Abstract: To address the challenges posed by complex space target structures and pronounced scattering diversity, a pseudo full-polarization information reconstruction method is introduced for reduced-polarization ISAR based on three-dimensional polarimetric correlation direction diagrams. By utilizing correlation features between different polarimetric channels as mechanistic constraints and employing a cascaded residual dense feature extraction model, the method enhances the transmission and fusion of spatial and polarimetric features within multi-scale deep networks. Simulation results demonstrate that, compared to conventional deep learning-based pseudo full-polarization reconstruction approaches, the proposed method increases reconstruction accuracy by at least 3.269 dB and achieves more precise inference of texture structures, thus providing more comprehensive polarimetric scattering information for the fine recognition of space targets.