To address the issue of unstable imaging performance of spatially modulated fully polarized imaging systems in different target scenes, this paper proposes a method for optimizing filter bandwidth calculation based on simulated interference watermark fringes. This method aims to ensure that the system maintains optimal performance across various target scenarios and provides theoretical guidance for selecting system imaging lenses. The approach involves generating simulated interference images by superimposing two-dimensional sine functions of varying sparsity, which represent simulated interference watermark fringes, onto intensity images. The fast Fourier transform and the frequency domain low-pass filtering algorithm are used to demodulate the simulated interference diagram, producing the total polarization image. The polarization image with the highest structural similarity to the intensity image is selected for further analysis. Additionally, an optimal filter bandwidth calculation method based on improved particle swarm optimization algorithm is proposed. The adaptive selection of the optimal filter bandwidth is achieved by introducing a vaccine extraction selection strategy and a simulated annealing mechanism. Combined with the selected image resolution, incident light wavelength and Savart polarizer single beam light offset in different target scenes, the optimal focal length of the imaging lens was calculated, and the theoretical selection was completed. In the experimental section, the stability of the spatial modulation-based full-polarization imaging system, built using the optimal filter bandwidth selection, is compared with that of the traditional experience-based selection. The experimental results show that the spatial modulation total polarization imaging system based on the optimal filter bandwidth has better performance. The area of spectral graph inversion is increased by 4.16 times and the similarity of the image structure based on the optimal filter bandwidth selection is improved by 63% compared with the self-empirical selection, which significantly improves the polarization imaging quality of the system.