|
|
QR code detection using deep learning
Černohous, Matěj ; Kříž, Petr (referee) ; Přinosil, Jiří (advisor)
This bachelor thesis deals with the design of an algorithm for detecting and decoding QR codes in images using deep learning techniques. The work involved the construction of 2 datasets, a YOLOv7 neural network model for detecting QR codes in images, a YOLOv4-tiny neural network model for detecting position markers of QR codes, and a Python program utilizing these models to read QR codes in images. For evaluation, the algorithm was compared with other options for QR code reading.
|
|
|
Prediction of radiotherapy response in rectal cancer by MR
Chmela, Radek ; Nohel, Michal (referee) ; Mézl, Martin (advisor)
This diploma thesis deals with the issue of predicting the response of rectal cancer to radiotherapy. The work is divided into four chapters. In the first two, the anatomy of the rectum, types of cancer and individual diagnostic methods are described, together with algorithms for detecting objects in images. In the third chapter, there is a description of the solution for automatic segmentation and prediction of the effectiveness of radiotherapy. In the fourth chapter, the achieved results are discussed.
|
|
|
Forensic analysis of handwriting for the Czech environment using artificial intelligence
Stejskal, Jan ; Přinosil, Jiří (referee) ; Burget, Radim (advisor)
The analysis of handwriting is an important area of research in modern science. However, it is a very complex process because handwritten text can take on various forms. The use of artificial intelligence for analyzing and identifying text from different authors is nothing new in the world. Research in this area is, however, slightly lagging behind in the Czech environment. For this reason, several convolutional network architectures were proposed and compared in this work in an effort to find the most suitable structure for solving this problem. Of all the trained and tested models, the model based on the ResNet18 architecture achieved the highest accuracy, with a success rate of 92.2 % on a self-made database of 1328 samples with a resolution of 750x256. This result suggests that with a sufficiently large and high-quality database, the problem can be solved even in the Czech environment with its more complicated character set.
|
|
|
Self-Supervised Learning for Recognition of Sports Poses in Image
Konečný, Daniel ; Beran, Vítězslav (referee) ; Herout, Adam (advisor)
Cílem této práce je rozpoznání sportovních pozic v obrazových datech za pomocí přístupu self-supervised learning pro docílení vyšší úspěšnosti klasifikace s použitím malého množství anotovaných vzorků. Učení za pomocí self-supervision je docíleno snímky stejné scény z různých úhlů ve stejných a různých časech. Konvoluční neuronová síť naučená s pomocí funkce triplet loss zakóduje sportovní pozice do latentních vektorů a plně propojená neuronová síť tyto vektory klasifikuje. Model natrénovaný pomocí self-supervised learning dosahuje o 30-40 % vyšší úspěšnosti než supervised model, když je trénovaný pouze na desítkách či jednotkách označených snímků z každé třídy. Hlavními přínosy této práce jsou nástroje pro přípravu datové sady pro tento specifický typ učení, dvě datové sady s více anotacemi a implementované modely využívající self-supervised learning. Výsledky ukazují, že učení za pomocí self-supervision je vhodný přístup pro řešení klasifikace za použití velmi malého množství označených snímků.
|
|
|
Exploitation of Neural Networks for Fusion of Image and Non-Image Data
Reich, Bořek ; Maršík, Lukáš (referee) ; Zemčík, Pavel (advisor)
This master thesis uses convolutional neural networks to fuse image and non-image data. Both deep learning detection systems that rely only on image data (images from the camera) and that use both image and non-image data (images from the camera and data from the millimeter-wave radar) are studied in this thesis. A unique dataset containing raw millimeter-wave radar data and corresponding time-synchronized images from the camera was created for the purpose of comparing these two types of methods (data fusion methods and methods that utilize only image data). Furthermore, a time synchronization method for millimeter-wave radar and cameras using only off-the-shelf hardware is proposed. Finally, the created dataset is used to verify the detection capability of the system that uses only camera data and the fusion system that uses both millimeter-wave radar and camera data.
|
|
| |
|
|
Exploitation of Machine Learning for Identification of Feeder Rod Movement
Vele, Patrik ; Vašíček, Zdeněk (referee) ; Šimek, Václav (advisor)
The aim of this diploma thesis is to create a device that uses machine learning methods to recognize the movements of a feeder fishing rod based on data from an inertial measurement unit. The introductory part is devoted to the feeder fishing technique, the selection of important movements and the possibilities of attaching the detection device to the rod. This is followed by the creation of a theoretical basis in the field of machine learning, familiarization with the inertial measurement unit and the issue of classification. The acquired knowledge is used to select appropriate techniques for solving the task of recognizing the movements of the rod. In the practical part, a detection device based on the ESP32 platform is designed and created. This is initially used as a motion sensor, which, in combination with the processing of the measured values, serves as a generator of a training data set. The work continues with the implementation of the convolutional neural network, the learning process on the created dataset and the integration of the most successful model into the detection device. The conclusion is devoted to testing in practice, evaluation and possibilities of future development. The result is a small, battery-powered device that, when attached to any feeder rod, provides highly successful detection of all key movements during the hunt. In addition, thanks to wireless communication via ESP-NOW, it is possible to send the results to various devices.
|
|
|
Computer Vision for Monitoring of 3D Printing
Heinz, Mikuláš ; Hradiš, Michal (referee) ; Smrž, Pavel (advisor)
This thesis deals with the automatic detection of errors that can occur during time-consuming 3D printing. It uses computer vision and artificial intelligence to achieve this. The main result is a system that uses Raspberry Pi and a connected camera to periodically record the printing process and sends the images to the user's computer for detection. On this computer, the image is analysed by a convolutional neural network model and information about found error is sent to the user via a SMTP protocol. The solution also includes a dataset with 385 images of 3D printing errors sorted by type.
|
|
|
Development of a universal cell for optical part inspection in a robotic workplace
Cahlík, Radim ; Rajchl, Matej (referee) ; Adámek, Roman (advisor)
Cílem této diplomové práce je vývoj univerzální buňky pro optickou kontrolu včetně softwaru pro detekci vad, který používá metody umělé inteligence. Dalším cílem je integrovat buňku do robotického pracoviště a otestovat ji na rozdílných dílech s různými vadami. První část práce popisuje vývoj konceptu optické buňky a její návrh. Poté pojednává o systému pro optickou kontrolu, počínaje popisem objektivu a kamery a konče analýzou kontrolního softwaru, který pro detekci vad využívá konvoluční neuronovou síť. Další část se zabývá vývojem robotického pracoviště včetně návrhu dvojosého robota a komunikace mezi zařízeními. Nakonec je rozebráno testování na dvou rozdílných dílech s cílem ověřit funkčnost.
|
|
|
Bearing diagnostics using machine learning
Zonygová, Kristýna ; Marada, Tomáš (referee) ; Zuth, Daniel (advisor)
The Master's thesis deals with the use of artificial intelligence methods in order to classify bearing failures. The SVC (Support Vector Classification), KNN (K-Nearest Neighbors Classifier), RFC (Random Forest Classifier) and CNN (Convolutional Neural Network) classification methods are described and tested on ball-bearing vibration signals from two different datasets. All methods achieve quite well accuracy (from 94.1 % to 99.8 %). Scripts in the Python programming environment that use libraries with free-licenses are also included. They provide the possibility of training classification methods (SVC, KNN, RFC or CNN) on your own data, or the use of already trained models.
|
guest ::
