Algorithm 1 Training algorithm for SAE

Require:  Dateset $Data=\left\{x_n\right\}$;The number of hidden units $N_h$; Number of iterations: m Ensure:  Optimization results  1.Initialize the matrix and randomize the bias.  2.Construct $g_{\theta 1}\left(x_n\right)=g\left(U_h^T{x}_n+b\right)$ // In the encoding process, $U_h^T$ is the weight matrix between the input layer and the hidden layer, and b is the bias vector generated during encoding.  3.Perform forward propagation to reconstruct $\overline{x}$:$h_{\theta 2}\left(g_{\theta 1}\left(x_n\right)\right)=h\left(U_r^T{g}_{\theta 1}\left(x_n\right)+b^{\prime }\right)$ // Encoding process, where $U_r^T$ is the decoding matrix between the input layer and the output layer, and $b^{\prime }$ is the bias vector generated during decoding.  4.Calculate loss.  5.Perform backpropagation to update model parameters.  6.Repeat steps 2–5 m times to output the optimization results and use them as feature vectors of the input vectors to extract higher representations in SAE.