Adaptive design framework for heterogeneous off-axis reflective imaging systems

Freeform off-axis reflective imaging systems represent one of the core instruments in the field of optical observation. However, the inherent asymmetry of off-axis reflective systems leads to a design process that is heavily dependent on experience and lacks a design framework across different configurations (e.g., two-mirror and three-mirror systems). This results in a fragmented workflow, difficult trade-offs between configurations, and a low overall efficiency. To address this, we propose an adaptive design framework for heterogeneous off-axis reflective imaging systems based on machine learning. First, an automated method for constructing a dataset of multi-mirror off-axis reflective imaging systems with good imaging quality is designed. Then, an adaptive encoding method is proposed. This method establishes a unified description for the optical parameters of different configurations, enabling a coherent design workflow. Finally, a random forest generation model for off-axis reflective imaging systems is constructed and trained, and the trained model is employed for prediction. Based on the predicted combinations of physical parameters, the improved Wassermann-Wolf (W-W) method is applied to derive freeform off-axis reflective systems. Experimental results show that this single framework can successfully generate both off-axis two-mirror and three-mirror systems from different input requirements, with the imaging quality meeting the design specifications. This provides an effective pathway for adaptively generating off-axis reflective imaging systems with varying numbers of mirrors, offering a more flexible and systematic methodology for advanced optical design.