<![CDATA[Batik Aceh is an Indonesian intangible cultural entity characterized by high visual morphological complexity and historical depth. Conventional motif identification processes face limitations due to observer subjectivity, high spectral variance, and the intricate geometric structures composing the ornaments. Therefore, this study aims to design an automated digital image classification architecture utilizing the Convolutional Neural Network (CNN) algorithm via VGG16 implementation empowered by Transfer Learning based on ImageNet pre-trained weights. To demystify the deep learning black-box nature, the computational ecosystem is integrated with Explainable AI using Gradient-weighted Class Activation Mapping (Grad-CAM) for visual inference transparency, alongside an interactive Virtual Try-On module deployed on the Streamlit web platform. The classification scope encompasses four distinct classes: Bukan Batik (Non-Batik), Pinto Aceh, Pucok Reubong, and Rencong. The dataset comprises 700 balanced images (175 per class) sourced from the Roboflow public repository and partitioned into 80% training (560 images), 10% validation (68 images), and 10% testing (72 images). The model was compiled with the Adam optimizer (learning rate 10⁻⁴), Categorical Cross-Entropy loss, and supervised by dynamic callbacks (EarlyStopping and ReduceLROnPlateau). Empirical evaluation on 72 unseen test images yielded an overall accuracy of 97.00%, an average precision of 97.50%, a recall of 97.00%, and an F1-score of 97.00%. Specifically, the Pinto Aceh and Pucok Reubong motifs achieved a perfect 100% score across all metrics, while Rencong reached a recall of 88.89%. Comprehensive Black-Box testing across 10 functional interface scenarios and White-Box code logic verification demonstrated 100% operational reliability, confirming that the proposed system is highly robust for digital cultural heritage preservation.]]>
Copyrights © 2026
