Algorithm 3 Construct training samples

Load the exiting dataset. for number of expansion times do Build a square lattice same with sampling grid. for $\left\{{\mathsf{\rho }}^{\left(\mathrm{i}\right)},{\mathsf{\theta }}^{\left(\mathrm{i}\right)}\right\}$ and $\left\{{\mathrm{V}}^{\left(\mathrm{i}\right)},{\mathsf{\sigma }}^{\left(\mathrm{i}\right)}\right\}$ in i-th sample do Convert polar coordinate $\left\{{\mathsf{\rho }}^{\left(\mathrm{i}\right)},{\mathsf{\theta }}^{\left(\mathrm{i}\right)}+num•{45}^{°}\right\}$ to rectangular coordinate $\left\{{\mathrm{x}}^{\left(\mathrm{i}\right)},{\mathrm{y}}^{\left(\mathrm{i}\right)}\right\}$. Set points $\left(\mathrm{x},\mathrm{y}\right)$ meets ${(\mathrm{x}-{\mathrm{x}}^{\left(\mathrm{i}\right)})}^2+{(\mathrm{y}-{\mathrm{y}}^{\left(\mathrm{i}\right)})}^2<{\mathrm{radius}}^2$ in square lattice as anomalies. Read the conductivity signal ${\mathsf{\sigma }}^{\left(\mathrm{j}\right)}$ in sequence again. Move the data in ${\mathrm{V}}^{\left(\mathrm{i}\right)}$ forward 7 units in sequence to form ${\mathrm{V}}^{\left(\mathrm{j}\right)}$. end for Save these $\left\{{\mathrm{V}}^{\left(1\right)},{\mathsf{\sigma }}^{\left(1\right)},\dots ,{\mathrm{V}}^{\left(\mathrm{j}\right)},{\mathsf{\sigma }}^{\left(\mathrm{j}\right)},\dots ,{\mathrm{V}}^{\left(num\right)},{\mathsf{\sigma }}^{\left(num\right)}\right\}$ as corresponding new samples. end for