Padang Cuisine Classification using Deep Convolutional Neural Networks and Transfer Learning

Abstract

Recent advances in artificial intelligence, particularly deep convolutional neural networks (DCNN), have revolutionized image classification tasks across various domains. However, the application of these techniques to culturally specific food classifications, such as Padang cuisine, remains underexplored. This study aimed to develop a robust model for accurately classifying Padang cuisine using a CNN architecture enhanced with Transfer Learning to address the challenge of distinguishing between visually and texturally similar dishes. The model was trained on a dataset comprising approximately 2500 images of nine distinct Padang dishes, including Rendang and Gulai. Images were preprocessed by resizing, normalizing, and augmented through techniques like rotation and zooming, to enhance model generalizability. A pretrained CNN model was fine-tuned using Transfer Learning to leverage the existing knowledge and improve classification accuracy. The enhanced CNN model achieved an overall accuracy of 92% in classifying Padang cuisine, which significantly outperformed traditional models. Despite this, misclassifications were noted in dishes with similar visual features, such as Sate and certain types of Gulai. The results demonstrate the effectiveness of combining CNNs and transfer learning to accurately classify culturally specific dishes. The findings not only advance the field of food image classification but also have practical implications for automated menu management and culinary education, particularly in preserving and promoting culinary heritage. The integration of AI into culinary heritage documentation represents a significant advancement in preserving cultural diversity and enhancement of technological applications in the culinary industry. Future research should explore larger and more diverse datasets to further refine model accuracy and broaden its applicability to other regional cuisines.