Application of deep learning in active alignment leads to high-efficiency and accurate camera lens assembly

Abstract

Active alignment plays a crucial role in minimizing decentering errors in optical systems and enhancing imaging quality. Traditional alignment techniques typically pay less attention to alignment speed and require precision apparatuses, such as laser locators, wavefront sensors and so on. To address these issues, this paper proposes a learning-search method that combines deep learning with search for achieving efficient alignment with a simple hardware system. With the powerful analytical capabilities of deep learning for images, our method achieves a speed of 9.2 seconds in experiments, which is a 56% improvement over the conventional search-based method. In terms of accuracy, it reaches an average weighted modulation transfer function (WMTF) of 0.594, with a difference from search-based method of no more than 0.003. Our method significantly improves alignment speed while maintaining accuracy, making it well-suited for large-scale applications.