Culinary Classification with Deep Learning: A Study of Pandang Cuisine Using MobileNetV3

Rich tastes, varied ingredients, and visually beautiful feasts define Pandang cuisine. Still, Pandang food is difficult to separate from its seeming homogeneity and professional cooking techniques. Apart from other purposes, computerized classification of these items would be especially beneficial for meal delivery, restaurant automation, nutritional monitoring, and cultural preservation. Accurate food recognition systems help to maintain traditional recipes, thereby enhancing nutritional monitoring, inventory control, food waste reduction, and culinary tourist support. Including depth wise separable convolutions and squeeze-and-excitation modules into MobileNetV3's optimal architecture provides outstanding accuracy with minimal computational cost, which fit for real-time applications on mobile and Edge devices. To improve model performance, this work used a 993 Kaggle picture dataset applying data preprocessing, augmentation, and a fine-tuning technique. Training, validation, and testing subsets were developed to guarantee balanced representation from the data. The carefully calibrated MobileNetV3 model for enhanced generalization consisted in extra thick layers, dropout regularization, and categorical cross-entropy loss. Regarding specific foods like those that Ayam_pop and Dendeng_batokok ranked highly recall, the results revealed an outstanding general accuracy of 90%. Still, outwardly identical things like gulai_ikan and gulai_tunjang brought attention to areas needing repairs and generated classification issues. The efficiency of MobileNetV3 for culinary classification as well as its opportunities to help technical innovation and cultural preservation is underlined in this paper. The results open the path for scalable food recognition systems able of including many cuisines, thereby improving uses in the food sector, health monitoring, and beyond.