Analisis Pengaruh Data Augmentasi Pada Klasifikasi Tenun Menggunakan Deep Learning Berbasis Convolutional Neural Network

This research develops a classification model of Timorese weaving motifs using Deep Learning method based on Convolutional Neural Network (CNN). Timor's diverse weaving motifs reflect the richness of local culture, but manual classification takes a long time and is prone to subjectivity. To improve model performance, Data Augmentation techniques, such as flipping, rotation, and zooming,, are applied to enrich the variety of pre-processed Timor weaving image datasets. In addition, the CNN model was developed using Transfer Learning techniques to improve training efficiency. Experimental results show that CNN without augmentation achieves 95.00% accuracy, 95.00% precision, 95.08% recall, and 95.04% F1-score, with a computation time of 2.37 minutes at 30 epochs. Meanwhile, applying Data Augmentation increased the model accuracy to 96.66%, precision 96.66%, recall 96.87%, and F1-score 96.77%, and reduced the computation time to 2.11 minutes. Analysis of the effect of augmentation data shows that increasing the variety of images contributes to the improvement of model generalization. Therefore, the use of CNN with Data Augmentation is a more optimal solution in the classification of Timorese weaving motifs. This research has the potential to support cultural preservation as well as the development of an artificial intelligence-based weaving motif identification system.