• Technical Name
  • AI Deep-Learning Neural Networks for Detecting O2O Food Similarity
  • Operator
  • Feng Chia University
  • Booth
  • Online display only
  • Contact
  • 吳沛儒
  • Email
  • wupj@fcu.edu.tw
Technical Description The technology of heterogeneous product pickups of machine learning is developed for omnichannel new retail logistics to resolve the O2O omnichannel new retail problems. The proposed technology involves a YOLO food category detector, which has the best accuracy (mAP=99.49%) and the second-fastest speed after I/O latency has been considered (12.33 FPS). This technique further integrates Triplet Mapping Network in order to embed the features and calculate the Triplet loss to check whether online pictures of products are similar to the offline versions.
The similarity detection results of the proposed technology to compare online photos of food products against actual offline products can achieve 95% accuracy. The proposed technology can enable couriers to pick up food at brick-and-mortar stores that actually resembles the online pictures of the products that consumers have ordered.
Scientific Breakthrough The fine-grained image analysis has been applied in the research of human identity validation, human face identification, similar face clustering, and detailed classification in the same class. In the process of food image AI research, the fine-grained image analysis has been used to detect the categories of the products on shelves and the categories of the foods in restaurants. However, such technologies have not been applied to an O2O omnichannel new retail ecosystem. The proposed technology successfully integrates YOLO food category detector and triplet mapping network, which has the ability of object detection and the fine-grained image analysis, to check whether online pictures of products are similar to the offline versions.
Industrial Applicability Online to offline (O2O) omnichannel new retail has grown very fast, but returns remain an issue due to differences between product illustrations posted online and the actual goods that consumers receive. Hence, the technology of heterogeneous product pickups of machine learning is developed for omnichannel new retail logistics to resolve the O2O omnichannel new retail problems. This technology can greatly benefit the O2O food ordering new ecosystem because when delivery personnel picks up a food order, they will use the technology to compare the online image with the offline product. This process will therefore result in fewer returns and increased customer satisfaction, which will benefit the platform tremendously.
本網站使用您的Cookie於優化網站。繼續瀏覽網站即表示您同意本公司隱私權政策,您可至隱私權政策了解詳細資訊。