Indonesian food identification and detection in the smart nutrition box using faster-RCNN

Food detection and localization are useful to recognize consumer preference and the amount of consumption. In the end, they can be integrated to food production to reduce oversupply and food waste. In this paper, the Faster R-CNN model is exploited to locate and classify food objects in a tray box as a first stage for food loss quantization. To perform the objective, the model was trained on images scraped from the Internet. Then, performance evaluation was conducted on both complex image and tray box image. On tray box images data, the model was able to achieve Average Precision of 0.455 and Average Recall of 0.628 for IoU 0.50:0.95.