Deep learning for recognition of Javanese batik patterns

• Main Authors: Mardani, Danis Aditya, Pranowo, Santoso, Albertus Joko
• Format: Conference Proceeding
• Language: English
• Subjects: Artificial neural networks; Case depth; Convolution; Datasets; Deep learning; Feature extraction; Image classification; Machine learning; Neural networks; Pattern recognition; Programming languages; Training
• ISSN: 0094-243X; 1551-7616
• DOI: 10.1063/5.0000686

Batik is one of the cultural heritages of a special Indonesian nation. Because of its diversity and uniqueness on October 2, 2009, Batik was first established as Masterpieces of the Oral and Intangible Heritage Humanity by UNESCO. To maintain sustainability, continuous research is needed. Although the topic of research on batik is already common, the introduction of batik patterns still has challenges that need to be resolved. One of the challenges of pattern recognition is in terms of classifying batik motifs. To simplify the work of computers in classifying, in this case the implementation of deep learning is needed by using the convolutional neural network (CNN) method. The convolutional neural network (CNN) method is one of the architectures in deep learning, this method is more effective for classifying images such as batik patterns because the convolutional neural network method has a convolution operation. In this operation the image will be extracted every feature so that it can produce patterns that can facilitate classification. In the process of training the convolutional neural network method requires heavy computation and not a short amount of time, therefore the use of GPU performance is needed to speed up the training time. The experimental process begins by compiling five classes of data sets of batik images, the class consisting of batik parang rusak, batik kawung, batik nitik, batik ceplok, and batik lereng with a total of 750 batik images as data sets. The data set was then trained using the Python programming language and GPU CUDA. The test results using cross-validation can achieve an accuracy of 90.14%. So that the results of the above tests can be concluded that deep learning using the CNN method can be used to classify batik patterns well.