Evaluation and development of deep neural networks for super-resolution of microscopy and astrophysics images
Due to physical constrains of an Electron Microscope, capturing high-resolution scans of a subject takes a very long time. On the other hand, running a Gravitational N-body simulation of hundreds of millions of particles, required for state-of-the-art research, takes millions of CPU hours. Thus, in...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis de maestría |
| Fecha de publicación: | 2021 |
| País: | España |
| Institución: | Universidad del País Vasco |
| Repositorio: | Addi. Archivo Digital para la Docencia y la Investigación |
| OAI Identifier: | oai:addi.ehu.eus:10810/58976 |
| Acceso en línea: | http://hdl.handle.net/10810/58976 |
| Access Level: | acceso abierto |
| Palabra clave: | machine learning neural networks image super resolution Wasserstein GAN |
| Sumario: | Due to physical constrains of an Electron Microscope, capturing high-resolution scans of a subject takes a very long time. On the other hand, running a Gravitational N-body simulation of hundreds of millions of particles, required for state-of-the-art research, takes millions of CPU hours. Thus, in this work we propose a new Image Super-Resolution framework based on Generative Adversarial Networks to super-resolve both images scanned by a microscope and snapshots of gravitational N-body simulations. We incorporate techniques from residual neural networks to increase the learning capabilities, and introduce the Wasserstein GAN training method to improve stability. Comparisons have shown that our model performs equally or better than state-of-the art methods in both of these use cases, and provides balanced results that are realistic but don't have much distortion. |
|---|