Comparison of Autoencoder with NMF to Explore Odor Components for Odor Reproduction Using Mass Spectrometry

カテゴリ: 全国大会

論文No: 3-191

グループ名: 【全国大会】平成30年電気学会全国大会論文集

発行日: 2018/03/05

タイトル(英語): Comparison of Autoencoder with NMF to Explore Odor Components for Odor Reproduction Using Mass Spectrometry

著者名: DANI PRASETYAWAN(Tokyo Institute of Technology),Takamichi Nakamoto(Tokyo Institute of Technology)

著者名(英語): DANI PRASETYAWAN(Tokyo Institute of Technology),TAKAMICHI NAKAMOTO(Tokyo Institute of Technology)

キーワード: Odor Reproduction,Odor Components,Mass Spectrometry,Essential Oils,Non-negative Matrix Factorization,Autoencoder

要約(日本語): Odor reproduction enables us to create odor based on sensed odor. It could be done by blending odor components to approximate target odor with the created one as close as possible. Therefore, it is essential to find an appropriate set of odor components for reproducing odor with the minimum number of odor to cover wide range of odors. Mass spectrometry data of 185 essential oils were collected. NMF and autoencoder were used to extract odor component. Increasing number of basis vectors in NMF and hidden neurons in autoencoder affects in increasing similarity between original data and reconstructed data.

原稿種別: 英語

PDFファイルサイズ: 179 Kバイト