Přístupnostní navigace
E-application
Search Search Close
Bachelor's Thesis
Author of thesis: Bc. Josef Hruška
Acad. year: 2024/2025
Supervisor: doc. Ing. Pavel Škrabánek, Ph.D.
Reviewer: Ing. Jan Turčínek, Ph.D.
Unsupervised deep learning methods can detect anomalies by learning patterns exclusively from normal, non-anomalous data. In industry, where anomalous data are rare, this approach is particularly promising for visual defect detection on manufactured parts. This thesis provides an overview of image anomaly detection methods using unsupervised deep learning, with the primary goal of developing a custom solution for detecting defects on a mass-produced plastic pre-filter. To ensure robust training and evaluation of the models, an extensive dataset of both normal and anomalous pre-filter images is collected. The anomalous images are then annotated to create binary masks for pixel-level evaluation. Two models are developed. The first model is an author's custom autoencoder based on direct image reconstruction, and second is a modified feature reconstruction model built on an existing architecture. The accuracy of the models is evaluated both at the image classification and at the pixel level. The best performing model is implemented for further experimental testing.
Unsupervised deep learning, visual quality inspection, image anomaly detection
Date of defence
18.06.2025
Result of the defence
Defended (thesis was successfully defended)
Grading
A
Process of defence
Student seznámil komisi s výsledky své bakalářské práce a odpověděl na otázky oponenta. Následovala diskuze k závěrečné práci a otázky komise: jak se v procesu kontrolují svary? Jak jste řešil vliv osvětlení? Jaký má montážní linka takt? Student odpověděl na všechny dotazy komise.
Language of thesis
English
Faculty
Fakulta strojního inženýrství
Department
Institute of Automation and Computer Science
Study programme
Fundamentals of Mechanical Engineering (B-ZSI-P)
Specialization
Fundamentals of Mechanical Engineering (STI)
Composition of Committee
prof. Ing. Vladimír Vašek, CSc. (předseda) Ing. Jan Turčínek, Ph.D. (člen) Ing. Petr Hadraba, Ph.D. (člen) Mgr. Monika Dosoudilová, Ph.D. (člen) Ing. Ondřej Rubeš, Ph.D. (člen) Ing. Luděk Janák (člen) Ing. Roman Parák, Ph.D. (člen) doc. Ing. Pavel Škrabánek, Ph.D. (místopředseda) doc. Ing. Ivan Švarc, CSc. (člen) Ing. et Ing. Stanislav Lang, Ph.D. (člen)
Supervisor’s reportdoc. Ing. Pavel Škrabánek, Ph.D.
Grade proposed by supervisor: A
Reviewer’s reportIng. Jan Turčínek, Ph.D.
Grade proposed by reviewer: A
Responsibility: Mgr. et Mgr. Hana Odstrčilová