SipBit: A Sensing Platform to Recognize Beverage Type, Volume, and Sugar Content Using Electrical Impedance Spectroscopy and Deep Learning

Abstract

We present SipBit, a sensing platform that digitally recognizes beverages and their attributes, an essential component in facilitating novel human-food interactions. SipBit consists of an electrical impedance measurement unit and a recognition method based on deep learning techniques. First, impedance measurements of a beverage are acquired using Electrical Impedance Spectroscopy. Then, a multi-task network cascade algorithm was employed to identify eight different beverage types in various volume levels and sugar concentrations. Results show that the multi-task network cascade discriminates beverage types with an accuracy of 96.32%, and estimates volumes with a root mean square error of 13.74ml and sugar content with a root mean square error of 7.99gdm-3. Future work will include: 1) developing utensils embedded with SipBit for automatic beverage and attribute recognition, and 2) further developing SipBit to recognize additional beverage types and their attributes, thus enabling a new avenue for designing human-food interactive technologies.