Researchers at Pohang University of Science and Technology and the University of Washington developed a de-occluding broadband metalens designed to remove physical obstructions such as raindrops, fences, and dust, according to a paper published in Nature Communications on August 15, 2026.
Conventional approaches to clearing occlusions rely on bulky compound-lens arrays or computational inpainting methods, which compromise compactness or image fidelity. While metalenses offer compact form factors, achieving broadband, obstruction-free imaging has remained difficult because a single metalens cannot simultaneously focus distant scenes and defocus nearby occlusions across a broadband spectrum.
The team addressed this limitation by combining a learned phase profile with multi-band spectral filtering. The optical setup divides the spectrum of each color channel into distinct pass and stop bands. Light from distant objects travels through the pass bands to be focused, whereas light originating from nearby obstructions falls into the stop bands and is rejected. An integrated neural network then processes the captured light to further refine overall image quality.
Testing against a conventional hyperbolic metalens under obstructed viewing conditions demonstrated quantitative performance gains. The new design improved peak signal-to-noise ratio by 32.29%, raised object detection benchmark results by 13.54%, and increased scores across two semantic segmentation benchmarks by 48.45% and 20.35%, respectively.
The study, co-led by first authors Seungwoo Yoon and Dohyun Kang under corresponding authors Seung-Hwan Baek and Junsuk Rho, targeted compact platforms including mobile robots, drones, and medical endoscopes. Project funding was provided by the National Research Foundation of Korea, the Korean Ministry of Science and ICT, the Institute of Information & Communications Technology Planning & Evaluation, Samsung Electronics, POSCO, and the U.S. National Science Foundation.
