Georgia Institute of Technology
Deep Learning Enhanced Biofilm Topography through Convolutional Neural Network
Abstract
dc:description.abstractBiofilms are surface attached communities microbes. One approach to study the formation and growth of biofilms is to observe its surface topography, such as with white light profilometry. However, this technique requires taking images of many fields of view and stitching them together, a time-consuming process. We thus sought to develop a convolutional neural network to that can convert low-resolution images to high-resolution images. Our results show that the technique succeeds with a low mean absolute error (~10^-4). We also found that the model prediction error is highly related to the biofilm's topographic roughness. As a result, highly rough surfaces are crucial resources for training deep learning super-resolution model. Roughness enriches the complexity of the biofilm surface, and a model trained on biofilm formed by strains with high roughness yield a lower error on other strains.
Degree
thesis:*- Level thesis:degree_level
- Masters
- Department dc:contributor.department
- Physics
- Grantor dc:publisher
- Georgia Institute of Technology
- Year dc:date.issued
- 2023
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Zhao, Lin
- Advisors dc:contributor.advisor
-
- Yunker, Peter J.
- Yunker, Peter
- Committee members dc:contributor.committeemember
-
- Curtis, Jennifer
- Gumbart, James C.
Subjects
dc:subject × 4Rights
- Language dc:language.iso
- en_US
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- https://hdl.handle.net/1853/75114