検索対象:     
報告書番号:
※ 半角英数字
 年 ~ 
 年

主成分分析により次元削減した加工データを学習データに用いたニューラルネットワークによるスペクトル解析

Spectral analysis by a neural network using processed data whose dimensions have been reduced by principal component analysis as learning data

大場 正規 

Oba, Masaki

Gd$$_{2}$$O$$_{3}$$、TiO$$_{2}$$、ZrO$$_{2}$$のデータから加工で得た462種類の学習データをPCAにより次元削減を行った。学習後、実試料のデータ62種類をテストデータとして元素の含有比を解析した。前回同様、実試料の真値と解析値の校正曲線を作成し、含有比を解析する。今回用いたニューラルネットワークは、入力層-中間層(2層)-出力層という構成で、中間層は各100ノード2層用いた。学習データのPCAの結果、7944次元(ピクセル)の学習データを5次元に大幅に削減することができた。これを用いて学習させ、テストデータを解析した結果、真値との差およそ$$pm$$10%で、前回とほぼ同様な値であった。

Dimension reduction was performed using PCA on 462 types of training data obtained by processing Gd$$_{2}$$O$$_{3}$$, TiO$$_{2}$$, ZrO$$_{2}$$data. After learning, the content rates of elements were analyzed using 62 types of data from actual samples as test data. Similar to last time, create a calibration curve of the true value and analytical value of the actual sample and analyze the content ratio. The neural network used this time had a configuration of input layer, middle layer (2 layers), and output layer, and the middle layer used 2 layers with 100 nodes each. As a result of PCA on the training data, we were able to significantly reduce the 7944 dimensions (pixels) of the training data to 5 dimensions. As a result of training using this and analyzing test data, the difference from the true value was approximately 10%, which was almost the same as the previous value.

Access

:

- Accesses

InCites™

:

Altmetrics

:

[CLARIVATE ANALYTICS], [WEB OF SCIENCE], [HIGHLY CITED PAPER & CUP LOGO] and [HOT PAPER & FIRE LOGO] are trademarks of Clarivate Analytics, and/or its affiliated company or companies, and used herein by permission and/or license.