Development if intelligent classifier and estimator for tualang honey purity

Norazian, Subari (2014) Development if intelligent classifier and estimator for tualang honey purity. Masters thesis, Universiti Sains Malaysia (Contributors, UNSPECIFIED: UNSPECIFIED).


Download (829kB) | Preview


Honey is a natural substance well-known as supplement for maintaining good health. It is also useful as an ingredient in medicine. However, the market price of pure honey is expensive, causing irresponsible parties to adulterate pure honey by adding various sugar substances. It is very challenging to come out with a suitable method to prove the presence of adulterants in honey products. Most previous studies involved close data observation from experts that is time-consuming. This research proposes the development of intelligent classifier to aid the task of differentiating pure honey from adulterated ones. Besides intelligent classifier, this research has also developed an intelligent estimator for the purpose of giving a percentage estimation of pure honey that exists in adulterated honey sample. The pure honey classifier and estimator are developed using Artificial Neural Network (ANN) approach. Ten types of pure honey from different brands and sugar compounds have been used to prepare various pure and adulterated honey (at different percentages of pure honey) samples. Electronic nose (E-Nose) and Fourier Transform Infrared Spectroscopy (FTIR) raw data have been gathered from various honey samples. The B-Nose, FTIR low level Fusion data of B-Nose and FTIR of raw and normalized data have been used to train a number of ANNs to produce intelligent classifiers (pure or adulterated honey) and estimators (fraction of pure honey). The research results showed that intelligent pure honey classifiers developed using E-Nose, FTIR and Fusion data gives classification accuracies of 100% with 0.390 seconds training time, 99.72% with 0.359 seconds training time and 100% with 0.094 seconds training time, respectively. The result comparison show that the intelligent pure honey classifier gives the best performance is the one trained with Fusion data. The developed intelligent pure honey fraction estimator systems gave average absolute errors of 16.55%, 6.11% and 4.88% for E-Nose, FTIR and Fusion data, respectively. The research result has revealed that intelligent pure honey classifier and estimator developed based on raw E-Nose and FTIR low level Fusion data is better than those using single E-Nose or FTIR data. However, correct pure honey fraction estimation of 30.9% suggests that the performance of intelligent pure honey fraction estimator still needs to be improved. Overall, this research has shown that ANN has the potential to aid the tasks of differentiating pure honey from adulterated ones and estimate the fraction of pure honey in adulterated honey samples.

Item Type: Thesis (Masters)
Uncontrolled Keywords: Neural networks; System analysis
Subjects: Q Science > QA Mathematics > QA76 Computer software
Depositing User: Ms. 'Arifah Nadiah Che Zainol Ariff
Date Deposited: 21 Mar 2016 06:30
Last Modified: 24 Aug 2021 02:16
Download Statistic: View Download Statistics

Actions (login required)

View Item View Item