Full Text Available
Note: Clicking the button above will open the full text document at the original institutional repository in a new window.
Object detection and classification in side scan sonar imagery using convolutional neural networks
Thesis (MEng)--Stellenbosch University, 2024.
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Published: |
Stellenbosch : Stellenbosch University
2025
|
| Subjects: | |
| Tags: |
No Tags, Be the first to tag this record!
|
| _version_ | 1867614077894787072 |
|---|---|
| access_status_str | Open Access |
| author | Ruzario, Souzah Emmanuel |
| author2 | Engelbrecht, J. A. A. |
| author_browse | Engelbrecht, J. A. A. Ruzario, Souzah Emmanuel |
| author_facet | Engelbrecht, J. A. A. Ruzario, Souzah Emmanuel |
| author_sort | Ruzario, Souzah Emmanuel |
| collection | Thesis |
| dc_rights_str_mv | Stellenbosch University |
| description |
Thesis (MEng)--Stellenbosch University, 2024. |
| format | Thesis |
| id | oai:scholar.sun.ac.za:10019.1/131900 |
| institution | Stellenbosch University (South Africa) |
| last_indexed | 2026-06-10T12:46:18.613Z |
| license_str | Other — see source repository |
| provenance_str_mv | Harvested via OAI-PMH from SUNScholar — Stellenbosch University Repository |
| publishDate | 2025 |
| publishDateRange | 2025 |
| publishDateSort | 2025 |
| publisher | Stellenbosch : Stellenbosch University |
| publisherStr | Stellenbosch : Stellenbosch University |
| record_format | dspace |
| source_str | SUNScholar — Stellenbosch University Repository |
| spelling | oai:scholar.sun.ac.za:10019.1/131900 Object detection and classification in side scan sonar imagery using convolutional neural networks Ruzario, Souzah Emmanuel Engelbrecht, J. A. A. Engelbrecht, H. A. Stellenbosch University. Faculty of Engineering. Dept. of Electrical and Electronic Engineering. Computer vision Image processing -- Digital techniques Neural networks (Computer science) UCTD Thesis (MEng)--Stellenbosch University, 2024. ENGLISH ABSTRACT: Sidescan sonar facilitates high resolution sea floor mapping. However, the analysis of these images is still heavily dependent on human interpreters and their expertise. Object detection and classification machine l earning models provide a potential solution for automated analysis of side scan sonar images, but are limited in feasibility by the lack of freely available data to train them. This project aims to develop an object detection and classification system to analyse side scanning sonar data utilising YOLOv3, YOLOv3-SPP and YOLOv4 models trained and tested on manually constructed real sonar image data sets containing sand and rock as classes. Models are further trained and tested on synthetic sonar image data sets containing an additional sea mine class, generated by artificially inserting sonar images of underwater mines into images from the real sonar data sets. Experiments on the random parameter, channel parameter, window size and anchor boxes are performed. Results show that YOLO models with different configurations can perform object detection and classification accurately on both real and synthetic sonar images. Results also suggest that the use of synthetically generated or real training data containing an adequate variety of examples for each class helps produce more accurate models. AFRIKAANSE OPSOMMING:Syskandering sonar fasiliteer hoe¨ resolusie seebodem kartering, maar die ontleding van hierdie beelde is steeds sterk afhanklik van menslike interpreteerders en hul kundigheid. Masjienleermodelle vir objekbespeuring en klassifikasie bied ’n potensie¨le oplossing vir geoutomatiseerde ontleding van syskandering-sonarbeelde, maar word in uitvoerbaarheid beperk deur die gebrek aan vrylik beskikbare data om hulle op te lei. Hierdie projek het ten doel om ’n objek-opsporing- en klassifikasiestelsel te ontwikkel om syskandering-sonardata te ontleed deur gebruik te maak van YOLOv3-, YOLOv3-SPP- en YOLOv4-modelle wat opgelei en getoets is op regte sonarbeelddatastelle wat sand en rots as klasse bevat. Modelle word verder opgelei en getoets op sintetiese sonarbeelddatastelle wat ’n bykomende seemyn klas bevat, gegenereer deur sonarbeelde van onderwatermyne kunsmatig in beelde van die regte sonardatastelle in te voeg. Eksperimente op die ewekansige parameter, kanaalparameter, venstergrootte en ankerbokse word uitgevoer. Resultate toon dat YOLO-modelle met verskillende konfigurasies voorwerpopsporing en klassifikasie akkuraat kan uitvoer op beide werklike en sintetiese sonarbeelde. Resultate dui ook daarop dat sinteties gegenereerde of werklike opleidingsdata met voldoende veran-derlikevoorbeelde vir elke klas teenwoordig in die datastel akkurate modelle produseer. Masters 2025-04-08T08:07:38Z 2025-04-08T08:07:38Z 2024-12 Thesis https://scholar.sun.ac.za/handle/10019.1/131900 Stellenbosch University xviii, 138 pages : illustrations application/pdf Stellenbosch : Stellenbosch University |
| spellingShingle | Computer vision Image processing -- Digital techniques Neural networks (Computer science) UCTD Ruzario, Souzah Emmanuel Object detection and classification in side scan sonar imagery using convolutional neural networks |
| title | Object detection and classification in side scan sonar imagery using convolutional neural networks |
| title_full | Object detection and classification in side scan sonar imagery using convolutional neural networks |
| title_fullStr | Object detection and classification in side scan sonar imagery using convolutional neural networks |
| title_full_unstemmed | Object detection and classification in side scan sonar imagery using convolutional neural networks |
| title_short | Object detection and classification in side scan sonar imagery using convolutional neural networks |
| title_sort | object detection and classification in side scan sonar imagery using convolutional neural networks |
| topic | Computer vision Image processing -- Digital techniques Neural networks (Computer science) UCTD |
| url | https://scholar.sun.ac.za/handle/10019.1/131900 |
| work_keys_str_mv | AT ruzariosouzahemmanuel objectdetectionandclassificationinsidescansonarimageryusingconvolutionalneuralnetworks |