Angga, Reza Putri (2026) Hibrida EfficientNetB4 dan Vision Transformer untuk Klasifikasi Penyakit Tanaman Cabai Menggunakan Citra Daun. Undergraduate thesis, UPN Veteran Jawa Timur.
|
Text (Cover)
22083010006-cover.pdf Download (1MB) | Preview |
|
|
Text (Bab 1)
22083010006-bab1.pdf Download (157kB) | Preview |
|
![[img]](https://repository.upnjatim.ac.id/style/images/fileicons/text.png) |
Text (Bab 2)
22083010006-bab2.pdf Restricted to Repository staff only until 13 May 2028. Download (1MB) | Request a copy |
|
|
Text (Bab 3)
22083010006-bab3.pdf Restricted to Repository staff only until 13 May 2028. Download (3MB) | Request a copy |
|
|
Text (Bab 4)
22083010006-bab4.pdf Restricted to Repository staff only until 13 May 2028. Download (9MB) | Request a copy |
|
|
Text (Bab 5)
22083010006-bab5.pdf Download (126kB) | Preview |
|
|
Text (Daftar Pustaka)
22083010006-daftarpustaka.pdf Download (142kB) | Preview |
|
|
Text (Lampiran)
22083010006-lampiran.pdf Restricted to Repository staff only Download (2MB) | Request a copy |
Abstract
Chili pepper (Capsicum annuum L.) is a commodity with a high consumption rate in Indonesia. However, its productivity often declines due to plant pests and diseases (PPDs), which are indicated by visual changes in the leaves. Manual disease identification is often difficult due to the similarity of symptoms across disease categories and limited access to agricultural extension workers. This study proposes a hybrid method combining EfficientNetB4 and Vision Transformer (ViT) to classify chili plant diseases using leaf images. EfficientNetB4 is used to extract local features, while Vision Transformer models global relationships between features to produce more comprehensive feature representations. The dataset consists of 5,600 images, including 4,000 secondary and 1,600 primary data, covering three disease categories, yellowish, curl leaf, spot leaf and one healthy category, healthy. This study evaluates the model’s generalization ability across five testing scenarios based on combinations of secondary and primary data. The baseline scenario achieved 98.25% accuracy but decreased to 82% when tested using 100% primary data without retraining, indicating a domain shift. The gradual addition of primary data improved model performance, with accuracies of 99.07% in scenario two, 99.17% in scenario three, and 98.67% in scenario four. Although the differences are not statistically significant, scenario three shows the best numerical performance with the highest evaluation metrics and lowest error rate. The model is implemented in a web-based platform called BotaniQ using Flask, enabling disease classification along with symptom information and organic treatment recommendations.
| Item Type: | Thesis (Undergraduate) | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Contributors: |
|
||||||||||||
| Subjects: | Q Science > Q Science (General) Q Science > QA Mathematics > QA76.6 Computer Programming Q Science > QA Mathematics > QA76.87 Neural computers T Technology > T Technology (General) |
||||||||||||
| Divisions: | Faculty of Computer Science > Departemen of Data Science | ||||||||||||
| Depositing User: | Reza Putri Angga | ||||||||||||
| Date Deposited: | 13 May 2026 08:05 | ||||||||||||
| Last Modified: | 13 May 2026 08:05 | ||||||||||||
| URI: | https://repository.upnjatim.ac.id/id/eprint/51760 |
Actions (login required)
 |
View Item |