Janssen, Roos S. (2021) Domain Knowledge Integration in Object Detection Tasks. Design Project, Industrial Engineering and Management.
|
Text
Master_IEM_Design Project_2021_RoosJanssen.pdf Download (9MB) | Preview |
|
![[img]](https://fse.studenttheses.ub.rug.nl/style/images/fileicons/text.png) |
Text
toestemming.pdf Restricted to Registered users only Download (124kB) |
Abstract
While rapid innovations have greatly improved the quality of image classification and object detection models, there still exists a large gap between the manner in which humans and machines learn. BrainCreators, an artificial intelligence company specialized in object detection for visual quality and asset inspectors, wishes to lift knowledge transfer to a higher level by integrating domain knowledge in the trainable machine learning pipeline. BrainCreators hopes that transferring knowledge to machine learning models in more than one way might improve predictive performance and decrease the amount of training data required to learn the patterns in the data, thereby improving the knowledge transfer process from the client domain to the model. Domain knowledge can be represented by knowledge graphs, describing objects, attributes, and relationships. In this project, an exploration of frameworks to integrate domain knowledge in the trainable machine learning pipeline was conducted, after which the Hybrid Knowledge Routed Modules (HRKM) framework was chosen. A new implementation of the framework was created, leading to a modular software deliverable that could potentially be used in the BrainMatter platform. The framework was applied to a dataset of a Dutch telecom company used for cabinet inspection, in order to evaluate the framework on an industrial use case.
| Item Type: | Thesis (Design Project) |
|---|---|
| Supervisor name: | Mohebbi, M. and Jayawardhana, B. |
| Degree programme: | Industrial Engineering and Management |
| Thesis type: | Design Project |
| Language: | English |
| Date Deposited: | 29 Jan 2021 14:02 |
| Last Modified: | 29 Jan 2021 14:02 |
| URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/23853 |
Actions (login required)
 |
View Item |