Klasifikasi Motif Kain Jumputan Palembang Menggunakan Metode CNN dengan Arsitektur Resnet-50

Abstract

This study develops an automated classification system for Palembang jumputan textile motifs based on computer vision to address inter-motif pattern similarities that often challenge non-expert users and hinder the digital documentation of textile cultural heritage. Unlike traditional textile studies that typically employ generic Convolutional Neural Networks (CNNs), this research applies transfer learning using the ResNet-50 architecture on a primary dataset consisting of five motif classes: lilin, titik 7, titik 9, bunga tabur, and akoprin daun. The dataset is divided into training, validation, and testing sets, followed by preprocessing and image augmentation to enhance data variability. The model is trained with learning rate tuning, and the best configuration achieves a training accuracy of 95.57%, a validation accuracy of 87.33%, and a testing accuracy of 88%. Evaluation using a classification report and confusion matrix indicates excellent performance for the titik 9 and bunga tabur motifs, with precision and recall values approaching 1.00, while misclassifications still occur in the lilin motif due to visual similarity. These results confirm the effectiveness of ResNet-50 for jumputan motif classification and support cultural preservation through faster and more consistent motif identification.