Deep learning-assisted quadratic metalens enables high-quality imaging within a 170° field of view

Quadratic phase metalens can transform the rotational symmetry of incident light into the translational symmetry of focal spots, showing significant potential for wide-field-of-view imaging. However, its inherent severe aberrations affect the actual image quality. To address this, this paper constructs an image restoration model based on a U-Net neural network, aiming to enhance the imaging performance of the quadratic phase metalens by incorporating post-processing methods. The results demonstrate that within the half-field-of-view range of 0° to 85°, the imaging quality of the restored images is significantly improved — the average structural similarity (SSIM) increases from 0.74 to above 0.78, and the average peak signal-to-noise ratio (PSNR) rises from 16.0 dB to above 21.5 dB. Particularly at a large incident angle of 85°, the SSIM of the restored image can still reach 0.7455, with a PSNR exceeding 20.0 dB, achieving high-contrast imaging over a 170° field of view. This method does not require additional optical components, thus significantly enhancing the potential applicability of metalenses in compact systems such as augmented reality/virtual reality (AR/VR) and biomedical imaging.