*4.1. Sensors*

As mentioned in Section 1, we only used low-power sensors attached to the smartphone and a wrist-wearable device to consider constraints of power consumption and obtrusiveness of the user. The distribution rate of the wrist-wearable device is much higher than other forms of wearable devices

and is in a natural position to collect daily life data consistently. Moreover, as we use our hands to eat something, the wrist is an appropriate position to collect food intake-related movement and the position of hands, and parametric temperature or humidity. We combined the four kinds of sensors for the wrist-wearable device (Figure 2), which are composed of MPU-9250 motion sensor of InvenSense (Seoul, republic of Korea), BME280 environment sensor of Bosch (Seoul, republic of Korea), and APDS-9900 illumination sensor of Avago Technologies (Seoul, republic of Korea). Table 4 shows the type of sensors with their power consumption and collecting frequency. The device can collect data continuously for about 6 h without charging.

**Figure 3.** An overview of the proposed method.


**Table 4.** Sensors attached to wrist-wearable devices for recognition.
