This chapter explores a two-stage intelligent channel estimation scheme for OFDM systems, encompassing pilot-position channel estimation and full temporal-frequency-domain channel interpolation. In the first stage, a CNN-based pilot-position channel estimation network (PCENet) is designed. It extracts and learns features from the transmitted and received signal matrices at pilot positions, yielding the CSI matrix for these positions. The second stage leverages the pilot-position estimation results to design a fully connected network (FCN)-based full temporal-frequency-domain interpolation network (INet). This network learns the correlation features between frequency-domain subcarriers and time-domain symbols, ultimately outputting the full temporal-frequency-domain CSI matrix.