KLASIFIKASI PENYAKIT DAUN CABAI PADA CITRA LAPANGAN MENGGUNAKAN ARSITEKTUR TRANSFER LEARNING CONVOLUTIONAL NEURAL NETWORK DENGAN FINE-TUNING DAN DATA AUGMENTATION

Bawazir, Fadhil Mohammad (2026) KLASIFIKASI PENYAKIT DAUN CABAI PADA CITRA LAPANGAN MENGGUNAKAN ARSITEKTUR TRANSFER LEARNING CONVOLUTIONAL NEURAL NETWORK DENGAN FINE-TUNING DAN DATA AUGMENTATION. Masters thesis, UPN Veteran Jawa Timur.

[img] Text (cover)
24066020010 - Cover.pdf
Download (8MB)
[img] Text
24066020010 - Bab1.pdf
Download (34kB)
[img] Text (Bab 2)
24066020010 - Bab2.pdf
Restricted to Repository staff only until 10 August 2029.
Download (24kB)
[img] Text (bab 3)
24066020010 - Bab3.pdf
Restricted to Repository staff only until 10 August 2029.
Download (313kB)
[img] Text (bab 4)
24066020010 - Bab4.pdf
Restricted to Repository staff only until 10 August 2029.
Download (2MB)
[img] Text
24066020010 - Bab5.pdf
Download (9kB)
[img] Text
24066020010 - Daftar Pustaka.pdf
Download (140kB)
[img] Text (lampiran)
24066020010 - Lampiran.pdf
Restricted to Repository staff only until 10 August 2029.
Download (3MB)

Abstract

Early detection of chili leaf diseases plays an important role in improving crop productivity and reducing yield losses. However, manual disease identification remains prone to subjectivity and misdiagnosis due to the visual similarity of disease symptoms. This study aims to comparatively evaluate the performance of three Transfer Learning-based Convolutional Neural Network (CNN) architectures, namely MobileNetV2, ResNet50V2, and EfficientNetB0, using identical training configurations across four experimental scenarios: Transfer Learning, Fine-Tuning, Data Augmentation, and the combination of Fine-Tuning and Data Augmentation. The experiments were conducted using 4,000 field images of chili leaves consisting of four balanced disease classes. Model performance was evaluated using accuracy, precision, recall, F1-score, model size, the number of parameters, and inference time. The results demonstrate that the Fine-Tuning strategy consistently achieved the best performance across all evaluated architectures. EfficientNetB0 with Fine-Tuning produced the highest performance, achieving an accuracy and Macro Average F1-score of 92.45%, a model size of 52.90 MB, and an average inference time of 21.96 ms per image. MobileNetV2 with Fine-Tuning served as an efficient lightweight alternative, achieving an accuracy of 91.85%, a model size of 30.80 MB, and the fastest inference time of 17.08 ms per image. In contrast, applying Data Augmentation without Fine-Tuning resulted in decreased performance across all evaluated architectures. The best-performing model was subsequently implemented in a Python-based desktop application prototype for chili leaf disease classification. This study provides empirical evidence regarding the relationship between classification performance and computational efficiency, serving as a reference for selecting appropriate Transfer Learning architectures for field-based chili leaf disease classification. Keywords: Chili Leaf Disease, Convolutional Neural Network, Transfer Learning, Fine-Tuning, Data Augmentation, Image Classification.

Item Type: Thesis (Masters)
Contributors:
ContributionContributorsNIDN/NIDKEmail
Thesis advisorRahajoe, Ani Dijah0012057301anidijah.if@upnjatim.ac.id
Thesis advisorAgussalim, Agussalim0911088501agussalim.si@upnjatim.ac.id
Subjects: T Technology > T Technology (General)
T Technology > T Technology (General) > T58.6-58.62 Management Information Systems
Divisions: Faculty of Computer Science > Magister Information Technology
Depositing User: Bawazir Fadhil Mohammad
Date Deposited: 11 Aug 2026 01:26
Last Modified: 11 Aug 2026 01:26
URI: https://repository.upnjatim.ac.id/id/eprint/58495

Actions (login required)

View Item View Item