FINE-TUNING RESNET50V2 WITH ADAMW AND ADAPTIVE TRANSFER LEARNING FOR SONGKET CLASSIFICATION IN LOMBOK

This study aims to develop a classification system for traditional Lombok songket fabric patterns using the ResNet50V2 architecture, optimized through fine-tuning and the AdamW optimizer. The data were collected directly from songket artisans in Lombok and categorized into three groups based on the origin of the patterns: Sade, Sukarara, and Pringgasela. The model was trained with data augmentation techniques, including rotation, shifting, and zooming, to increase data diversity. During the training process, fine-tuning was applied to the last layer of ResNet50V2, and optimization was performed using AdamW with a learning rate of 0.0001. The model was evaluated using a confusion matrix, classification report, and analysis of accuracy and loss. The experimental results showed that the model achieved 100% accuracy at the 15th epoch. Furthermore, experiments with different parameters (epochs, batch size, and learning rate) demonstrated that the 15th epoch provided the best results with 100% accuracy, while using higher epochs (30 and 40) did not necessarily yield better outcomes. This model is effective in identifying songket fabric patterns with good classification results for each class. Although the results are excellent, increasing the dataset size and exploring more complex model architectures could further enhance performance. Overall, this study demonstrates the significant potential of deep learning technology in classifying songket patterns with reliable accuracy in real-world applications.