usage during training of neural networks, testing on convolutional neural network (CNN) UNet model and possibilities of extension of this model to SALMnet (Structure-Aware Lane Marking Detection Network) via addding SGCA module (semantic-guided channel attention) and PDC module (pyramid deformable convolution). Training results from synthetic datasets show very accurate results, reaching around 95\,\% in accuracy (even 99\,\% for easier images). Trainings with real datasets show lower accuracy, depending on the difficulty of the dataset itself. TuSimple has easier and clearer images and reaches about 62\,\%. CuLane is much more complex and results show accuracy around 37\,\%.
|
|
|
Analysis of Landscape Deforestation Using Satellite Imagery
Javorka, Martin ; Španěl, Michal (referee) ; Beran, Vítězslav (advisor)
Today is important to protect forest resources and tracking deforestation is essential. Re- mote sensing has an important role in this monitoring effort. This thesis studies four different techniques for detecting deforestation from satellite imagery - using both optical and radar data. The specifics of Earth observation data and geospatial analyses are described. The analytical techniques are used for detecting deforestation in the study area of Chocske vrchy. Image segmentation with Unet neural network model is used to classify there all deforested patches.
|
|
|
Image segmentation methods with limited data sets
Horečný, Peter ; Kolařík, Martin (referee) ; Burget, Radim (advisor)
The goal of this thesis was to propose an image segmentation method, which is capable of effective segmentation process with small datasets. Recently published ODE neural network was used for this method, because its features should provide better generalization in case of tasks with only small datasets available. The proposed ODE-UNet network was created by combining UNet architecture with ODE neural network, while using benefits of both networks. ODE-UNet reached following results on ISBI dataset: Rand: 0,950272 and Info: 0,978061. These results are better than the ones received from UNet model, which was also tested in this thesis, but it has been proven that state of the art can not be outperformed using ODE neural networks. However, the advantages of ODE neural network over tested UNet architecture and other methods were confirmed, and there is still a room for improvement by extending this method.
|
|
|
Deep learning methods for vessel and optic disc segmentation in ophthalmologic sequences
Rozhoňová, Andrea ; Odstrčilík, Jan (referee) ; Hesko, Branislav (advisor)
The aim of the following thesis was to study the issue of optical disc and retinal vessels segmentation in ophthalmologic sequences. The theoretical part of the thesis summarizes the principles of different approaches in the field of deep learning, which are used in connection with the given issue. Based on the theoretical part, methods for optical disk segmentation and retinal vessel segmentation based on the convolutional neural networks Linknet, PSPNet, Unet and MaskRCNN are proposed. The practical part of the thesis deals with the description of their implementation and subsequent evaluation.
|
|
|
Detection and localization of microbial colonies by means of deep learning algorithms
Čičatka, Michal ; Vičar, Tomáš (referee) ; Mézl, Martin (advisor)
Due to massive expansion of the mass spectrometry and constant price growth of the human labour the optimalisation of the microbial samples preparation comes into question. This master thesis deals with design and implementation of a machine learning algorithm for segmentation of images of microbial colonies cultivated on Petri dishes. This algorithm is going to be a part of a controlling software of a MBT Pathfinder device developed by the company Bruker s. r. o. that automates the process of smearing microbial colonies onto a MALDI target plates. In terms of this thesis a several models of neural networks based on the UNet, UNet++ and ENet architecture were implemented. Based on a number of experiments investigating various configurations of the networks and pre-processing of the training datatset there was chosen an ENet model with quadruplet filter count and additional convolutional block of the encoder trained on a dataset pre-processed with round mask.
|
|
|
Intracranial aneurysm detection in TOF-MRI data
Vývoda, Jan ; Vičar, Tomáš (referee) ; Jakubíček, Roman (advisor)
Práce obsahuje teoretický přehled informací o intrakraniálních aneurysmatech, jejich diagnostice a terapii. Dále shrnuje metody detekce objektů jak klasickými metodami, tak metodami strojového učení a také obsahuje stručný popis některých metod detekce intrakraniálních aneurysmat. V praktické části je vyhodnoceno a porovnáno několik navrhnutých postupů segmentace a detekce těchto aneurysmat pomocí neuronové sítě U-net.
|
|
|
Analysis of ultra-high frequency ECG using deep learning
Koščová, Zuzana ; Antin, Christoph Hoog (referee) ; Plešinger, Filip (advisor)
Analýza ultravysokofrekvenčného EKG (UHF-ECG) poskytuje informácie o elektrickej komorovej dyssynchrónii. Okrem toho analýza UHF-ECG v reálnom čase umožňuje priamu optimalizáciu stimulačnej elektródy počas implantácie kardiostimulátora. V tejto diplomovej práci opisujeme poruchy komorového vedenia, súčasnú metódu analýzy UHF-ECG a hlavne predstavujeme niekoľko modelov hlbokého učenia na to, aby sme zistili, ktoré kroky UHF-ECG analýzy môžu byť hlbokým učením nahradené. Dáta použité na vývoj a validáciu modelov hlbokého učenia pochádzajú z 2 súkromných nemocníc (FNUSA-ICRC, Brno, Česko, FNKV Praha, Česko) a z 3 verejne dostupných databáz. Najprv boli predstavené dve metódy hlbokého učenia na detekciu QRS komplexu a odhad trvania QRS komplexu v jednom kroku inferencie. Pri úlohe detekcie sme získali celkové F1-skóre 98,84 ± 0,51 \% a pri úlohe odhadu trvania QRS komplexu strednú absolútnu chybu (MAE) 12,25 ± 2,16 ms. Táto metóda zvyšuje výkonnosť analýzy UHF-ECG a vďaka tomu môže výrazne skrátiť čas merania. Okrem toho bol vyvinutý regresný model na odstraňovanie stimulačných impulzov založený na tzv. conditional generative adversarial networks. Výsledky boli vyhodnotené na základe korelácie 15 priemerných vysokofrekvenčných obálok v oblasti QRS komplexu medzi výstupom modelu a cieľovým signálom. Výsledky ukazujú vyššiu koreláciu na spontánnych signáloch a pokles korelácie so zvyšujúcim sa frekvenčným pásmom. Napokon boli vytvorené dva modely konvolučených neurónových sietí (CNN) na odhad komorovej elektrickej dyssynchrónie (VED). Konkrétne CNN s vrstavmi v 1D a 2D. MAE medzi naším riešením a anotáciou je 12,61 ±18,95 ms a 12,27 ±17,73 ms pre 1D a 2D CNN. MAE na spontánnych signáloch je pre oba modely približne o 5 ms nižšia ako na stimulovaných údajoch, čo naznačuje potrebu odstrániť stimulačné impulzy. Tieto modely hlbokého učenia prinášajú redukciu pipeline predspracovania a zároveň poskytujú výstup v jednom kroku inferencie. V prípade modelu detekcie QRS a odhadu trvania QRS je zlepšenie výkonu oproti súčasnému riešeniu evidentné a tieto kroky súšasnej analýzy UHF-ECG by mohli byť hlbokým učením nahradené. Avšak pre odstránenie stimulačných impulzov a odhad parametrov VED je potrebné zlepšiť výkon pre reálne použitie.
|
|
|
Image Super-Resolution Using Deep Learning
Mojžiš, Tomáš ; Beran, Vítězslav (referee) ; Španěl, Michal (advisor)
The aim of this thesis is to create a deep neural net capable of super-resolution on images acquired by electron microscopes. The thesis consists of two parts - finding appropriate data and creating a dataset for the super-resolution task and designing a neural net architecture capable of solving the super-resolution task. Within the thesis, two datasets comprised of images acquired by electron microscopes were created. The datasets differ in the approach to data augmentation. They allow to train a neural network which fulfills the super-resolution task. To solve this task, two U-Net based and one GAN based architecture were trained. The resolution of images was upscaled by a factor of two and four. The best artificially upscaled images were created by neural network Real-ESRGAN. The values of metrics were not higher than the tested interpolation method, but the images seem more visually pleasing especially when they were upscaled four times. Thanks to this thesis, two datasets were created allowing to train other possible neural network architectures to improve the quality of the artificially upscaled images. The neural networks trained in this thesis can be utilized in the process of acquiring higher quality data from low resolution electron microscope images.
|
|
|
Recognition of Driving Lane Borders in Video from On-Board Camera
Fridrich, David ; Kohút, Jan (referee) ; Herout, Adam (advisor)
This paper talks about lane detection. Specifically custom generator of synthetic images, usage during training of neural networks, testing on convolutional neural network (CNN) UNet model and possibilities of extension of this model to SALMnet (Structure-Aware Lane Marking Detection Network) via addding SGCA module (semantic-guided channel attention) and PDC module (pyramid deformable convolution). Training results from synthetic datasets show very accurate results, reaching around 95\,\% in accuracy (even 99\,\% for easier images). Trainings with real datasets show lower accuracy, depending on the difficulty of the dataset itself. TuSimple has easier and clearer images and reaches about 62\,\%. CuLane is much more complex and results show accuracy around 37\,\%.
|
guest ::
