FoG Recognition Based on Time-frequency Analysis and Convolutional Neural Network Combined with Attention Mechanism

Freezing of gait is one of the most common movement disorders in Parkinson’s disease. In this paper, with the use of attention mechanism, a neural network detection method is proposed for time-frequency data of FoG events. The acceleration signal collected by the IMU is firstly converted into a two-dimensional image representing the time-frequency information through continuous wavelet transform, and then sent to the convolutional neural network combined with the attention module for feature extracting and model training. Finally, the effectiveness is demonstrated with an accuracy of 91.37% and geometric mean of 88.73% on the test set.