Classification of adulterated para rubber sheet and adulterated honey using hyperspectral imaging system

Name: Siano, Dharell Bandarlipe

LCSH: Kasetsart University -- Dissertations. D.Eng. (Agricultural Engineering) 2021

Classification :.LCCS: TA1637

LCSH: Kasetsart University. -- Department of Agricultural Engineering -- Dissertations

LCSH: Hyperspectral imaging

LCSH: Image processing

LCSH: Supervised learning (Machine learning)

LCSH: Rubber

LCSH: Hevea

LCSH: Honey

LCSH: Agriculture -- Data processing

Abstract: The study was divided into two topics that focused on the classification of adulterated Para rubber sheets that were coagulated using formic acid or sulfuric acid and the classification of adulterated honey from different botanical origins using NIR-HSI. Hyperspectral imaging is a non-destructive technique that has the capability of acquiring spectral and spatial information of the samples that can be used to develop a classification model with high accuracy and a classifier map for two-dimensional visual classification. Both topics used reflectance and transflectance modes of measurement to obtain spectral and spatial information from 864.53 nm to 1700 nm. In the first topic, the best classification model was obtained from transflectance mode using SNV+2D pretreated spectra that was analyzed using PLS-DA with rp 2 = 0.88, SEP = 0.176, and RPD = 2.86. The classification accuracy for both modes of measurement was 98.33 % on which PRS-SA was classified better than PRS-FA. Although transflectance mode showed better results in terms of RPD, SEP, and rp 2 , the t-test revealed that there was no significant difference between the two models developed at each mode which depicted that both modes of measurement could equally be applied in the classification of Para rubber sheets with a high level of accuracy. Moreover, the classifier mapped images provided two-dimensional color visualization of each type of Para rubber sheet that could be easily recognized. In the second topic, the best model was observed in transflectance mode using SDA in a LOOCV method with test set accuracy of 100 %. Using the selected features by the SDA, an increase in the test set accuracy was observed, resulting in 96 % for linear SVM, 96 % for Gaussian SVM, 91 % for cubic SVM, and 92 % for kNN. The method for the development of classifier maps using the ten-discriminant function for each mode of measurement can create a two-dimensional image that has a distinct color pertaining to a particular class that can easily discriminate one class from another.

Kasetsart University. Office of the University Library

Address: Bangkok

Email: tdckulib@ku.ac.th

Name: Anupun Terdwongworakul

Role: Thesis Advisor

Name: Wanrat Abdullakasim

Role: Thesis CO-Advisor

Role: Thesis CO-Advisor

Name: Kaewkarn Phuangsombut

Role: Thesis CO-Advisor

Created: 2021

Modified: 2025-07-26

Issued: 2025-07-26

วิทยานิพนธ์/Thesis

application/pdf

URL: https://www.lib.ku.ac.th/KUthesis/2564/dharell-ban-all.pdf

CallNumber: TA1637 .S54

eng

DegreeName: Doctor of Engineering

Level: Doctoral Degraee

Descipline: Agricultural Engineering

Grantor: Kasetsart University

©copyrights Kasetsart University

ลำดับที่.	ชื่อแฟ้มข้อมูล	ขนาดแฟ้มข้อมูล	จำนวนเข้าถึง	วัน-เวลาเข้าถึงล่าสุด
1	dharell-ban-all.pdf	2.75 MB

ใช้เวลา

0.023936 วินาที