Generative Adversarial Networks Based Multi-Sensor Remote Sensing Image Translation for Disaster Damage Mapping
Project/Area Number |
19K20308
|
Research Category |
Grant-in-Aid for Early-Career Scientists
|
Allocation Type | Multi-year Fund |
Review Section |
Basic Section 61010:Perceptual information processing-related
|
Research Institution | Institute of Physical and Chemical Research |
Principal Investigator |
He Wei 国立研究開発法人理化学研究所, 革新知能統合研究センター, 研究員 (10819387)
|
Project Period (FY) |
2019-04-01 – 2022-03-31
|
Project Status |
Discontinued (Fiscal Year 2021)
|
Budget Amount *help |
¥4,160,000 (Direct Cost: ¥3,200,000、Indirect Cost: ¥960,000)
Fiscal Year 2021: ¥910,000 (Direct Cost: ¥700,000、Indirect Cost: ¥210,000)
Fiscal Year 2020: ¥1,430,000 (Direct Cost: ¥1,100,000、Indirect Cost: ¥330,000)
Fiscal Year 2019: ¥1,820,000 (Direct Cost: ¥1,400,000、Indirect Cost: ¥420,000)
|
Keywords | inpainting / flood area detection / dataset preparation / GAN generation / image restoration / image translation / multi-sensor / deep learning / disaster response |
Outline of Research at the Start |
This project consists of three aspects: 1) SAR2SAR translation via GAN based method. 2) SAR2optical image translation via multi-temporal SAR-optical image fusion based method. 3) Disaster damage mapping.
|
Outline of Annual Research Achievements |
Our purpose is to predict the post-disaster optical image with landslide details, from the input of pre-disaster SAR-optical image pairs and post-disaster SAR image. Previous deep learning based methods can recover the optical image in good visual, unfortunately with the landslides disappeared. To reconstruct physically meaningful details, I calculate a weight matrix of post-disaster SAR image to measure the importance of each pixel, and utilize the weight matrix to guild the reconstruction of optical image. Until now, I have finished the dataset preparation, and on the basis of the dataset, I developped the multi-temporal SARoptical method. On the basis of the related works, we also developped several image quality improvement methods for remote sensing image denoising, resotration, and reconstruction. Firstly, we try to reconstruct the hypersepctral image from the low-spatial-resolution hyperspectral image and low-spectral-resolution multispectral image. Secondly, we try to reconstruct the hyperspectral image from color image and the measurements via computational camera. The related publications include 1 paper accepted by Pattern Recognition, 1 paper accpepted by IEEE transactions on image processing.
|
Report
(3 results)
Research Products
(13 results)