EUVIMEDEUVIMED entdecken
Open Access · CC BY-SA 4.0

ResNet-50 for Fine-Grained Peafowl Classification Using Single-Feather Images

Muhammad Aryasatya Nugroho, Wahyu Syaifullah Jauharis Saputra, Shindi Shella May Wara

bit-Tech · 10.08.2026 · 9 (1) · 173-185

DOI: 10.32877/bt.v9i1.3820 ↗

Kurz gefasst

Fachartikel in der Zeitschrift bit-Tech, online erschienen am 10.08.2026. Verfasst von Muhammad Aryasatya Nugroho und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-SA 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Identifying peafowl species from single-feather images is a fine-grained visual classification problem in which species-level differences must be inferred from subtle local characteristics such as color gradation, texture, and feather fiber patterns, without relying on complete body morphology. Although deep learning has been widely used for bird and biological image classification, the extent to which standard transfer learning with ResNet-50 can provide a reliable baseline for peafowl identification from single-feather images remains underexplored. This study empirically evaluates ResNet-50 fine-tuning strategies for classifying peafowl species using single-feather images. The primary contribution of this work lies in establishing an initial empirical baseline for feather-based peafowl classification and analyzing how different degrees of ResNet-50 backbone adaptation influence classification performance. A primary dataset consisting of 773 images from three peafowl species, namely blue, green, and purple peafowl, was collected and divided into training, validation, and test sets using an 80:10:10 stratified split. Three transfer learning scenarios were evaluated: freeze total, partial fine-tuning, and full fine-tuning. The results showed that full fine-tuning achieved the highest test accuracy of 98.70%, with macro precision, recall, and F1-score values of 98.92%, 98.89%, and 98.89%, respectively. Partial fine-tuning achieved 96.10% accuracy with a macro F1-score of 96.67%, while freeze total achieved 88.31% accuracy with a macro F1-score of 89.06%. These findings indicate that increasing the degree of ResNet-50 backbone adaptation improves classification performance for single-feather peafowl images. This study provides a focused empirical baseline for future research on single-feather fine-grained recognition, particularly through more diverse datasets, external validation, and comparison with attention-based or transformer-based architectures.

Zitieren

APA

Nugroho, M. A., Saputra, W. S. J., & Wara, S. S. M. (2026). ResNet-50 for Fine-Grained Peafowl Classification Using Single-Feather Images. bit-Tech, 9(1), 173-185. https://doi.org/10.32877/bt.v9i1.3820

Vancouver

Nugroho MA, Saputra WSJ, Wara SSM. ResNet-50 for Fine-Grained Peafowl Classification Using Single-Feather Images. bit-Tech. 2026;9(1):173-185. doi:10.32877/bt.v9i1.3820

BibTeX
@article{Nugroho2026,
  author = {Nugroho, Muhammad Aryasatya and Saputra, Wahyu Syaifullah Jauharis and Wara, Shindi Shella May},
  title = {ResNet-50 for Fine-Grained Peafowl Classification Using Single-Feather Images},
  journal = {bit-Tech},
  year = {2026},
  volume = {9},
  number = {1},
  pages = {173--185},
  doi = {10.32877/bt.v9i1.3820}
}
Kostenlos starten →