End-to-end learned hybrid refractive-diffractive optical design for fast and high-quality varifocal imaging

Modern varifocal imaging systems face a stringent trade-off between compactness and high-performance imaging. While an electrowetting liquid lens (ELL) offers a promising solution to achieve fast tunable imaging and miniaturization, it is limited by finite focus range and inherent aberrations. To address the complex interplay between focus range, image quality, and system compactness, we propose an end-to-end (E2E) joint optimization framework for a hybrid refractive-diffractive system, achieving synergistic optimization of the dielectric failure suppression ELL, the diffractive optical element (DOE), and the reconstruction algorithm. We introduce a decoupled physics-informed network (DPI-Net). Guided by degradation priors, this model employs a decoupled strategy to recover information explicitly from both amplitude and phase domains. This approach effectively corrects the inherent chromatic aberrations of the devices and the complex dynamic aberrations within a large focus range. Consequently, we realize a compact and lightweight system characterized by a large focus range, rapid response speed, and high imaging quality. This methodology holds significant promise for applications in smart vision and miniaturized optoelectronic imaging systems.