TRANSFORMING WOVEN IKAT FABRIC: ADVANCED CLASSIFICATION VIA TRANSFER LEARNING AND CONVOLUTIONAL NEURAL NETWORKS

Abstract

The woven ikat fabric from Nusa Tenggara Timur is a local wisdom that must be preserved. Due to its vast array of motifs, users often encounter challenges in its recognition. For this study, the TenunIkatNet dataset was employed. One prominent recognition method involves classification based on the motif type and geographical origin. The efficacy of the classification is heavily contingent upon the method of extraction employed. The Convolutional Neural Network (CNN) method is used for feature extraction and classification processes. This research compares the classification performance of the VGG16 baseline model and the proposed model. The proposed model modifies the baseline at the fully connected layer and the training process from the first convolution layer. Incorporating elements such as Global Average Pooling (GAP), Batch Bormalization (BN), and Dropout has proven instrumental in mitigating overfitting. The transfer learning strategy is used for feature extraction and classification because the model has been intelligently trained on a large dataset. The research findings unequivocally indicate that the performance of the modified model supersedes that of the baseline model. Based on the evaluation metrics, the proposed model is superior to the baseline model with precision, recall, accuracy, and F1-score, respectively 98.73%, 98.54%, 98.54%, and 98.53%