EGCLAMNETL2: MODEL PENGENALAN CITRA DAUN FITOMEDISIN MENGGUNAKAN CONTRAST ENHANCEMENT DAN REGULARIZED DEEP LEARNING BERBASIS ENSEMBLE BOOSTING

AYUMI, VINA and Ermatita, Ermatita and Abdiansah, Abdiansah (2023) EGCLAMNETL2: MODEL PENGENALAN CITRA DAUN FITOMEDISIN MENGGUNAKAN CONTRAST ENHANCEMENT DAN REGULARIZED DEEP LEARNING BERBASIS ENSEMBLE BOOSTING. Doctoral thesis, Sriwijaya University.

	Text RAMA_21001_03013682126016.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (15MB) | Request a copy
	Text RAMA_21001_03013682126016_TURNITIN.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (21MB) | Request a copy
	Text RAMA_21001_03013682126016_0013096707_0001108401_01_front_ref.pdf - Accepted Version Available under License Creative Commons Public Domain Dedication. Download (1MB)
	Text RAMA_21001_03013682126016_0013096707_0001108401_02.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (1MB) | Request a copy
	Text RAMA_21001_03013682126016_0013096707_0001108401_03.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (1MB) | Request a copy
	Text RAMA_21001_03013682126016_0013096707_0001108401_04.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (1MB) | Request a copy
	Text RAMA_21001_03013682126016_0013096707_0001108401_05.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (829kB) | Request a copy
	Text RAMA_21001_03013682126016_0013096707_0001108401_06.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (348kB) | Request a copy
	Text RAMA_21001_03013682126016_0013096707_0001108401_07_ref.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (356kB) | Request a copy
	Text RAMA_21001_03013682126016_0013096707_0001108401_08_lamp.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (9MB) | Request a copy

Abstract

Phytomedical knowledge is knowledge related to the use of plants for treatment that are passed down from generation to generation. The study of phytomedicine is concerned not only with knowledge management but also image processing. To manage this knowledge, it is necessary to implement computer vision technology to process phytomedicine plant image data so that it can be managed easily. This study conducted experiments using four different models, namely MNET, MNETL2, GCLAMNET and EGCLAMNETL2. MNET experiments conducted using existing parameter settings with values include batch size: 32, epoch: 50, optimizer value: Adam, learning rate value = 0.0001 and input shape value: 224x224L2 regularization (0.05) in the MNETL2 model has a significant impact on loss values. The loss value in the MNETL2 model tends to decrease compared to MNET significantly at each epoch. There are three parameters that need to be initialized in the proposed GCLA model: clipLimit, tileGridSize and midGamma. The clipLimit parameter sets the contrast limiting threshold value. The clipLimit parameter is defined with a value of (3.0). The tileGridSize parameter is used for the number of tiles in each row and column of phytomedicine image pixels. The tileGridSize parameter is defined with a value (8.8) while the midGamma parameter is defined with a value (1.0). Several previous experiments were conducted to determine the appropriate parameter coefficients for each model before being used on EGCLAMNETL2. The proposed EGCLAMNETL2 model architecture as a model of phytomedicine plant leaf image recognition uses an ensemble boosting approach. Based on the experimental results, EGCLAMNETL2 got the best test accuracy compared to the previous model, which was 90.72%, MNET got an accuracy of 83.73%, GCLAMNET got an accuracy of 69.39% and MNETL2 got an accuracy of 84.27%.

Item Type:	Thesis (Doctoral)
Uncontrolled Keywords:	phytomedicine, deep learning, image processing, contrast enhancement, regularization, ensemble boosting
Subjects:	T Technology > T Technology (General) > T58.5-58.64 Information technology > T58.5 General works Management information systems Cf. HD30.213 Industrial management Cf. HF5549.5.C6+ Communication in personnel management Cf. TS158.6 Automatic data collection systems (Production control)
Divisions:	03-Faculty of Engineering > 21001-Engineering Science (S3)
Depositing User:	Vina Ayumi
Date Deposited:	28 Nov 2023 06:36
Last Modified:	28 Nov 2023 06:36
URI:	http://repository.unsri.ac.id/id/eprint/131330

Actions (login required)

View Item