Continuous Human Motion Recognition Based on FMCW Radar and Transformer

Figure 14

Training loss for policy (1) and policy (2). Strategy (1) only needs to predict the type of the action, so it only needs to be trained for 150 epochs. Strategy (2) not only needs to predict type of the action but also needs to predict the time of the action, so it needs to train 800 epochs.

(a) Loss function trained by strategy (1)

(b) Loss function trained by strategy (2)