The channel sandstone of Qianfoya Formation in northeastern Sichuan Basin has good exploration prospects. However, sandstone reservoirs have strong horizontal and vertical heterogeneity and complex spatial distribution characteristics. At the same time, the longitudinal wave impedance of logging curves is not sensitive to sandstone and mudstone.Conventional post-stack seismic inversion has difficulty solving the problem of quantitative prediction of lithology and reservoirs in this area. Therefore, in this paper, seismic information from different frequency bands and geological sedimentary knowledge are fully utilized to explore a geologically constrained frequency division inversion method for predicting complex sandstone and reservoirs. Firstly, we use known logging data to clarify the parameter characteristics of lithology and reservoirs. Then, based on the frequency division of seismic data using the matching pursuit algorithm, the neural network algorithm is used to establish a nonlinear mapping relationship between lithology and frequency division volumes to directly predict lithology. At the same time, characteristic attributes related to the development of channel sands are introduced as constraints during the learning process to reduce the risk of over-fitting and make the results more predictive. Finally, based on sandstone prediction, conventional longitudinal wave impedance inversion is combined to further predict reservoirs with high porosity inside sandstone. The results of practical application exhibit high resolution in the longitudinal and transverse directions of the profile, high consistency with wells, and reasonable sedimentary patterns in the plane. This method explores a new technological approach for predicting similar complex sandstone reservoirs in other blocks.