Researchers at King Abdullah University of Science and Technology (KAUST) have developed a new artificial intelligence method called DeepLens that dramatically speeds up and simplifies the design of optical lenses for imaging systems. This innovative approach automates the complex process, significantly reducing the time and cost associated with lens development, particularly for mobile phone cameras, while also promising enhanced image quality.
DeepLens utilizes a technique known as “curriculum learning,” inspired by human learning processes, to iteratively refine lens designs. Instead of requiring human-designed templates, as with traditional methods, the AI autonomously generates designs for compound optical systems. These systems comprise multiple refractive lens elements, each specifically shaped to achieve optimal performance.
According to Xinge Yang, one of the developers, traditional automated methods offered only minor improvements to existing optical designs. DeepLens, however, offers a substantial leap forward, potentially shrinking months of work by experienced engineers down to a single day.
The effectiveness of DeepLens has been demonstrated in creating both conventional optical designs and advanced computational lenses with extended depth-of-field. For example, it has been used to design a mobile phone lens system with a wide field of view, incorporating highly aspheric lens elements and a short back focal length.
Currently, DeepLens is designed for refractive lenses, but the KAUST team is working to expand its capabilities to include hybrid optical systems. This would integrate refractive lenses with diffractive optics and metalenses, allowing for further miniaturization and new imaging functionalities such as spectral cameras and joint-color depth imaging. The researchers believe this technology will have a far-reaching impact, setting new performance standards for imaging systems across various applications, from mobile phones to advanced imaging technologies. The research findings have been published in Nature Communications.
Leave a Reply