Full Text Available
Note: Clicking the button above will open the full text document at the original institutional repository in a new window.
Rail surface anomaly detection : a deep learning . approach for computer vision
Dissertation (MEng (Mechanical Engineering))--University of Pretoria, 2018.
Saved in:
| Other Authors: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2026
|
| Subjects: | |
| Tags: |
No Tags, Be the first to tag this record!
|
| _version_ | 1867613707658330112 |
|---|---|
| access_status_str | Open Access |
| author2 | Heyns, P.S. (Philippus Stephanus) |
| author_browse | Heyns, P.S. (Philippus Stephanus) |
| author_facet | Heyns, P.S. (Philippus Stephanus) |
| collection | Thesis |
| description | Dissertation (MEng (Mechanical Engineering))--University of Pretoria, 2018. |
| format | Thesis |
| id | oai:repository.up.ac.za:2263/110034 |
| institution | University of Pretoria (South Africa) |
| language | English |
| last_indexed | 2026-06-10T12:40:25.890Z |
| license_str | Not specified — see source repository |
| provenance_str_mv | Harvested via OAI-PMH from UPSpace — University of Pretoria Institutional Repository |
| publishDate | 2026 |
| publishDateRange | 2026 |
| publishDateSort | 2026 |
| record_format | dspace |
| source_str | UPSpace — University of Pretoria Institutional Repository |
| spelling | oai:repository.up.ac.za:2263/110034 Rail surface anomaly detection : a deep learning . approach for computer vision Heyns, P.S. (Philippus Stephanus) u13026888@tuks.co.za Deetlefs, Richard Deep learning computer vision rail surface anomaly detection unsupervised segmentation real-time inspection. Dissertation (MEng (Mechanical Engineering))--University of Pretoria, 2018. Rail surface defects have become more of an issue in recent years due to new manufacturing techniques which produce head-hardened rails and as industry demands higher speeds, heavier loads and increased traffic. These defects can cause catastrophic accidents, which have consequences such as death, injury, huge cost implications and loss of public confidence. Computer vision systems have become popular, as cameras are non-contact full-field sensors which are low in cost, have high sampling rates and provide appealing performance. However, accurate inspection remains challenging due to dynamic non-linear environmental and rail surface conditions in which images are captured, which result in a heterogeneous image dataset. It is also difficult to select useful features which satisfy the variations due to different failure modes. In addition, there is a class imbalance issue, as most captured images do not contain any defects. In this dissertation, we develop deep generative models that are trained exclusively using healthy images of a rail surface so that we learn useful features to capture the complex nature of the images which are acquired. We propose multiple models which operate with images at different resolutions. We present a new dataset which will be made publicly available. Experimental results demonstrate that our proposed models can perform accurate detection using our dataset. The proposed algorithms are highly parallel and computationally efficient, which enables real-time inspection at speeds that exceed the world’s fastest railway trains: Fuxing Hao CR400AF/BF that has a continuous operation speed of approximately 400 km/h. Mechanical and Aeronautical Engineering MEng (Mechanical Engineering) 2026-05-15T17:26:08Z 2026-05-15T17:26:08Z 19/03/25 2018 Dissertation http://hdl.handle.net/2263/110034 en application/pdf |
| spellingShingle | Deep learning computer vision rail surface anomaly detection unsupervised segmentation real-time inspection. Rail surface anomaly detection : a deep learning . approach for computer vision |
| title | Rail surface anomaly detection : a deep learning . approach for computer vision |
| title_full | Rail surface anomaly detection : a deep learning . approach for computer vision |
| title_fullStr | Rail surface anomaly detection : a deep learning . approach for computer vision |
| title_full_unstemmed | Rail surface anomaly detection : a deep learning . approach for computer vision |
| title_short | Rail surface anomaly detection : a deep learning . approach for computer vision |
| title_sort | rail surface anomaly detection a deep learning approach for computer vision |
| topic | Deep learning computer vision rail surface anomaly detection unsupervised segmentation real-time inspection. |
| url | http://hdl.handle.net/2263/110034 |