Recently, artificial neural network (ANN) attracts intensive attentions on solving electromagnetic (EM) inverse problems. In an inverse design of frequency selective surface (FSS) model with ANN, the inputs are S-parameters, while the outputs are structure parameters or material parameters. However, faced with applications where S-parameters vary in a large frequency range with different curve shapes, such as multi-band microwave devices, simple sampling with equal spacing may cause the input dimension to be too large and will require more complex neural network. In this paper, a cognition-driven sampling method is introduced to solve this problem. A parameter-extraction modeling of dual-passband FSS using both equidistant sampling and proposed method is presented and the well-designed FSS is further fabricated to validate the technique.