Title:	ANALYSIS OF EATING HABITS USING SOUND INFORMATION FROM A BONE-CONDUCTION SENSOR
Author(s):	Hao Zhang, Guillaume Lopez, Masaki Shuzo, Jean-Jacques Delaunay, Ichiro Yamada
ISBN:	978-972-8939-49-6
Editors:	Mário Macedo
Year:	2011
Edition:	Single
Keywords:	Monitoring of eating habits, wearable sensing system, wavelet features, mRMR, PNN.
Type:	Full Paper
First Page:	18
Last Page:	27
Language:	English
Cover:
Full Contents:	click to dowload
Paper Abstract:	In recent years, an increasing number of people have been suffering from lifestyle-related diseases such as metabolic syndrome. Although it is important to monitor eating habits to prevent lifestyle-related diseases, no objective sensing or analysis systems have been established yet. Thus, we have been developing an analysis method to differentiate activities at meal times using wearable sensors. The sensing hardware consists of a bone-conduction microphone connected to a portable IC recorder that collects vibration signals from internal body sounds. In the differentiation process, we adopted a wavelet function to extract features that may be relevant characterize activities at meal times, to produce 70 features from the coefficients after discrete wavelet transformation. Then, we selected an optimal features set using the minimal-redundancy-maximal-relevance (mRMR) criterion, and we finally used a probabilistic neural network (PNN) to classify activities at meal times. Experiments were first carried out on small data set collected from six people. Then, we did further evaluation. The classification of activities at meal time using our proposed differentiation process, resulted in an accuracy of 87% with the larger database including data from 10 food types of 15 participants in the evaluation, and the selected features set was independent of individual differences. Additional analysis was carried out applying principal component analysis (PCA) on extracted features, and we succeeded in obtaining information on food hardness.
	Go Back