2019 Fiscal Year Final Research Report
Signal Recognition Mechanisms by Selecting Higher-Order Spectral Features Through Learning
| Project/Area Number |
16K00322
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Multi-year Fund |
|---|
| Section | 一般 |
|---|
| Research Field |
Soft computing
|
|---|
| Research Institution | University of Tsukuba |
|---|
Principal Investigator |
|
|---|
| Project Period (FY) |
2016-04-01 – 2020-03-31
|
| Keywords | 高次スペクトル / 信号認識 / ニューラルネットワーク / 学習 |
|---|
| Outline of Final Research Achievements |
Higher-Order Spectral features of signals provide nonlinear signal features unavailable in power spectra. However, their high-dimensionality, sparseness and high computational cost were obstacles for general use for signal classification. In this work, we attempted to solve this issue by joint use of Local Higher-Order Moment Spectrum (LHOMS) kernel functions and gradient based learning of neural networks. A modified image feature extraction method by LHOMS kernel was applied to iris authentication. It was found that the method allows authentication robust to additive noise to the iris image. Also, a novel method for inheriting the probability density of parameters in transfer learning of convolutional neural networks was introduced. The novel transfer learning improved the training efficiency when compared with the existing methods.
|
| Free Research Field |
パターン認識，適応的な信号処理，ニューラルネットワーク
|
| Academic Significance and Societal Importance of the Research Achievements |
本研究で得られた成果は，画像や音などのメディア信号の認識に際して用いる特徴量として，高次スペクトル特徴量の抽出とニューラルネットワークを用いた分類学習に関する新たな知見を提供するものである．提案手法は生体認証や画像認識の問題に適用され，それぞれその有効性を示している．メディア信号の学習に基づく自動認識や分類はその重要性がますます増大しており，本研究の成果はそれらを行う手段としてに新たな選択肢を提供するものである．
|
