We propose a method based on nonsubsampled shearlet transform (NSST) and parameter-adaptive simplified pulse coupled neural network (SPCNN) for underwater polarization image fusion. Firstly, the degree of linear polarization and polarized light intensity images of underwater objects are acquired. Then NSST decomposition is performed on the two images to obtain their multi-scale and multi-direction subband coefficients. The high frequency direction subband coefficients of the two images are fused by the parameter adaptive SPCNN model. The low frequency subband coefficients of the two images are fused by an adaptive weighted fusion method based on regional energy. Finally, the fused image is reconstructed by inverting NSST to the high frequency direction subbands and low frequency subbands. Experimental results show that compared with other polarization image fusion methods, the proposed method can detect more details and significant features of underwater objects, and improve subjective visual perception and objective evaluation.