Classification of Batik Patterns Using Inception-ResNetV2 with Data Augmentation

Document Type

Article

Publication Date

1-1-2026

Abstract

One of Indonesia’s cultural heritages with significant artistic and historical value is batik. The background of this research is the diversity of cultures and customs in various regions of Indonesia, which is reflected in the diverse batik patterns currently used. Classifying batik patterns is very important in preserving, recognizing, and promoting batik as a valuable cultural asset of the country. The problem statement is how classification facilitates the identification of the history, significance, and characteristics of each batik pattern currently in use. This problem can be overcome by applying deep learning to help identify batik patterns throughout the archipelago. The key conclusion is that a deep learning strategy is necessary, which involves training with extensive visual data. The methods used include deep learning by utilizing the Inception-ResNetV2 architecture. Its contribution is adopting an architecture that is designed to minimize the number of parameters and improve computational performance. The network can perform more efficiently overall when combined with residual connections from ResNetV2. Various types of batik and classes from the archipelago form the dataset. After the calculation was carried out for 9 min and 6 s, a batik pattern model was obtained with an average accuracy of 98.19%, precision of 98.20%, recall of 98.19%, and F1-Score of 98.16%. Mean Squared Error (MSE) 0.0023, Root Mean Squared Error (RMSE) 0.0483, Mean Absolute Error (MAE) 0.0035. The experimental data were then used to test the confidence level, achieving an average accuracy of 76–99%.

Keywords

batik pattern, data augmentation, deep learning, Inception-ResNetV2, process innovation, product innovation

Publication Title

Journal of Advances in Information Technology

DOI

10.12720/jait.17.1.42-54

Volume

17

Issue

1

First Page

42

Last Page

54

Share

COinS