Development of a Balinese Endek Classification Application Using Convolutional Neural Networks

Endek is a unique Balinese handicraft made using a special technique to produce beautiful and distinctive patterns compared to other Indonesian fabrics. Endek has immense potential both domestically and internationally, as evidenced by its presence at international events such as Paris Fashion Week. However, the Bali Provincial Research and Innovation Agency states that there is a need for efforts to help preserve traditional Balinese Endek, such as an application that makes it easy for the public and tourists to identify types of traditional Balinese Endek motifs in order to counter the existence of Endek motif imitations. Therefore, the author conducted research to create a mobile application that can detect the type of Balinese Endek using machine learning technology. This application will be built using the extreme programming method, and the convolutional neural network algorithm will be used to create the Endek type classification feature to obtain good accuracy. In this research, eight classes were used, where seven classes are types of Balinese Endek motifs and one additional class is a random image that is not Endek to be able to detect images that are not Endek or Endek fabrics that have not been included in the dataset. The application created can display the results of Endek classification which also contains additional information about the type of Endek. This application is made using Flutter, a framework developed by Google. The advantage of Flutter is that it can produce cross-platform apps with high performance. Model creation uses TensorFlow, which compares the performance of three types of transfer learning models: VGG16, VGG19, and MobileNetV3Small. Of the three models used, the MobileNetV3Small model achieved the highest accuracy of 90% in the testing results, so the MobileNetV3Small model was used in the application.