Leveraging Deep Learning for Cultural Preservation: A Mobile Application for Padang Cuisine

Abstract

Padang cuisine, originating from West Sumatra, Indonesia, is recognized as one of the most widespread traditional food types due to its prevalence in restaurants across the country. Despite the increasing interest in classifying Indonesian food using artificial intelligence, there have been limited studies that have explicitly focused on classifying Padang dishes using deep learning approaches. This study aimed to develop an intelligent mobile application capable of identifying various Padang dishes from images using transfer learning-based convolutional neural networks (CNNs). Four pre-trained CNN architectures—EfficientNetV2M, MobileNetV2, VGG19, and ResNet152V2—were fine-tuned and evaluated on a dataset of Padang food images. This dataset comprised a total of 1,108 images, categorized into nine distinct Padang dishes, collected from both publicly available repositories and original photographs taken for this study. Among these models, ResNet152V2 achieved the best performance after optimization, with a validation loss of 0.4142 and a test accuracy of 91.33%. The optimized model was converted to TensorFlow Lite and deployed as a mobile application, enabling real-time recognition of Padang dishes. This study presented a deep-learning-based mobile solution for recognising nine traditional Padang dishes with high accuracy, demonstrating the potential of AI-driven applications to support culinary heritage preservation and promote cultural tourism in Indonesia.