A Food Package Recognition Framework for Enhancing Efficiency Leveraging the Object Detection Model

Anwar Ur Rehman, Ignazio Gallo, Pengue Lorenzo

Abstract: Detecting food packaging in images is an important task for many applications in the food industry, including quality control, inventory management, and marketing research. In this paper, to enhance the automation process of the food package detection system, we employ a Raspberry Pi camera to generate a multi-class new Food Package (FP) dataset comprised of 2000 different food package images. We also proposed a Deep Learning based system to precisely detect various food packages using FP and the existing Oktoberfest Food (OF) dataset. To validate our system's efficiency, an experimental study achieved a higher precision, recall, and mean Average Precision (mAP) of 88.0%, 98.1%, and 90.3%, respectively, on the FP dataset test set as compared to OF dataset. It demonstrates the effectiveness of the proposed system in identifying different food package materials, shapes, and sizes. The results also show that the system can accurately detect food packages in real-world food industry scenarios with an average processing rate of 36 Frame Per Second (FPS). The system can also perform well under various lighting conditions, making it suitable for use in different environments. The FP dataset is available publicly in roboflow.