RECOGNITION OF BOTI WOVEN FABRIC MOTIFS FROM EAST NUSA TENGGARA USING DETECTION OBJECT AND FEW-SHOT LEARNING APPROACH

• Main Author: Gusriana, Prana
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/80979
• Institution: Institut Teknologi Bandung

Weaving is one of Indonesia’s cultural heritages that has been around for generations and has very diverse manufacturing techniques and motifs. Currently, woven fabrics can support the creative economy of local communities by becoming the main commodity in several regions in Indonesia. However, it is possible that there could be counterfeiting, piracy, and claims of woven fabric by irresponsible parties which would be detrimental to local weaving craftsmen and the general public. One of the reasons for this is the lack of general public awareness of the diversity of weaving cultures. Boti Village is one of the traditional villages in South Central Timor Regency, East Nusa Tenggara. The people of Boti Village still maintain their ancestral customs and weaving has become a part of the life of the people of Boti Village. Current developments in the field of artificial intelligence can be utilized in the field of art to be able to recognize types of motifs typical of a region with the aim of increasing general public awareness of the diversity of weaving culture and also as an effort to preserve weaving culture. Research related to the recognition of typical woven cloth motifs of a region has previously been carried out a lot. However, there are still several problems that need to be resolved, namely problems related to woven fabric datasets which are still quite difficult to obtain and also problems related to woven fabric motifs which can continue to increase due to the creativity and cultural activities of local communities. Therefore, in this final project, a model for recognizing motifs from woven cloth typical of Boti Village will be built by combining the object detection approach using YOLOv8 and the few-shot learning approach using Siamese Neural Network to overcome those problems. In this final project, hyperparameter tuning and different support set selection experiments will be carried out to obtain the best model performance as measured using the metrics precision, recall, mAP50, and mAP50-95. The proposed method can be trained with a small dataset but has performance that is not much different or even better than the current state-of-the-art object detection method in recognizing the 4 main motifs of the typical woven cloth of Boti Village with the best performance of precision, recall, mAP50, and mAP50-95 were 98.01%, 100%, 99.31%, and 88.17% respectively. In addition, the proposed method is also flexible enough to be able to recognize new types of motifs quite well.