PENGELOLAAN DATASET CITRA BATIK DAYAK KALIMANTAN UNTUK KLASIFIKASI MOTIF BERBASIS MACHINE LEARNING DENGAN ROBOFLOW

Dayak Batik from Kalimantan is a significant Indonesian cultural heritage with rich philosophical meaning and complex visual characteristics, making it suitable for digitalisation through computer vision technology. A major challenge in developing batik image analysis is the lack of structured, well-annotated datasets for machine learning, since documentation of Dayak Batik images remains limited compared to curated Javanese Batik datasets. This study aims to manage a structured image dataset of Dayak Batik based on visual attributes using the Roboflow platform to support motif classification analysis. The method involved reusing 1,200 images across five classes based on regional origin, namely West Kalimantan, East Kalimantan, South Kalimantan, North Kalimantan, and Central Kalimantan, with 240 images per class. The stages included uploading the dataset to Roboflow, splitting it into training, validation, and testing sets using a stratified 70:20:10 ratio, preprocessing through Auto-Orient and resizing to 224x224 pixels, annotation and labelling based on regional attributes using Roboflow Annotate, and data augmentation through flipping, cropping, and rotation applied to the training set. The augmentation process increased training data from 838 to 2,514 images, bringing the total dataset to 2,876 images, while validation and testing sets of 241 and 121 images remained unaugmented to ensure objective evaluation. A Convolutional Neural Network model based on ResNet-50 was trained for 28 epochs and achieved a best validation accuracy of 87.1 percent, demonstrating competitive performance compared to similar studies. These findings confirm the dataset supports the model learning process, while contributing to cultural preservation and advancement of image classification technology.