Institutional Repository of Vilnius University
Deep learning application for population forecasting using aerial imagery and data compression on local hardware /
Abstract
dc:descriptionThe main goal of this study is to identify the best data compression algorithm out of the three chosen for compressing raw GeoTIFF aerial images, which are then used for deep learning model training for population forecasting on two local computers, depending on various metrics collected, energy consumed and carbon emitted. Three different data compression algorithms were chosen and used Deflate and Zstandard (ZSTD) both being lossless, and JPEG, being the only lossy algorithm in the chosen stack. To achieve this goal, a thorough analysis of the research papers and articles on topics related to aerial imagery, data compression, CNN deep learning architecture, and local computer hardware was done in order to gather information required to properly overview the methodologies used for conducting the experiment. Next, the methodologies and testing scenarios were created, where the aerial imagery, data compression algorithms, deep learning model chosen and local hardware used, were specified and described. Aster conducting multiple testing scenarios on two different local computers and collecting valuable compression and training data together with energy consumption, the results of the experiment were then thoroughly analyzed and compared according to the various system metrics and other useful data collected described in the paper. Each algorithm’s performance metrics in both raw GeoTIFF aerial image compression and deep learning model training are visualized for an easier comparison. The visualizations and further collected data analysis indicated that the lossy nature of the JPEG data compression algorithm did not hinder the deep learning model’s performance and was the most superior out of the three data compression algorithms chosen for the population forecasting task described in the paper.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2025
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Brasiūnas, Lukas,
Subjects
dc:subject × 1Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- eng
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD239029919&prefLang=en_US