Perbandingan Arsitektur Vgg16 dan ResNet50 dalam Klasifikasi Jenis Ulos Batak Toba

Abstract

Ulos is a traditioal cloth typical of the Toba ethnic Batak tribe that is often used during various traditional ceremonies. Ulos has various types, each with different functions in its use. The motifs on each ulos also vary, although at first glance the look similar if not observed in detail. The problem that often arises is the public’s mistake in recognizing the type of ulos, so ulos are often considered the same without knowing the difference. To overcome this, this research aims to classify ulos to facilitate the identification of the types of Toba Batak ulos. This research compares the Visual Geometry Group (VGG16) and Residual Network (ResNet50) models in the classification process. The research was conducted on six types of Toba Batak ulos, namely Ulos Bintang Maratur, Ulos Mangiring, Ulos Ragi Hidup, Ulos Ragi Hotang, Ulos Sadum, and Ulos Sibolang, which are distinguished by the way they are made, namely manual and machine weaving. The test was conducted using images of ulos taken with a camera on full motifs without interference from other objects and full motifs with interference from other objects. The dataset used amounted to 3,189 data divided into training data and test data. The results showed that the classification model using VGG16 and ResNet50 both achieved maximum performance with accuracy, precision, recall, and F1-Score values of 1.0 on the Confusion matrix assessment. In testing on mobile applications, the VGG16 and ResNet50 models were able to identify each type of Toba Batak ulos well at close range. However, at longer distances, VGG16 showed superiority in distinguishing the types of ulos compared to ResNet50.