Improving Food Recognition in Smart Refrigerators with Deep Learning
Author Information
Author(s): Dai Xiaoyan
Primary Institution: Advanced Technology Research Institute, Kyocera Corporation, Yokohama, Japan
Hypothesis
Can enhancing deep learning models improve food recognition accuracy in smart refrigerators?
Conclusion
The study demonstrates that the proposed model significantly improves food recognition accuracy in various real-world conditions.
Supporting Evidence
- The proposed model BroadFPN-YOLACT improved recognition rates significantly compared to existing models.
- Simu-Augmentation effectively generated training data that closely simulates real-world scenarios.
- The study included a custom dataset with over 30 types of small food items to ensure diverse testing conditions.
- BroadFPN-YOLACT achieved a mean Average Precision (mAP) of 95.0% for objects at distances of 60-100 cm.
Takeaway
This study shows how a smart refrigerator can better recognize food using advanced technology, helping to reduce food waste.
Methodology
The study utilized a modified YOLACT model with a new Feature Pyramid Network and a novel data augmentation technique called Simu-Augmentation.
Limitations
The current system primarily focuses on store-bought products and lacks functionality for interpreting expiry dates.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website