APPLICATION OF SIAMESE NETWORK TO CLASSIFY SMALL DATASET OF THE MOTIFS ON THE CENTER OF SUKHOTHAI CERAMICS

Abstract

This study proposes a novel system for identifying central motifs on Sukhothai ceramics using a Siamese neural network. Traditional motif recognition techniques face challenges due to limited information and incomplete patterns. The Siamese network identifies similarities between images, making it well-suited for comparing unknown motifs to a database of known motifs. The proposed system utilizes a deep convolutional neural network (CNN) for feature extraction. Data augmentation techniques are employed to enrich the dataset and address limitations caused by the small number of available motif images. The Siamese network architecture is trained to compute similarity between image pairs, enabling the system to effectively categorize unknown motifs based on their resemblance to known examples. Experimental results demonstrate that the CNN with a dropout layer 0.3 achieves the highest test accuracy (0.82), indicating its effectiveness in motif identification. This research has potential applications in ceramic conservation, research, and data retrieval, aiding archaeologists, and the public in studying and cataloging Sukhothai ceramic motifs. This approach offers a promising solution for identifying patterns on Sukhothai ceramics, despite data limitations.