HIGH-LEVEL DATA FUSION FOR ‘INSIDIOUS FRUIT ROT9 (IFR) DETECTION OF HARUMANIS CULTIVAR

Nurul Syahirah Binti Khalid, Nurul Syahirah (2020) HIGH-LEVEL DATA FUSION FOR ‘INSIDIOUS FRUIT ROT9 (IFR) DETECTION OF HARUMANIS CULTIVAR. Doctoral thesis, UNIVERSITI MALAYSIA PERLIS.

Text HIGH-LEVEL DATA FUSION FOR ‘INSIDIOUS FRUIT ROT' (IFR) DETECTION OF HARUMANIS CULTIVAR_NURUL SYAHIRAH KHALID OCR.pdf - Submitted Version Restricted to Registered users only Download (27MB) | Request a copy

Abstract

Harumanis cv. is known as the king of mangoes where every year, large amount of this mangoes is produced and its quality assessments need to be evaluated before being sold or exported. Existing approach to evaluate the quality of the fruit is solely depending on manual observation based on personal experiences, where only external parameters are considered and this could lead to some errors due to inconsistencies made by human inspection. Furthermore, this approach has problems in evaluating the internal characteristics quality of Harumanis cv., as Harumanis cv. are prone to physiological disorder that can only be detected destructively. One of the mainly physiological disorder attack is the internal breakdown known as ‘Insidious Fruit Rot’ (IFR) which has affected the overall production of the fruit. Therefore, there is a need to have a non-destructive method in grading the fruit according to its internal quality based on the presence of IFR. Charged Coupled Device (CCD) camera, Near-Infrared (NIR) Spectroscopy and Electronic nose (E-nose) are the analytical techniques which have been used separately in the food quality appraisal to measure specific gravity and aroma of fruits. In this work, fusion of these three sensors will be used as a mean to increase the reliability of classification of Harumanis cv. condition internally as compared to depending only to a single analytical technique. To get higher accuracy for quality appraisal, high-level data fusion is proposed to fuse these multiple representations using Dempster-Shafer (D-S) and Majority Voting (MVT) fusion techniques. Two types of based classifier; Support Vector Machine (SVM) and K-Nearest Neighbours (KNN) were used to analyze the performance of the single-sensor used. Experimental procedures are then performed to validate the proposed approach. Four testing stages were included; firstly, specific gravity fusion from image processing and NIR spectroscopy; next, specific gravity and aroma fusion from NIR spectroscopy and E-nose; thirdly, fusion of aroma from E-nose and specific gravity from image processing; and lastly, fusion of specific gravity from image processing and NIR spectroscopy with aroma from E-nose. The experimental results show that the proposed method has better performance classification compared with using single-sensor technique in general, where the best fusion method is by applying D-S for specific gravity from image processing technique and specific gravity from NIR spectroscopy with 96.87% accuracy for overall cases. Meanwhile, the best technique in fusing all single-sensors is by applying D-S fusion (95.80% accuracy) compared to MVT method (87.92% accuracy). This proves that IFR can be detected non-destructively by fusing related techniques and this method is hope to be beneficial in helping farmers and agricultural experts in grading and evaluating Harumanis cv. internally in the future.

Item Type:	Thesis (Doctoral)
Subjects:	Q Science > Q Science (General) S Agriculture > S Agriculture (General)
Divisions:	Universiti > Universiti Malaysia Perlis (UniMAP)
Depositing User:	Encik Mohd Zulkarnain Hassan bin Mohd Zainudin
Date Deposited:	16 Oct 2024 16:44
Last Modified:	16 Oct 2024 16:44
URI:	https://repositori.mohe.gov.my/id/eprint/20

Actions (login required)

View Item