| | SLO | ENG | Cookies and privacy

Bigger font | Smaller font
First page > Show document

Show document Help

Title:Metoda za izboljšanje prostorsko-časovne ločljivosti okoljskih geoprostorskih podatkov z uporabo lokalnih meritev in satelitskih slik : doktorska disertacija
Authors:ID Cukjati, Jernej (Author)
ID Žalik, Borut (Mentor) More about this mentor... New window
Files:.pdf DOK_Cukjati_Jernej_2023.pdf (5,61 MB)
MD5: 0AE7DA2DB162EBE7AFBA7A4D737A8F52
 
Language:Slovenian
Work type:Doctoral dissertation
Typology:2.08 - Doctoral Dissertation
Organization:FERI - Faculty of Electrical Engineering and Computer Science
Abstract:V doktorski disertaciji predstavimo novo metodo za izboljšavo prostorsko-časovne ločljivosti okoljskih geoprostorskih podatkov. Geoprostorske podatke pogosto dobimo tudi iz meritev, ki jih zajamemo z lokalnimi ali s satelitskimi senzorji. Pomanjkljivost teh zajemov so redki lokalni senzorji in nizka časovna ločljivost satelitskih slik. Prostorsko in časovno ločljivost izboljšamo s souporabo podatkov iz meritev obeh podatkovnih virov. Najprej opazovano območje razdelimo v mrežo pikslov in nad lokalnimi senzorji sestavimo Voronoijev diagram. Središča Voronoijevih celic ustrezajo lokacijam lokalnih senzorjev, ki v danem časovnem trenutku vračajo veljavne izmerjene vrednosti. Za nabor pikslov znotraj vsake posamezne Voronoijeve celice zgradimo ločene regresijske modele s strojnim učenjem. Razlagalne spremenljivke regresijskih modelov so pretekli podatki iz meritev lokalnih senzorjev trenutno izbrane Voronoijeve celice in njenih sosed, ciljne vrednosti pa so iz izbranega nabora pikslov satelitskih slik. Po izračunu vrednosti okoljske spremenljivke v vseh pikslih na opazovanem območju dobimo geolocirano rastrsko sliko okoljske spremenljivke. Predlagano metodo smo uporabili na podatkih meritev lokalnih senzorjev in satelitskih slik toplogrednega plina NO2. Regresijske modele smo zgradili s tremi metodami: algoritmom najbližjih sosedov, linearno regresijo in večplastno naprej usmerjeno nevronsko mrežo. Najvišjo točnost smo dosegli z nevronsko mrežo. Predlagano metodo smo primerjali s petimi referenčnimi metodami, ki so bile predstavljene v zadnjih treh letih. Te metode so: geografsko-časovno obtežena nevronska mreža, prilagodljiva grafovska konvolucijska povratna nevronska mreža, časovna grafovska konvolucijska nevronska mreža z mehanizmom pozornosti, nevronska mreža za izmenjevanje sporočil, združena z mrežami dolgega kratkoročnega spomina, in globoko ansambelsko strojno učenje. Po točnosti je najboljše rezultate dala naša metoda. Najbolj se nam je približala metoda, sestavljena iz nevronske mreže za izmenjavo sporočil in nevronske mreže z dolgim kratkoročnim spominom. Od te smo bili točnejši za približno 5 %.
Keywords:računalništvo, strojno učenje, k-najbližji sosedje, linearna regresija, naprej usmerjena nevronska mreža, daljinsko zaznavanje
Place of publishing:Maribor
Place of performance:Maribor
Publisher:[J. Cukjati]
Year of publishing:2023
Number of pages:XI, 86 str.
PID:20.500.12556/DKUM-84443 New window
UDC:004.85:004.932:528.83(043.3)
COBISS.SI-ID:166705667 New window
Publication date in DKUM:02.10.2023
Views:690
Downloads:133
Metadata:XML DC-XML DC-RDF
Categories:KTFMB - FERI
:
Copy citation
  
Average score:(0 votes)
Your score:Voting is allowed only for logged in users.
Share:Bookmark and Share



Hover the mouse pointer over a document title to show the abstract or click on the title to get all document metadata.

Licences

License:CC BY-NC-ND 4.0, Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International
Link:http://creativecommons.org/licenses/by-nc-nd/4.0/
Description:The most restrictive Creative Commons license. This only allows people to download and share the work for no commercial gain and for no other purposes.
Licensing start date:08.06.2023

Secondary language

Language:English
Title:Improving the spatiotemporal resolution of geospatial data with the integration of the Internet of Things and remote sensing data
Abstract:This Doctoral thesis introduces a novel approach, which increases the spatio-temporal resolution of the environmental geospatial data. Geospatial data can also be obtained from the Internet of Things (IoT) measurements and satellite images. However, their main drawbacks are the sparsity of IoT sensors and low temporal resolution of the satellite images. We combine the data from both measurement sources to address this issue. Firstly, the observed area is gridded and a Voronoi diagram is constructed. The Voronoi centers correspond to the location of the IoT sensors which are returning the valid measurements in the considered time. Secondly, a separate regression model is constructed for each set of pixels inside each Voronoi cell with machine learning. The regression model uses the measurements from the central and neighboring IoT sensors as explanatory variables. On the other hand, the measurement data from the satellite pixels, located inside the considered Voronoi cell are, used as the target values. A satellite-like image is constructed after determining the value for each pixel in the considered area. The proposed approach was used to assess the concentration of NO2 and tested with three different methods: k-nearest neighbors, linear regression, and a multilayered feed-forward neural network. The highest accuracy was obtained from the regression models built by the multilayered feed-forward neural network. The proposed method was compared with five similar recent methods: Geographically and Temporally Weighted Generalized Regression Neural Network, Adaptive Graph Convolutional Recurrent Network, Attention Temporal Graph Convolutional Network, Message Passing Neural Networks with Long Short-Term Memory, and the Deep Ensemble Machine Learning framework. Our method, with the usage of the feed-forward neural network, performed the best, while the second most accurate method consisted of Message Passing Neural Networks and Long Short-Term Memory. With this method we were approximately 5 % more accurate.
Keywords:computer science, machine learning, k-nearest neighbors, linear regression, feedforward neural network, remote sensing


Comments

Leave comment

You must log in to leave a comment.

Comments (0)
0 - 0 / 0
 
There are no comments!

Back
Logos of partners University of Maribor University of Ljubljana University of Primorska University of Nova Gorica