Razpoznava drevesnih vrst iz oblakov točk z uporabo globokih nevronskih mrež

magistrsko delo

Vid Topolovec Klemenčič (Author), Štefan Kohek (Mentor), Damjan Strnad (Co-mentor)

Abstract

V magistrskem delu obravnavamo razpoznavo drevesnih vrst iz oblakov točk z uporabo novejših nevronskih mrež in primerjamo uspešnost razpoznave s sorodnimi deli. Razpoznavo smo zasnovali na dva različna načina. Pri prvem načinu smo izbrali in pripravili nevronsko mrežo za obdelavo 3D podatkov oz. oblakov točk, medtem ko smo pri drugem načinu izbrali in pripravili nevronsko mrežo za obdelavo 2D podatkov oz. slik. Pripravili smo tudi skupno podatkovno zbirko z združitvijo prosto dostopnih zbirk, ki vsebujejo posamezna drevesa v obliki oblakov točk, in med učenjem obogatili podatke. Po zaključenem učenju s pripravljeno skupno podatkovno zbirko nismo dosegli podobnih zaključkov kot v primerjanem predhodnem delu. V našem primeru je izbrana nevronska mreža, namenjena obdelavi 3D podatkov, dosegla 4 % višjo skupno točnost od izbrane nevronske mreže, ki je obdelovala 2D podatke.

Keywords

globoke nevronske mreže;klasifikacija drevesnih vrst;3D oblaki točk;2D slike;obogatitev podatkov;magistrske naloge;

Data

Language:	Slovenian
Year of publishing:	2024
Typology:	2.09 - Master's Thesis
Organization:	UM FERI - Faculty of Electrical Engineering and Computer Science
Publisher:	[V. Topolovec Klemenčič]
UDC:	004.93:004.032.26(043.2)
COBISS:	206147843
Views:	129
Downloads:	23
Average score:	0 (0 votes)
Metadata:

Other data

Secondary language:	English
Secondary title:	Tree species recognition from point clouds using deep neural networks
Secondary abstract:	In this thesis we discuss the classification of tree species from point clouds using more recent neural networks and compare the success of the classification with related work. We have divided the classification into two distinct parts. For the first part, we selected and prepared a neural network for processing 3D data or point clouds. For the second part, we selected and prepared a neural network for processing 2D data or images. We also created a common database by merging freely available databases containing individual trees in the form of point clouds and augmenting the data during the learning process. After completing the study with the prepared common database, we did not reach the same conclusions as in the related work. In our example, the neural network selected for 3D data processing achieved a 4 % higher overall accuracy than the neural network selected for 2D data processing.
Secondary keywords:	deep neural networks;tree species classification;3D point clouds;2D images;data augmentation;
Type (COBISS):	Master's thesis/paper
Thesis comment:	Univ. v Mariboru, Fak. za elektrotehniko, računalništvo in informatiko, Računalništvo in informacijske tehnologije
Pages:	1 spletni vir (1 datoteka PDF (XI, 91 f.))
ID:	23823041