Classification of Central Lombok Songket Motifs Using Convolutional Neural Network

Abstract

Songket is a traditional Indonesian textile renowned for its high aesthetic and symbolic value. In Central Lombok, songket fabrics exhibit diverse motifs that reflect the region’s rich cultural identity. However, manual classification of these motifs is time-consuming and requires expert knowledge, limiting its scalability for digital preservation and cultural heritage documentation. This study proposes an automated classification system for Central Lombok songket motifs using a Convolutional Neural Network (CNN) based on the GoogLeNet (InceptionV3) architecture. The dataset comprises seven distinct motifs—namely Iket, Subhanale, Alang, Subhanale-Laek, Pangkeros, Mawar, and Merak—collected directly from the Sukarara weaving center. A total of 7000 images were used for training and 1400 for testing. The model was trained with 75% and 80% proportions of the dataset and evaluated using accuracy, precision, recall, F1-score, and confusion matrix metrics. Experimental results indicate that the CNN model achieved 97.97% accuracy with 75% training data and improved to 98.91% with 80% training data. These findings demonstrate that GoogLeNet is highly effective in classifying traditional songket motifs with high accuracy and computational efficiency. The proposed system offers significant potential for supporting the digital preservation of cultural assets and facilitating the development of AI-based tools for heritage documentation and creative industries.