Samodejna segmentacija satelitskih slik na podlagi šumnih oznak

Šuler, Kristijan (author), Čehovin Zajc, Luka (mentor), Račič, Matej (comentor)

V delu preučujemo vpliv šuma v oznakah na samodejno razpoznavo področij v satelitskih slikah. Pridobivanje oznak je na tem področju namreč izziv. Veliko jih je pridobljenih iz virov, ki niso usklajeni s slikovnimi podatki. Prihaja do prostorskih odstopanj ter zamenjave razredov posameznih območij. V študiji obravnavamo več uveljavljenih metod strojnega učenja, ki smo jih preizkusili na različnih vrstah šuma, ki je lahko prisoten pri oznakah satelitskih slik. Posebej se osredotočimo na metode globokega učenja, ki dosegajo dobre rezultate v računalniškem vidu. Te metode so do neke mere že robustne na šum v oznakah, dodatno pa preizkusimo tudi ogrodje DivideMix, ki je narejeno prav za učenje na šumnih podatkih. Vpliv šuma eksperimentalno ovrednotimo na realnem problemu določanja dejanske rabe kmetijskih in gozdnih zemljišč v Republiki Sloveniji. Rezultati študije kažejo, da so metode globokega učenja robustne na nizke do srednje vrednosti šuma v oznakah. Kadar pa je šuma v oznakah veliko, lahko z ogrodjem DivideMix dosežemo izboljšanje. Obenem so se za zelo robustne izkazale tudi klasične metode strojnega učenja.

• Language: Slovenian
• Keywords: računalniški vid, šumne oznake, satelitske slike, samodejna segmentacija
• Work type: Master's thesis/paper
• Typology: 2.09 - Master's Thesis
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2024
• PID: 20.500.12556/RUL-166035
• COBISS.SI-ID: 220047107
• Publication date in RUL: 18.12.2024

ŠULER, Kristijan, 2024, Samodejna segmentacija satelitskih slik na podlagi šumnih oznak [online]. Master’s thesis. [Accessed 5 April 2025]. Retrieved from: https://repozitorij.uni-lj.si/IzpisGradiva.php?lang=eng&id=166035

Secondary language

• Language: English
• Title: Automatic segmentation of satellite images using noisy labels

Abstract

In this work, we examine the impact of label noise on the actual recognition of areas in satellite images. Acquiring labels in this field is challenging, as many of the labels are obtained from sources which are not aligned with the image data. Spatial deviations and misclassifications of specific regions also occur. In this thesis, we discuss several established machine learning methods, which were then tested on the different types of noise that can be present in satellite image labels, with an in-depth focus on those deep learning methods that achieve satisfactory results in computer vision. These methods are already more or less robust when it comes to label noise. Additionally, we tested the DivideMix framework, which is specifically designed for learning from noisy data. The impact of noise is experimentally evaluated on the real problem of determining the actual use of agricultural and forest land in the Republic of Slovenia. The results of this thesis show that deep learning methods are robust to low to medium levels of label noise. However, when the level of label noise is high, the DivideMix framework can be used to improve results. Next to that, classical machine learning methods have also proven to be very robust.

• Keywords: computer vision, noisy labels, satellite images, automatic segmentation