Algorithm 2 The FA method for feature alignment

Input: The source domain feature matrix ${\mathbf{X}}_{\mathrm{S}}=\left[{\mathbf{x}}_{\mathrm{S}}^1,\dots ,{\mathbf{x}}_{\mathrm{S}}^{N_{\mathrm{S}}}\right]\in {\mathfrak{R}}^{d\times N_{\mathrm{S}}}$ The source domain label matrix ${\mathbf{T}}_{\mathrm{S}}=\left[T_{\mathrm{S}}^1,\dots ,T_{\mathrm{S}}^{N_{\mathrm{S}}}\right]\in {\mathfrak{R}}^{d\times N_{\mathrm{S}}}$ The target domain feature matrix ${\mathbf{X}}_{\mathrm{T}}=\left[{\mathbf{x}}_{\mathrm{T}}^1,\dots ,{\mathbf{x}}_{\mathrm{T}}^{N_{\mathrm{T}}}\right]\in {\mathfrak{R}}^{d\times N_{\mathrm{T}}}$ The transfer samples of source domain ${\mathbf{M}}_{\mathrm{S}}=\left[{\mathbf{m}}_{\mathrm{S}}^1,\dots ,{\mathbf{m}}_{\mathrm{S}}^{N_{\mathrm{m}}}\right]\in {\mathbf{X}}_{\mathrm{S}}$ The transfer samples of target domain ${\mathbf{M}}_{\mathrm{T}}=\left[{\mathbf{m}}_{\mathrm{T}}^1,\dots ,{\mathbf{m}}_{\mathrm{T}}^{N_{\mathrm{m}}}\right]\in {\mathbf{X}}_{\mathrm{T}}$ The regularization parameter $\lambda$ Output: Transformation matrix $\mathbf{P}$ New target domain $X_T^{\prime }$ Recognition accuracy (%) of unlabeled data in new target domain Procedure: 1: Calculate the basis transformation matrix $\mathbf{P}$ according to (8) 2: Calculate the new target domain $X_T^{\prime }$ according to $X_T^{\prime }$=${\mathbf{X}}_{\mathrm{T}}P$ 3: Calculate the predicted output of unlabeled samples of new target domain $\hspace{1em}{T}_{\mathrm{T}}$← SVMClassifier ${(\mathbf{X}}_{\mathrm{S}},{\mathit{T}}_{\mathit{S}},X_T^{\prime }$) 4: Calculate the recognition accuracy (%) of unlabeled data in new target domain