Syamjovanka, Revelin Putri (2026) Optimization of Soybean Quality Classification Using EfficientNet-B0 with Mixup and Bayesian Hyperparameter Optimization. Undergraduate thesis, UPN Veteran Jawa Timur.
|
Text (Cover)
Cover.pdf Download (3MB) | Preview |
|
|
Text (Bab 1)
Bab 1.pdf Download (140kB) | Preview |
|
![[img]](https://repository.upnjatim.ac.id/style/images/fileicons/text.png) |
Text (Bab 2)
Bab 2.pdf Restricted to Repository staff only until 29 June 2029. Download (6MB) |
|
|
Text (Bab 3)
Bab 3.pdf Restricted to Repository staff only until 29 June 2029. Download (5MB) |
|
|
Text (Bab 4)
Bab 4.pdf Restricted to Repository staff only until 29 June 2029. Download (3MB) |
|
|
Text (Bab 5)
Bab 5.pdf Download (124kB) | Preview |
|
|
Text (Daftar pustaka)
Daftar pustaka.pdf Download (154kB) | Preview |
|
|
Text (Lampiran)
Lampiran.pdf Restricted to Repository staff only until 29 June 2029. Download (421kB) |
Abstract
products. Traditionally, quality assessment has been performed manually, making the results inefficient and highly dependent on subjective judgment. To address this limitation, an artificial intelligence (AI)-based system is required to automate and improve the classification of soybean quality. This study employs the EfficientNet-B0 deep learning architecture to categorize soybeans into five distinct quality classes: whole, spotted, immature, broken, and damaged-skin beans. To enhance model performance the MixUp data augmentation method was applied with bayesian optimization was utilized to hyperparameter tuning. The model was evaluated under three distinct scenarios first, using EfficientNet-B0 as the baseline; second, incorporating MixUp data augmentation and third, combining MixUp with Bayesian Optimization. The secondary dataset used for model development was sourced from Kaggle, comprising a total of 5,513 images. Model performance was evaluated using accuracy, precision, recall, and F1-score metrics. The results demonstrated that the baseline model achieved an accuracy of 91.67%, whereas the model with MixUp alone showed a slightly lower accuracy of 89.31%. However, incorporating Bayesian Optimization significantly improved the performance of the MixUp configuration, yielding an accuracy of 90.94%, a precision of 90.98%, recall of 90.97%, and F1-score of 90.92%. These findings indicate that hyperparameter tuning effectively optimizes the MixUp technique during training. Finally, the optimized model from the third scenario was deployed into a web-based system using FastAPI. To evaluate the system's real-world functionality, web testing was conducted using 15 primary images collected directly. The system successfully automates the entire sorting pipeline, from image ingestion to displaying the predicted class alongside its confidence score. Consequently, the developed system offers a more efficient, rapid, and objective approach to assessing soybean seed quality.
| Item Type: | Thesis (Undergraduate) | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Contributors: |
|
||||||||||||
| Subjects: | Q Science > QA Mathematics > QA75 Electronic computers. Computer science Q Science > QA Mathematics > QA76.6 Computer Programming |
||||||||||||
| Divisions: | Faculty of Computer Science > Departemen of Informatics | ||||||||||||
| Depositing User: | Revelin Putri Syamjovanka | ||||||||||||
| Date Deposited: | 30 Jun 2026 02:23 | ||||||||||||
| Last Modified: | 30 Jun 2026 02:23 | ||||||||||||
| URI: | https://repository.upnjatim.ac.id/id/eprint/54299 |
Actions (login required)
 |
View Item |