Automated Pattern Recognition of Inaul Traditional Fabrics Using Convolutional Neural Networks

The Philippines hosts over 50 unique traditional weaving practices across its regions. Inaul, a handwoven fabric of the Maguindanaon and Iranun peoples of Mindanao, is recognized for its cultural significance and 63 unique pattern variations. Despite its local prominence and growing recognition to date, Inaul remains underdocumented, particularly in reliable digital sources. This study contributes to cultural preservation by creating a digital profile of Inaul and identifying the optimal lightweight convolutional neural network model among MobileNetV3, EfficientNet-Lite, and MobileViT for automated pattern recognition. Using transfer learning and stratified 5-fold cross-validation, MobileViT achieved the highest accuracy from test set of 72.3%, followed by EfficientNet-Lite at 62.9% and MobileNetV3 at 48.0%. Further optimization through enhanced augmentation and hyperparameter tuning improved MobileViT's accuracy from validation set to 89.33% and F1-score to 0.814. This work digitizes Inaul's textile heritage, empowers indigenous artisans by facilitating pattern recognition and documentation, and demonstrates the role of artificial intelligence in preserving the cultural legacy of the Maguindanaon and Iranun peoples.