PERBANDINGAN ARSITEKTUR CNN MOBILENETV3-LARGE DAN EFFICIENTNETB2 MENGGUNAKAN EKSTRAKSI FITUR LBP PADA KLASIFIKASI JENIS RIMPANG

Amin, M. Ryan Nurdiansyah N (2025) PERBANDINGAN ARSITEKTUR CNN MOBILENETV3-LARGE DAN EFFICIENTNETB2 MENGGUNAKAN EKSTRAKSI FITUR LBP PADA KLASIFIKASI JENIS RIMPANG. Undergraduate thesis, Fakultas Ilmu Komputer.

	Text (Cover) 21081010321-cover.pdf Download (1MB)
	Text (Bab 1) 21081010321-bab1.pdf Download (256kB)
	Text (Bab 2) 21081010321-bab2.pdf Restricted to Repository staff only until 17 December 2028. Download (1MB)
	Text (Bab 3) 21081010321-bab3.pdf Restricted to Repository staff only until 17 December 2028. Download (1MB)
	Text (Bab 4) 21081010321-bab4.pdf Restricted to Repository staff only until 17 December 2028. Download (3MB)
	Text (Bab 5) 21081010321-bab5.pdf Download (270kB)
	Text (Daftar Pustaka) 21081010321-daftarpustaka.pdf Download (214kB)
	Text (Lampiran) 21081010321-lampiran.pdf Restricted to Repository staff only until 17 December 2028. Download (237kB)

Abstract

Indonesia has a diverse range of rhizome species used in food and health applications, but the visual similarity between species makes manual identification difficult. This study compares the performance of the CNN architectures MobileNetV3-Large and EfficientNetB2 on original rhizome images as well as images enhanced with texture features using Local Binary Pattern (LBP). The dataset consists of 400 images from eight classes (50 images per class), processed through resizing, augmentation, normalization, and split into 80% training, 10% validation, and 10% testing. Experiments were conducted in 16 test results with learning rates of 0.1, 0.01, 0.001, and 0.0001, evaluated using accuracy, precision, recall, and F1-score. In general, large learning rates (0.1 and 0.01) degraded the performance of both architectures, while smaller learning rates combined with LBP produced much better results. The best configuration was obtained by EfficientNetB2 with LBP at a learning rate of 0.001 (Test Result 16), achieving an accuracy of 0.9850. precision of 0.9859, recall of 0.9850. and F1-score of 0.9850. followed by MobileNetV3-Large with LBP at a learning rate of 0.001 (Test Result 8) with 0.9800 accuracy, 0.9815 precision, 0.9800 recall, and 0.9799 F1-score. These findings indicate that combining LBP with CNN architectures and using small learning rates can improve rhizome classification accuracy and has the potential to support more reliable automatic identification systems.

Item Type:

Thesis (Undergraduate)

Contributors:

Contribution	Contributors	NIDN/NIDK	Email
Thesis advisor	Via, Yisti Vita	NIDN0025048602	yistivia.if@upnjatim.ac.id
Thesis advisor	Nurlaili, Afina Lina	NIM0013129303	afina.lina.if@upnjatim.ac.id

Subjects:

T Technology > T Technology (General)

Divisions:

Faculty of Computer Science > Departemen of Informatics

Depositing User:

M.Ryan Nurdiansyah N.A

Date Deposited:

17 Dec 2025 05:31

Last Modified:

17 Dec 2025 05:31

URI:

https://repository.upnjatim.ac.id/id/eprint/48317

Actions (login required)

View Item