|
Type |
Oral Presentation |
Area |
Oral Presentation of Young Analytical Chemists II |
Room No. |
Room 321 |
Time |
FRI 09:40-: |
Code |
ANAL2.O-5 |
Subject |
Evaluation of Fuzzy Rule Building Expert System Tree and Restricted Boltzmann Machine for NIR Spectroscopic Identification of Geographical Origins of Agricultural Products |
Authors |
Woosuk Sohng, Yonjin Shin1, Hoeil Chung2,* Chemistry, Hanyang University, Korea 1Pohang University of Science and Technology, Korea 2Department of Chemistry, Hanyang University, Korea |
Abstract |
Fuzzy rule building expert system tree (FuRES) and Restricted Boltzmann machine (RBM) have been evaluated as a potential method for NIR spectroscopic identification of geographical origins of agricultural samples. FuRES is a tree algorithm with fuzzy expression of classification entropy. FuRES has benefits from simulated annealing and gradient optimization. Restricted Boltzmann machine (RBM) is a restricted form of Boltzmann machines with gradient descent and back-propagation. It can be trained in either supervised or unsupervised ways, depending on the task. In this study, RBM is applied for supervised classification and unsupervised way to extract features. For the evaluation, NIR spectra of imported and domestic agricultural samples (8 different samples: adzuki, angelica root, bellflower root, bracken, carrot, green kernel black bean, kidney bean and perilla seed) were used. For each sample, the discrimination accuracies were acquired using both methods and compared with those using conventional methods such as linear discriminant analysis (LDA) and support vector machine (SVM). The advantages and disadvantages of FuRES and RBM, and their potential in vibrational spectroscopic discriminant analysis will be discussed. |
E-mail |
sohng@hanyang.ac.kr |
|