Construction methods of deep virtual datasets for single-plane diffractive elements

Abstract

The miniaturization and simplification of optical imaging systems have become critical demands in modern optics, yet conventional refractive optics remain bulky. The Single-Plane Diffractive Optical Element (SPDOE) enables image formation with a single optical component and features simpler microstructural fabrication compared to other advanced elements. However, owing to its distinct diffractive nature, the SPDOE inevitably introduces aberration-induced blur and diffraction-related background blurring in the imaging process. In this study, a virtual dataset construction method capable of accurately characterizing the degradation features of the SPDOE was proposed. Combined with a simplified neural network, high-quality real-time imaging was ultimately achieved. The aberration-induced image degradation was simulated by convolving the full-field point spread function (PSF) with the target image, while the diffraction-related background blur was modeled based on the proposed PSF degradation method derived from diffraction efficiency. An SPDOE with an f-number of 5 and a focal length of 50 mm was fabricated, operating within the visible wavelength range of 486–656 nm. Experimental results demonstrate that a structural similarity index (SSIM) of up to 0.9012 was achieved between the synthetic degraded images constructed using this method and the actual SPDOE captured images. A peak signal-to-noise ratio(PSNR) of 28.1 dB was obtained from tests conducted on real captured images, the accuracy of the constructed method was quantitatively validated. Compared with conventional deconvolution based on PSF models, the proposed deep learning approach achieved over fivefold improvement in real-time performance. This method addresses the limitations of small-sample SPDOE datasets, significantly reduces image acquisition and reconstruction time, and eases pixel alignment, providing a theoretical foundation for high-quality real-time imaging with SPDOE.