|
|
Application-based Anomalous Communication Detection
Dostál, Michal ; Homoliak, Ivan (referee) ; Očenášek, Pavel (advisor)
This bachelor thesis deals with the analysis, design and implementation of a system for detecting anomalous network communication activities using high-level characteristics. The thesis contains a theoretical basis for the detection of anomalies using countries, autonomous systems and applications that are used to communicate. It also contains information about the techniques and methods of machine learning needed for implementation. The practical part describes the design, use and implementation of individual technologies. The result of this work is detection based on multiple machine learning methods, mostly classification.
|
|
|
Brno Communication Agent
Křištof, Jiří ; Fajčík, Martin (referee) ; Smrž, Pavel (advisor)
The aim of this thesis is the implementation of a communication agent, which provides information about Brno. The communication agent uses three - tier architecture . For the question answering , machine learning and neural network techniques are used . User tests determined the success rate 84 %. 58 % of the primary users were satisfied with the system. Main benefit of the work is facilitating the retrieving of information about Brno for its residents and visitors .
|
|
|
Object Detection Networks For Localization And Classification Of Intracranial Hemorrhages
Nemcek, Jakub
Intracranial hemorrhages represent life-threatening brain injuries. This paper presents twostate-of-the-art object detection systems (Faster R-CNN and YOLO v2) which are trained to localizeand classify hemorrhages in axial head CT slices by providing labelled rectangular bounding boxes.Publicly available datasets of head CT data and ground truth bounding boxes are used to evaluate andcompare the performance of both detectors. The Faster R-CNN shows better results by achieving anaverage Jaccard coefficient of 58.7 %.
|
|
|
Rf Transmitter Classification Based On Front-End Impairments
Youssefová, Kristina
This paper focuses on classifying Radio Frequency transmitters depending on their Radiofrequencyimperfections using a machine learning algorithm. The theoretical part of the papercan be divided into two branches. In the first branch, possible imperfections in the radio frequencytransmitters are presented. In the second one, the support vector machine algorithm is explained.The practical part deals with the implementation of support vector machines in the MATLAB programand the evaluation of results.
|
|
|
Traffic Sign Classification Using Deep Learning
Sicha, Marek
The thesis focuses on the classification of traffic signs in images and video sequences.The goal is real-time processing and usage of software in the vehicle. Neural networks and thePython programming language were chosen to solve the problem. To solve the problem a machinelearning method was chosen, more precisely a convolutional neural network. A neural network inthe Python programming language was created for the classification of traffic signs, using the Kerasand Tensorflow libraries. The neural network architecture is chosen for optimization for use on asingle-board computer with limited performance.
|
|
|
Multilingual Analysis Of Hypokinetic Dysarthria In Patients With Parkinson’s Disease
Kováč, Daniel
This article deals with the multilingual analysis of hypokinetic dysarthria (HD) in patientswith Parkinson’s disease (PD). The goal is to identify acoustic features that have high discriminationpower and that are independent of the language of a speaker. The speech corpus contains 59 PD patientsand 44 healthy controls (HC) speaking in Czech (cs) and American English (en-US). Based onnon-parametric statistical tests and logistic regression, we observed the best discrimination power hasthe speech index of rhythmicity (extracted from a reading text) and harmonic-to-noise ratio (extractedfrom a sustained vowel). We were able to identify PD with 67% sensitivity and 79% specificity inthe Czech corpus and with 78% sensitivity and 67% specificity in the English one. The performanceof the model was significantly lower when combining both datasets, thus suggesting language playsa significant role during the automatic assessment of HD.
|
|
|
Evaluation Of Cnn And Cldnn Architectures On Radio Modulation Datasets
Pijáčková, Kristýna
This paper presents an evaluation of deep learning architectures designed for modulationrecognition. The evaluation inspects, whether the architectures behave in the same way as they didon the dataset they were designed on. The architectures are trained and tested on two different radiomodulation datasets. This results in proposing additional binary classification as a method to reducemisclassification of QAM modulation types in one of the datasets.
|
|
|
Detection Of Intracranial Haemorrhages In Head Ct Data Based On Deep Learning
Nemček, Jakub
In this paper, we present a method for detection of intracranial haemorrhages in the head CT data using convolutional neural networks. We introduce three 2D image classifiers that perform in three perpendicular anatomical planes and classify the CT slices into healthy or pathological, whereby they provide the information about the position of the haemorrhage in the 3D CT image. The accuracies of the three models are 90.19%, 88.15%, and 80.90% for the axial, sagittal and coronal plane.
|
|
|
Active Upper Limb Prosthesis
Brázdil, Štěpán
This work focuses on the field of prosthetics, especially the issue of active prosthesis control. The goal is to perform a comprehensive analysis, design and construction of a mobile prosthetic system that, based on the analysis of sensory data (such as an EMG signal from multiple channels), can correctly classify a gesture and subsequent moving of the limb model. The measurements and test are executed on the real constructed system, which consists of the InMoov robotic hand, the Raspberry Pi controller and accessories. The results are critically evaluated, and future improvements are discussed.
|
|
|
Segmentation Of Ribs In Thoracic Ct Scans
Kašík, Ondřej
This paper deals with rib segmentation in thoracic CT data. For the segmentation method of rib centerlines detection is chosen. The first step of this approach is to extract the centerlines of all the bones located in the scan. These centerlines are divided into short primitives, which are subsequently classified into couple of categories, depending on whether they represent the centerline of the rib. More than 95% of all primitives are classified correctly. In the last step, the rib centrelines are used as the seed points of the region growing algorithm in three-dimensional space.
|
guest ::
