Classification of Ornament or Pepatran in Balinese Traditional Architecture Pattern Using Learning Vector Quantization Algorithm
I Gusti Agung Gede Arya Kadyanan, Nyoman Gunantara, Ida Bagus Gede Manuaba, Komang Oka Saputra, Wayan Firdaus Mahmudy · 2024
Bali is very famous for traditional architectur. Not everyone is able to identify the various Balinese ornament patterns or Balinese pepatran just based on their visual attributes. This is because the patterns have high similarities even though they have different classes. To solve this problem, a machine learning model was developed using the LVQ method. In this model, preprocessing is carried out involving scaling, grayscale conversion and sobel edge detection to obtain strong features and enable the model to obtain high accuracy. A total of 210 pepatran images, which 168 pepatran images or 80% of the images used as training data and 42 or 20% of the pepatran images used as testing data. Twelve classes representing each used pepatran comprise the data in this research. The model has been shown to be effective in identifying Balinese pepatran patterns. Performance testing and Accuracy Testing were two testing techniques used to evaluate the successfully constructed system. An accuracy of 92.86% is obtained with 40 Balinese pepatran that can be correctly recognized based on accuracy testing using the influence of changes in learning rate, maximum epoh, and epsilon so that the best learning rate parameters are 0.06, epoh 20, and epsilon 0.0001 from 42 testing data used. The outcomes demonstrated that the system was able to identify and categorize traditional Balinese pepatran, with the image being classed and recognized depending on the outcome of Sobel feature extraction.