Klasifikasi Motif Batik menggunakan metode Deep Convolutional Neural Network dengan Data Augmentation

Samuel Febrian Tumewu (Program Studi Informatika)
Djoni Haryadi Setiabudi (Program Studi Informatika)
Indar Sugiarto (Program Studi Teknik Elektro)

Jurnal Infra
Vol 8, No 2 (2020)

Related researches before used Convolutional Neural Network (CNN) VGG to classify batik motif which limited only on geometrical pattern and implemented 2 augmentation consist of scale and rotation. Therefore, this research uses CNN Residual Network (Resnet) with 4 augmentation technique on both geometrical and non geometrical batik pattern.This research use (Resnet) as a main architecture of CNN to identify batik pattern. Batik motives for this research are from geometric category which is ceplok, kawung, lereng, nitik, and parang. And from nongeometri category are semen and lunglungan. Furthermore, the dataset will be applied scale, random erase, rotation, and flip augmentation to increase the quantity and variation of batik dataset.The results show that CNN Resnet with data augmentation on training dataset gives accuracy up to 84,52% on Resnet-18 and 81,90% on Resnet-50. furthermore, rotation augmentation adds 4,06%, random erase augmentation adds 9,38%, scale augmentation adds 6,52%, and flip augmentation adds 8,58% on accuracy