2019 Fiscal Year Research-status Report
Deep forest for disaster monitoring with multi-temporal and multi-modal earth observation data
| Project/Area Number |
19K20309
|
|---|
| Research Institution | Institute of Physical and Chemical Research |
|---|
Principal Investigator |
Xia Junshi 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 (00830168)
|
|---|
| Project Period (FY) |
2019-04-01 – 2022-03-31
|
| Keywords | Damage mapping / Machine learning / Deep forest / Multi-source datasets |
|---|
| Outline of Annual Research Achievements |
The research achievements can be summarized as follows:
1. We have made several damage responses with Group Earth Observation (GEO), such as 2018 Sulawesi, Sunda Strait earthquake/tsunami, 2019 Brazil Dam collapse etc. In these damage responses, we have applied the image preprocessing (i.e., noise filtering), classification and change detection, and post-processing to produce the damage mapping.
2. We have proposed a deep rotation forest, which uses multiple levels of rotation forest, for the classification of hyperspectral and Light Detection and Ranging (LiDAR). This kind of technique outperforms the rotation forest and deep neural networks, and will be extended for the damage mapping.
3. We have applied machine learning classifiers (i.e., random forest, rotation forest, and canonical correlation forest) for the four-levels damage mapping of 2018 Sulawesi earthquake/tsunami using multiple source datasets (i.e., optical and Synthetic Aperture Radar, SAR).
4. We have proposed to use transfer learning, including overall centroid alignment and deep learning, and image translation techniques for moderate-resolution SAR, and very high-resolution optical images in real-world cross-domain application (i.e., building damage mapping).
|
| Current Status of Research Progress |
Current Status of Research Progress
2: Research has progressed on the whole more than it was originally planned.
Reason
The project was performed smoothly. Besides the initially planned work (i.e., constructing the database), we have performed several disaster damage responses and applications of multiple source damage mapping and tranfer learning damage mapping.
|
| Strategy for Future Research Activity |
Now, two damage mapping datasets, including xView2 (with high-resolution optical) and our multiple-source (with high-resolution optical and SAR) datasets are available. In the next fiscal year, my research plan, which works towards the above datasets, includes that
1. The deep forest will be integrated with the deep neural network, which may provide the advantages of low computational cost and no-tunning parameters.
2. Domain adaptation will be investigated in the build damage mapping. In this case, damage grading map of the future events will be produced for the rapid disaster response.
|
| Causes of Carryover |
In the first year, I focus on the disaster damage response and construct the database. In the next year, I will focus the real applications using the database.
We will aconsider to submit the papers to several international conference. We will also consider to submit several journal papers.
|
