Abstract

Spices  such as kencur (Kaempferia galanga), turmeric, and ginger play vital roles in Indonesian cuisine; however, the physical similarities of their rhizomes often cause difficulties in identification. This research developed a Flutter-based mobile application to classify these three types of spices using deep learning technology with TensorFlow Lite. The dataset was obtained through video recording and frame extraction using OpenCV, producing image frames for each type of spice. The classification model uses the EfficientNet Lite2 architecture trained with a data split of 90% training, 5% validation, and 5% testing. Testing results showed excellent performance with 99.08% training accuracy (loss: 0.3334), 100% validation accuracy (loss: 0.3007), and 100% testing accuracy (loss: 0.3009). The application provides two classification modes: real-time detection using the camera and classification through the gallery, with a confidence threshold of 0.8 to ensure classification accuracy. Implementation on mobile devices demonstrated good performance in classifying spices under various image capture conditions.