Classification of Papuan Batik Motifs Using Deep Learning and Data Augmentation

Papuan Batik motifs began to appear in 1984 and are only known in general with one designation, namely Papuan batik, even though these various motifs can be classified according to the area of origin of manufacture, culture and flora and fauna that have meaning. local wisdom of the community. The ability to recognize every batik motif from Papua requires experience and knowledge from certain circles, so that knowledge and meaning are maintained, a tool is needed that can classify various batik motifs from Papua. This study uses four classes of datasets, namely Cendrawasih Motifs, Raja Ampat Motifs, Tifa Honai Motifs and Asmat Motifs which are non-geometric. It is proposed to classify these motifs using deep learning using Vgg16 and Resnet50 architectures with fine tuning, to add data with various combinations in order to obtain better performance. The test results showed that without data augmentation on the VGG16 architecture, an accuracy of 78.79% was obtained and the Resnet50 architecture obtained an accuracy of 81.82% with several combinations of augmentation techniques giving the same better results without data augmentation with the results on the VGG16 architecture giving 84 results., 85% and on the Resnet50 architecture it gives a yield of 87.88%.