Nugroho, Muhammad Aryasatya (2026) Model ResNet-50 dengan Selective Kernel untuk Klasifikasi Spesies Merak Menggunakan Citra Bulu. Undergraduate thesis, UPN Veteran Jawa Timur.
|
Text (Cover)
22083010085-cover.pdf Download (1MB) | Preview |
|
|
Text (Bab 1)
22083010085-bab1.pdf Download (155kB) | Preview |
|
![[img]](https://repository.upnjatim.ac.id/style/images/fileicons/text.png) |
Text (Bab 2)
22083010085-bab2.pdf Restricted to Repository staff only until 2 July 2029. Download (5MB) | Request a copy |
|
|
Text (Bab 3)
22083010085-bab3.pdf Restricted to Repository staff only until 2 July 2029. Download (4MB) | Request a copy |
|
|
Text (Bab 4)
22083010085-bab4.pdf Restricted to Repository staff only until 2 July 2029. Download (2MB) | Request a copy |
|
|
Text (Bab 5)
22083010085-bab5.pdf Download (126kB) | Preview |
|
|
Text (Daftar Pustaka)
22083010085-daftarpustaka.pdf Download (151kB) | Preview |
|
|
Text (Lampiran)
22083010085-lampiran.pdf Restricted to Repository staff only until 2 July 2029. Download (1MB) | Request a copy |
Abstract
Peafowl have distinctive feather visual characteristics, but the differences between species can be subtle because they are influenced by color, texture, fiber patterns, and local gradation. Manual identification may be subjective, especially when using only an image of a single feather without body structure as the main distinguishing feature. This study proposes a ResNet-50 model with Selective Kernel to classify peafowl species based on feather images into three classes, namely Blue, Green, and Purple. The dataset used consists of 763 peafowl feather images and was divided using stratified split with a proportion of 80% training data, 10% validation data, and 10% test data. All images were processed into a size of 224 × 224 pixels. This study compares Baseline ResNet-50 and ResNet-50 + Selective Kernel through three training scenarios, namely freeze total, partial fine-tuning, and full fine-tuning. The evaluation was conducted using accuracy, macro precision, macro recall, macro F1-score, confusion matrix, inference time, and statistical testing based on 25 repeated training runs. The main experimental results show that the SK-3 scenario, namely ResNet-50 + Selective Kernel with full fine-tuning, was selected as the final model because it achieved an accuracy of 98.70%, macro precision of 98.92%, macro recall of 98.89%, and macro F1-score of 98.89%. This model produced only one misclassification out of 77 test data, with an average inference time of 127.39 ms/image. However, the statistical test results show that the difference in mean accuracy between SK-3 and Baseline-3 was not statistically significant, with a Welch independent samples t-test p-value of 0.303629 and a paired t-test p-value of 0.294995. The final model was then integrated into a web application named PavoLens to display classification results, confidence score, class probabilities, inference time, and prediction history. The results of this study indicate that the proposed model can be applied to peafowl species classification based on feather images and implemented into a web system, although its performance improvement over the baseline has not been proven significant based on repeated training.
| 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 |
||||||||||||
| Divisions: | Faculty of Computer Science > Departemen of Data Science | ||||||||||||
| Depositing User: | Muhammad Aryasatya Nugroho | ||||||||||||
| Date Deposited: | 02 Jul 2026 07:46 | ||||||||||||
| Last Modified: | 02 Jul 2026 07:46 | ||||||||||||
| URI: | https://repository.upnjatim.ac.id/id/eprint/54370 |
Actions (login required)
 |
View Item |