Automated visual inspection in the industrial setup via deep learning

Semchyshyn, Pavlo

Домівка
→
Students Research & Project Works | Роботи студентів
→
Факультет прикладних наук
→
Бакалаврська програма "Комп'ютерні науки"
→
2023
→
Перегляд матеріалів

dc.contributor.author	Semchyshyn, Pavlo
dc.date.accessioned	2024-02-14T09:43:14Z
dc.date.available	2024-02-14T09:43:14Z
dc.date.issued	2023
dc.identifier.citation	Semchyshyn, Pavlo. Automated visual inspection in the industrial setup via deep learning / Pavlo Semchyshyn; Supervisor: Taras Firman; Ukrainian Catholic University, Department of Computer Sciences. – Lviv: 2023. – 41 p.: ill.	uk
dc.identifier.uri	https://er.ucu.edu.ua/handle/1/4419
dc.language.iso	en	uk
dc.title	Automated visual inspection in the industrial setup via deep learning	uk
dc.type	Preprint	uk
dc.status	Публікується вперше	uk
dc.description.abstracten	With the rise of Industry 4.0, much attention is attracted to the field of automated vi- sual inspection. Automation of the quality check in the production environment can reduce labor costs significantly, therefore, especially with the rise of deep learning- based algorithms, anomaly detection became one of the most researched topics in a machine learning community. Visual anomaly detection aims to detect inconsisten- cies in image data, which can be classified as anomalies. This can be used in many areas apart from manufacturing, including the detection of abnormal areas in med- ical imaging, surface inspection, or photo editing. This task becomes quite common when we have access only to normal samples as anomalies are rare compared to normal data and are usually hard to collect. Therefore, visual anomaly detection is usually solved in an unsupervised setting, where we take advantage of only normal data. One of the approaches to solving visual anomaly detection is image reconstruc- tion. Recently, diffusion models became state-of-the-art in the image generation task, being especially prominent in terms of image quality and diversity of the generated samples. In this study, we leverage diffusion models to the task of visual anomaly detection in a manufacturing setting, show its strengths and weaknesses as well as compare it with other existing methods and provide extensive benchmarks on the subject.	uk