2019 Fiscal Year Annual Research Report
大気歪み画像モデルを組み込んだ深層学習によるリモートセンシング画像の画質改善
| Project/Area Number |
19J13820
|
|---|
| Research Institution | Ritsumeikan University |
|---|
Principal Investigator |
LI YINHAO 立命館大学, 情報理工学研究科, 特別研究員(DC2)
|
|---|
| Project Period (FY) |
2019-04-25 – 2021-03-31
|
| Keywords | Image processing / Artificial intelligence |
|---|
| Outline of Annual Research Achievements |
The purpose of this research is to enhance the quality of remote sensing images using deep 3D convolutional neural networks incorporated with an atmospherically distorted image model.
In the first year, my research was mainly about the proposal and improvement of new technologies and methods.For the improvement of the image quality of large-size and high-dimensional images, based on the very hot artificial intelligence (AI) technology in recent years, I have proposed a lightweight deep learning method.
This method not only can maintain or exceed the existing methods in accuracy, but also has a faster processing speed and lower hardware requirements for saving the model, so it is more suitable for practical applications than traditional methods.
|
| Current Status of Research Progress |
Current Status of Research Progress
2: Research has progressed on the whole more than it was originally planned.
Reason
So far, the progress of the research is basically in line with the original plan. I studied and referred to various deep learning methods in recent years, and then proposed and improved my original model. From the results, my method outperforms state-of-the-art methods in both accuracy and speed.
In addition, I have published or expected to publish three papers. The first one is about an improved multi-frame super-resolution method which has been published in the IET Image Processing. The second one is a proposed three-dimensional image processing model which has been submitted to a top international conference. The third one is my proposed lightweight super-resolution model which will be submitted to the IEEE Image Processing.
|
| Strategy for Future Research Activity |
In this year, I am going to apply the proposed network to detection, segmentation or classification in geographic information and geological analysis. I plan to use the proposed method to automatically judge whether a region has landslides or major disasters based on satellite images before and after the earthquake.
According to the results, parameters and network architecture will be adjusted to optimize the proposed model. Then, I am going to do evaluations to demonstrate my method outperforms conventional methods in quality enhancement of remote sensing images. Finally, I will publish my findings in top international conferences and journals before I graduate.
|
