Klasifikasi Karakter Wayang Ramayana Menggunakan Cross Vision Transformer Dengan Optimasi Adam

Wirayudha, Lugas Fernanda (2026) Klasifikasi Karakter Wayang Ramayana Menggunakan Cross Vision Transformer Dengan Optimasi Adam. Undergraduate thesis, Universitas Pembangunan Nasional Veteran Jawa Timur.

	Text (Cover) 20081010144-coverfix.pdf Download (877kB)
	Text (Bab 1) 20081010144 Bab-1.pdf Download (263kB)
	Text (Bab 2) 20081010144 Bab-2.pdf Restricted to Repository staff only until 22 January 2029. Download (418kB)
	Text (Bab 3) 20081010144 Bab-3.pdf Restricted to Repository staff only until 22 January 2029. Download (358kB)
	Text (Bab 4) 20081010144 Bab-4.pdf Restricted to Repository staff only until 22 January 2029. Download (2MB)
	Text (Bab 5) 20081010144 Bab-5.pdf Download (238kB)
	Text (Daftar Pustaka) 20081010144-Dapus.pdf Download (227kB)
	Text (Lampiran) 20081010144-Lampiran.pdf Restricted to Repository staff only until 22 January 2029. Download (1MB)

Abstract

The development of deep learning technology has enabled the automation of image classification with high accuracy, including in the context of the preservation of digital culture. Wayang Ramayana as a cultural heritage rich in complex visual characters requires an introductory approach that is able to capture the details of ornaments and distinctive features at various scales. Conventional Convolutional Neural Networks (CNNs) are often limited in capturing the global contextual relationships between visual features. To address this, this study proposes the use of a Cross Vision Transformer (CrossViT) that integrates a cross-scale attention mechanism, complemented by Adam optimization to accelerate model convergence. This study uses a dataset of 1043 images of Ramayana puppets consisting of 60 characters, collected through live shooting and digital sources. The data was then augmented to 7873 images to increase variation and reduce overfitting. The CrossViT model is designed with two small (16×16) and large (32×32) patch processing branches, equipped with cross-attention for multi-scale feature integration. The training was conducted with a learning rate of 0.0001, a batch size of 32, and a loss categorical cross-entropy function. The model parameters were optimized using the Adam algorithm with decay rates of β1=0.9 and β2=0.999, as well as epsilon 1e-7 for numerical stability. The results of the evaluation showed that the model achieved 89% accuracy on the test data, with an average precision of 0.90, recall of 0.88, and an F1-score of 0.88. Some characters such as Gareng, Petruk, and Semar achieved perfect scores (F1-score 1.00), while characters such as Pratalamariam and Resi Subali still need refinement. Comparative analysis showed that the use of data augmentation and Adam optimization significantly improved accuracy compared to training without both components. These findings prove that the CrossViT architecture with parameter optimization using Adam is effective in classifying Ramayana puppet characters that have high visual complexity, and has the potential to be developed as a support system for digitization and preservation of puppet culture. Keywords: Image Classification, Cross Vision Transformer, Adam Optimization, Deep Learning Parameters, Wayang Ramayana, Digital Cultural Preservation

Item Type:

Thesis (Undergraduate)

Contributors:

Contribution	Contributors	NIDN/NIDK	Email
Thesis advisor	Rahmat, Basuki	NIP19690723 202121 1 002	basukirahmat.if@upnjatim.ac.id
Thesis advisor	Endah Wahanani, Henni	NIP19780922 202121 2 005	henniendah.if@upnjatim.ac.id

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:

Lugas Wirayudha Fernanda

Date Deposited:

29 Jan 2026 07:28

Last Modified:

29 Jan 2026 08:21

URI:

https://repository.upnjatim.ac.id/id/eprint/48941