Data-driven Filter Design and Implementation for Snapshot Hyperspectral Imaging
| Project/Area Number |
19K20307
|
|---|
| 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 | National Institute of Informatics |
|---|
Principal Investigator |
ZHENG YINQIANG 国立情報学研究所, コンテンツ科学研究系, 准教授 (30756896)
|
|---|
| Project Period (FY) |
2019-04-01 – 2021-03-31
|
| Project Status |
Completed (Fiscal Year 2020)
|
| Budget Amount *help |
¥4,160,000 (Direct Cost: ¥3,200,000、Indirect Cost: ¥960,000)
Fiscal Year 2020: ¥1,430,000 (Direct Cost: ¥1,100,000、Indirect Cost: ¥330,000)
Fiscal Year 2019: ¥2,730,000 (Direct Cost: ¥2,100,000、Indirect Cost: ¥630,000)
|
|---|
| Keywords | Spectral Imaging / Deep Learning / Filter Selection / Filter Design / spectral imaging / filter design / deep learning |
|---|
| Outline of Research at the Start |
Existing multi-channel devices, such as the widespread three-channel RGB cameras, are not necessarily optimal for hyperspectral reconstruction. This research proposal aims to optimize the filter response for multispectral-to-hyperspectral reconstruction using deep neural networks.
|
| Outline of Final Research Achievements |
The research purpose of this project is to find the best spectral response functions for accurate multispectral-to-hyperspectral reconstruction using deep neural networks, and when necessary, implement the deeply learned filters by using film manufacturing technologies. We have tried to indentify the best camera spectral response curves from a given camera database, and design the optimal IR-cut filter for RGB-based spectral reconstruction. We have also examined fusion based spectral reconstruction, and found the best camera spectral response curves. Finally, we have gone beyond spectral reconstruction and examined the effect of spectral response fuctions for high-level task of scene classification.
|
| Academic Significance and Societal Importance of the Research Achievements |
深層学習を用いてイメージングハードウェアの最適化はとても挑戦的な研究課題です。本研究では、カメラの感度曲線の最適化方法を開発する上、スペクトル再構成の精度を向上させた。更に、製造上の拘束も考慮したので、アルゴリズムによる設計結果はフィルターで忠実に実装が可能であることも示した。
|
Report
(3 results)
Research Products
(10 results)
