Deep Mining of Wearable Spatial Variability for Efficient Edge Computing.

Wearable biomechanical sensors can be placed on different body locations and each sensor may have multiple channels. Deep mining of the spatial variability in these sensor locations and channels, is essential for not only optimal system configuration towards energy efficient edge computing, but also comprehensive musculoskeletal dynamics understanding. Targeting these needs, we propose to leverage deep learning to investigate seven body locations and six channels for each location, thereby demonstrating the spatial variability among 42 combinations. The research findings indicate that the thigh location and the accelerometer axis-Y is the best configuration. Further, experimental results also indicate the diverse spatial variability among different sensor locations and sensor channels, which provide interesting and rich information about the biomechanical dynamics. This study will thus greatly advance our understanding of the spatial variability in wearable biomechanical sensors and channels, thereby minimizing the data analytics load and facilitating energy efficient edge computing for big biomechanical data mining.