| Presentation | 2021-05-28 Self-supervised representation learning with grayscale images Yuichiro Sumi, Takuto Kojima, Kentaro Kutsukake, Tetsuya Matsumoto, Hiroaki Kudo, Yoshinori Takeuchi, Noritaka Usami, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | Multicrystalline silicon solar cells, which are most commonly used in residential photovoltaic systems, include regions which show performance degradation due to the generation of dislocation clusters in silicon ingot. In this study, we performed self-supervised learning of ResNet using grayscale images such as PL (photoluminescence) or CT images, which are inherently intensity information, and investigated whether the features obtained by the self-supervised learning are useful for transfer learning for grayscale images. We performed transfer learning on the self-supervised ResNet using 45,000 grayscale transformed CIFAR-10 images for training, 5,000 for validation, and 10,000 for testing, and obtained a correct response rate of 64.7 %. When the transfer learning was conducted using 5,514 silicon PL images for training, 551 for verification, and 3,125 for testing, the correct response rate was 73.0 %. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | self-supervised learning / convolutional neural network / multicrystalline silicon / photoluminescence image / dislocation / generation point |
| Paper # | IMQ2021-1 |
| Date of Issue | 2021-05-21 (IMQ) |
| Conference Information | |
| Committee | IMQ |
|---|---|
| Conference Date | 2021/5/28(1days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | Online |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | |
| Chair | Toshiya Nakaguchi(Chiba Univ.) |
| Vice Chair | Mitsuru Maeda(Canon) / Kenya Uomori(Osaka Univ.) |
| Secretary | Mitsuru Maeda(Shizuoka Univ.) / Kenya Uomori(Sony Semiconductor Solutions) |
| Assistant | Hiroaki Kudo(Nagoya Univ.) / Masaru Tsuchida(NTT) |
| Paper Information | |
| Registration To | Technical Committee on Image Media Quality |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | Self-supervised representation learning with grayscale images |
| Sub Title (in English) | |
| Keyword(1) | self-supervised learning |
| Keyword(2) | convolutional neural network |
| Keyword(3) | multicrystalline silicon |
| Keyword(4) | photoluminescence image |
| Keyword(5) | dislocation |
| Keyword(6) | generation point |
| 1st Author's Name | Yuichiro Sumi |
| 1st Author's Affiliation | Nagoya University(Nagoya Univ.) |
| 2nd Author's Name | Takuto Kojima |
| 2nd Author's Affiliation | Nagoya University(Nagoya Univ.) |
| 3rd Author's Name | Kentaro Kutsukake |
| 3rd Author's Affiliation | Institute of Physical and Chemical Research(RIKEN) |
| 4th Author's Name | Tetsuya Matsumoto |
| 4th Author's Affiliation | Nagoya University(Nagoya Univ.) |
| 5th Author's Name | Hiroaki Kudo |
| 5th Author's Affiliation | Nagoya University(Nagoya Univ.) |
| 6th Author's Name | Yoshinori Takeuchi |
| 6th Author's Affiliation | Daido University(Daido Univ.) |
| 7th Author's Name | Noritaka Usami |
| 7th Author's Affiliation | Nagoya University(Nagoya Univ.) |
| Date | 2021-05-28 |
| Paper # | IMQ2021-1 |
| Volume (vol) | vol.121 |
| Number (no) | IMQ-48 |
| Page | pp.pp.1-4(IMQ), |
| #Pages | 4 |
| Date of Issue | 2021-05-21 (IMQ) |