Ensemble CNNs in Transform Domains for Small Data Image Super-resolution
Public DepositedContenu téléchargeable
open in viewerIn this research, we develop methods for single image super-resolution by combining ideas inspired by compressive sensing with super-resolution neural networks and ensemble learning. We are interested in problems where large data sets are not available such as healthcare and aerospace applications. The essence of our work is to use ideas similar to those leveraged by compressive sensing, namely sparse representations, for a robust model accommodating small training data. We develop and demonstrate techniques for combining classic sparse representations with modern ideas in deep neural networks, such as neural network ensembles, to improve the performance of image super-resolution task. Particularly, we report here a successful application of our model to improve the resolution of areal density maps of carbon nanotube sheets generated by a beta particle transmission system. We show that applying our models can reveal finer details in the material texture, helping to improve the detection of manufacturing defects and improving the quality control capabilities for carbon nanotube sheet production.
- Creator
- Contributeurs
- Degree
- Unit
- Publisher
- Identifier
- etd-71551
- Mot-clé
- Advisor
- Committee
- Defense date
- Year
- 2022
- Sponsor
- Date created
- 2022-08-08
- Resource type
- Source
- etd-71551
- Rights statement
- Dernière modification
- 2022-12-09
Relations
- Dans Collection:
Contenu
Articles
La vignette | Titre | Visibilité | Embargo Release Date | actes |
---|---|---|---|---|
Yingnan_Liu_s_thesis__11_.pdf | Public | Télécharger |
Permanent link to this page: https://digital.wpi.edu/show/3197xq223