Research Article
Deep Learning Model for Real-Time Multi-Class Detection on Food Ingredients Using Yolov4 Algorithm
@INPROCEEDINGS{10.4108/eai.11-10-2021.2319569, author={Syafli Shovin and Tristyanti Yusnitasari and Teddy Oswari and Reni Diah Kusumawati and Nurasiah Nurasiah}, title={Deep Learning Model for Real-Time Multi-Class Detection on Food Ingredients Using Yolov4 Algorithm}, proceedings={Proceedings of the 6th Batusangkar International Conference, BIC 2021, 11 - 12 October, 2021, Batusangkar-West Sumatra, Indonesia}, publisher={EAI}, proceedings_a={BIC}, year={2022}, month={8}, keywords={deep learning; transfer learning; yolov4; food ingredients detection}, doi={10.4108/eai.11-10-2021.2319569} }- Syafli Shovin
Tristyanti Yusnitasari
Teddy Oswari
Reni Diah Kusumawati
Nurasiah Nurasiah
Year: 2022
Deep Learning Model for Real-Time Multi-Class Detection on Food Ingredients Using Yolov4 Algorithm
BIC
EAI
DOI: 10.4108/eai.11-10-2021.2319569
Abstract
The position of food ingredients that may be piled and the similarity in terms of the shape, color, and texture of food ingredients become challenges to build a deep learning model that can perform optimal detection task on food ingredients. Therefore, the food ingredients will be trained and detected using the YOLOv4 algorithm because of its good performance. The data augmentation techniques from YOLOv4 algorithm also applied in this research to improve the variation and amounts of the collected dataset. To make the training more efficient, this research utilizes transfer learning method to adopt knowledge from YOLOv4 pre-trained model. The approach used in this research successfully creates a deep learning model for real-time multi-class detection on food ingredients with a reasonably good performance. The model shows performance with mAP@0.50 value of 84.90% and an average IoU of 72.77%.