Images have become a primary medium for information dissemination and presentation, with high-resolution images offering superior clarity and the ability to convey richer details. With advancements in electro-optical imaging technology, modern optoelectronic systems are expected to achieve miniaturization and lightweight designs while maintaining high-resolution imaging capabilities. In this paper, we propose a simplified design framework for a super-resolution imaging system based on optical-digital joint optimization. The approach integrates optical design, image restoration, and super-resolution reconstruction algorithms. Using the gradient backpropagation mechanism of deep learning, both optical parameters and network parameters are co-optimized to achieve an optimal match between the imaging system and the reconstruction algorithm. To enable the neural network to address both optical aberration correction and super-resolution reconstruction simultaneously, an improved two-branch generative adversarial network is proposed. This network effectively extracts features related to optical blurring and super-resolution reconstruction in a targeted manner. Using this method, we demonstrate an end-to-end joint optimization of a card-type telescope imaging system and a recovery reconstruction network, significantly simplifying the telescope system’s structure while maintaining imaging quality. High-detail super-resolution images are obtained, and the algorithm’s performance and effectiveness are validated through comparisons of PSNR and SSIM metrics between reconstructed and reference images, along with other related algorithms.

In this paper, we propose an end-to-end simplified design framework for super-resolution imaging systems. By employing optical-digital joint optimization and leveraging the gradient backpropagation mechanism of deep learning, both lens parameters and recovery network parameters are co-optimized. The optical system’s imaging degradation is modeled and analyzed, and images synthesized from this degradation model are used to train the super-resolution reconstruction network, ensuring that the training results align closely with real-world scenarios. In addition, to address the dual challenges of optical aberration recovery and super-resolution reconstruction, an improved two-branch generative adversarial network is proposed. This network is designed to target and extract blurred optical features while simultaneously focusing on super-resolution reconstruction. Our method enables the design of simplified optical systems capable of high-resolution imaging, with significant potential application in fields such as security monitoring and aerospace.

Using the proposed method, the optical system’s complexity is successfully reduced from four mirrors to two without compromising imaging quality (Fig. 9). Objective metrics confirm that the imaging quality is comparable to that of the original optical system (Table 1). The reconstructed super-resolution images exhibit detailed texture information, and comparisons with advanced super-resolution networks demonstrate superior performance in terms of objective metrics (Table 2). To further validate the superiority of the joint optimization method proposed in this paper, a detailed simulation analysis of the entire imaging and reconstruction process is conducted. The results show that when the photoelectric receiver is constrained by its inherent physical limitations (e.g., low resolution) or when information is lost during image transmission, the resulting image suffers from degraded resolution. In such cases, conventional image restoration techniques cannot recover high-resolution images with rich details (Fig. 15). This underscores the necessity of incorporating super-resolution reconstruction.

In this paper, we propose a simplified design framework for super-resolution imaging systems using optical-digital joint optimization to meet the demand for lightweight optoelectronic imaging systems. Deep learning is utilized to co-optimize the optical system and recovery reconstruction network, simplifying the system structure while enhancing imaging resolution. For the dual challenges of correcting aberrations and reconstructing super-resolution images, an improved two-branch parallel generative adversarial network is proposed, specifically targeting blur correction and feature reconstruction. This framework is applied to a card-type telescope system, successfully reducing the lens count in the back group from four to two while improving the original imaging quality. Compared to previous joint optimization methods, we provide a more comprehensive consideration of image acquisition and detector physical constraints, modeling and analyzing all aspects of the image formation process. As a result, the proposed approach delivers more detailed, high-resolution images.