Klasifikasi Kain (Uis) Suku Karo Menggunakan Algoritma MobileNetV2 Berbasis Website

Abstract

Karo cloth (Uis) is a high-value cultural heritage whose type identification is still often done manually, making it prone to errors and risking hampering preservation efforts. This research aims to design and build a website-based Karo Uis classification system that is accurate and easily accessible by implementing deep learning methods. The method used is MobileNetV2 Convolutional Neural Network (CNN) architecture due to its high efficiency. The model was trained and tested using a dataset consisting of 1,572 total images divided into five Uis classes, namely Bekabuluh, Gatip Dilaki, Julu Diberu, Jongkit Tudung, Uis Nipes, and Uis Gara. The system test results show excellent performance, where the MobileNetV2 model managed to achieve an accuracy rate of 90.65%. The performance of this model is also supported by macro average values for precision of 97%, recall 96%, and F1-Score 96%. These results prove that the MobileNetV2 architecture is able to classify Uis Karo types effectively based on their motif and color features. The developed system successfully functions as a digital media to help people recognize the types of uis automatically, which in turn can increase information accessibility and support efforts to preserve the cultural heritage of the Karo Tribe