Algorithm 1 Proposed meta-learning recognition module

Input: Training and testing dataset S, inner step size $\alpha$, meta update step size $\beta$ Output: Few-shot gesture classification model While not done do     Task generation $T_i=(T_i^S,T_i^Q)$ from S     Compute prototype ${ϵ}_n$     Generate classifier initialization ${\omega }_n$     Set $\theta =\left\{{\omega }_n,{\theta }_{g\left(x\right)}\right\}$     for $T_i\in T$ do         Calculate loss ${\mathrm{Loss}}_{T_i^S}$         Update the parameter ${\theta }_{T_i}=\theta -\alpha {\nabla }_{\theta }{\mathrm{Loss}}_{T_i^S}\left(\theta \right)$     end     Update ${\theta }^{\prime }=\theta -\beta {\nabla }_{{\theta }^{\prime }}{\sum }_{i=1}^M{\mathrm{Loss}}_{T_i^Q}\left({\theta }_{T_i}\right)$   end