| Presentation | 2021-11-05 Quaternion convolutional neural networks for PolSAR land classification Yuya Matsumoto, Ryo Natsuaki, Akira Hirose, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | We propose a quaternion convolutional neural network (QCNN) for Polarimetric synthetic aperture radar(PolSAR) land classification. Unlike a conventional real-valued CNN (RVCNN), a QCNN does not simply sum up thecomponents of input vectors in the convolutional processing. A QCNN performs an orthogonal transformation to thecomponents of input vectors by quaternionic rotation and scaling to learn the relationship between them. A QCNNalso can learn spatial textures of input data as well as a conventional RVCNN. In our experiments, we compare threeneural networks, namely, a fully-connected quaternion neural network (QNN), a RVCNN, and our proposed QCNN. Asexperimental results, our proposed QCNN show the best classification performance. We also show that quaternionicconvolution can extract spatial texture by visualizing quaternion kernels. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | Polarimetric synthetic aperture radar (PolsAR) / convolutional neural network (CNN) / quaternion neural network (QNN) |
| Paper # | EMT2021-43 |
| Date of Issue | 2021-10-28 (EMT) |
| Conference Information | |
| Committee | EMT / IEE-EMT |
|---|---|
| Conference Date | 2021/11/4(2days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | Online |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | Electromagnetic Theory, etc. |
| Chair | Hiroyuki Deguchi(Doshisha Univ.) / Akira Matsushima(Kumamoto Univ.) |
| Vice Chair | Hideki Kawaguchi(Muroran Inst. of Tech) |
| Secretary | Hideki Kawaguchi(Miyazaki Univ.) / (Mitsubishi Electric) |
| Assistant | Kazuki Niino(Kyoto Univ.) / Junichiro Sugisaka(Kitami Inst. of Tech.) |
| Paper Information | |
| Registration To | Technical Committee on Electromagnetic Theory / Technical Meeting on Electromagnetic Theory |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | Quaternion convolutional neural networks for PolSAR land classification |
| Sub Title (in English) | |
| Keyword(1) | Polarimetric synthetic aperture radar (PolsAR) |
| Keyword(2) | convolutional neural network (CNN) |
| Keyword(3) | quaternion neural network (QNN) |
| 1st Author's Name | Yuya Matsumoto |
| 1st Author's Affiliation | The University of Tokyo(UTokyo) |
| 2nd Author's Name | Ryo Natsuaki |
| 2nd Author's Affiliation | The University of Tokyo(UTokyo) |
| 3rd Author's Name | Akira Hirose |
| 3rd Author's Affiliation | The University of Tokyo(UTokyo) |
| Date | 2021-11-05 |
| Paper # | EMT2021-43 |
| Volume (vol) | vol.121 |
| Number (no) | EMT-226 |
| Page | pp.pp.76-81(EMT), |
| #Pages | 6 |
| Date of Issue | 2021-10-28 (EMT) |