Implementation of convolutional neural networks in accelerating units for real-time image analysis

Eldar Šabanovič

Doctoral dissertation

Dissertations are not being sold

Quantity

Description

The visible spectrum image analysis enables intelligent systems to derive in­for­mation as human vision. Image analysis based on Convolutional Neural Net­works (CNN) is used in many areas. The application of locally executed CNN algorithms in embedded systems still limited due to the high amount of required calculations. Currently, there are useful accelerating units for CNN calculations in embedded systems, but a list of useful and supported CNN elements and meth­odology for the creation of specialized CNN structures are needed for implementa­tion.

The object of work research is convolutional neural networks for real-time image analysis. The following main aspects of the research object are investigated in the present thesis: implementation in accelerating units and design methodol­ogy.

The thesis presents the implementation of convolutional neural networks for image analysis based on the proposed list of CNN elements and the methodology to solve selected tasks of image analysis. There are two application areas of CNN‑based image analysis algorithms, which have been presented in this work. The first area is the local future description of the retinal images, and the second one is the evaluation of the road type and conditions using images received from the camera in front of the car.

The main results of the thesis were published in seven scientific publica­tions – three in peer-reviewed scientific jour­nals, four in conference proceedings: one in journal, referenced by the Clarivate Analytics Web of Science database, with a citation index of 1.524, one in journal, referenced in the Clarivate Analytics Web of Science database, without citation index, one in journal referenced in Index Copernicus database, three articles in international conference proceedings, referenced in the Clarivate Analytics Web of Science Proceedings database, one article in conference pro­ceedings, referenced in other databases. The results of the research that have been defined in the dissertation have been presented at nine scientific conferences in Lithuania and abroad.

The dissertation consists of an introduction, three chapters, general conclu­sions, references, and three annexes.

Electronic version of the book:

DOI: https://doi.org/10.20334/2019-051-M

Book details

Data sheet

Year:
2019
ISBN:
978-609-476-204-8
Imprint No:
2019-051-M
Dimensions:
160×240
Pages:
130 p.
Cover:
Softcover
Language:
Lithuanian
16 other books in the same category:

Follow us on Facebook