Our method achieves an average recognition accuracy
of 84.9%. This result confirms that the recognition algorithm
performed sufficiently well to be used in the exhibition. Specifically,
recognition precision of peanut is the lowest (0.755).
This implies that sounds of chewing peanut are similar to
chewing other solid food. A possible reason is, for solid
food, varying physical properties directly impact the sounds of
mastication, and consequently influence food type recognition.
Similarly, the sound of swallowing water is characterized with
high frequency signal compared to chewing solid food, which
makes swallowing water effectively recognized, with highest
recall of 0.994. We plan to integrate adaptive methods into
our recognition approach to further improve accuracy in the
future.
• Liquid/Solid Food Classification: To enable the suggestion
for users to intake adequate hydration, it suffices to distinguish
liquid food from solid food. To test the recognition accuracy,
we conduct another set of experiments which includes
1652 solid food chewing events and 181 liquid swallowing
events from 12 subjects. For these experiments, we construct
a decision tree specific for this purpose.
TABLE VIII shows the results of classification between
solid and liquid foods. Only 4 liquid food events are
incorrectly recognized as solid food events among 181 liquid
food events; 5 solid food events are incorrectly recognized as
liquid food events among 1652 solid food events, resulting
in the accuracy for solids and liquids of 99.7% and 97.6%
respectively, which is more than enough to remind the user
with proper hydration intake.