
Next-Generation On-Device AI Vision Technology Cuts Power Consumption at the Data Generation Stage; Undergraduate-Led Paper Selected as Cover Article in International Journal* Nanophotonics
A joint research team led by Professor Badloe of the Department of Electronics and Information Engineering and Professor Lee Hyungun of the Department of AI Semiconductor Engineering at Korea University Sejong Campus as developed a next-generation imaging system capable of ultra-low-power, high-speed, low-data edge detection by integrating a metalens with a compressive sensing (CS)–based semiconductor image sensor.
The research is significant in that it addresses the power consumption and data bottleneck problems associated with large-volume image data processing in AI vision systems by reducing computational load at the point of data generation itself, rather than after the fact. In particular, the team proposed a novel approach in which only the necessary information is efficiently extracted as light reaches the sensor, departing from the conventional method of transmitting a full captured image and then processing it through digital computation.
The findings were published in Nanophotonics (Impact Factor: 6.6; top 15% in JCR), an international journal in the fields of optics and nanomaterials and were selected as the journal's front cover article. The paper has drawn additional attention for having been led by Yoon Se-jin, who was a fourth-year undergraduate student in the Department of Electronics and Information Engineering at the time of the research.
"Hybrid Metalens–Compressive Sensing Imaging System for Ultra-Low-Power Edge Detection," Nanophotonics 2026, nap2.70079)

Edge detection is a technique for extracting the boundaries of objects within an image and serves as a critical pre-processing step in various machine vision systems, including object recognition, autonomous driving, and robotic vision. Conventional digital approaches, however, require all data captured by an image sensor to be transmitted to a signal-processing device before algorithmic computation can be performed, resulting in high power consumption and processing delays caused by data bottlenecks between the sensor and processor. Applications requiring real-time processing of 4K and 8K high-resolution images have long faced the challenge of rising processing times and power consumption as resolution increases.
To address these issues, the research team proposed an opto-electronic co-design system at the system level, combining a spiral metalens with a compressive sensing CMOS image sensor (CS-CIS).
The technology precisely controls the phase of light through a metamaterial-based lens, enabling the extraction of edge information at the speed of light without any additional digital computation. A compressive sensing image sensor then acquires only the necessary information in a compressed form, substantially reducing data volume and power consumption while enabling high-speed imaging. As a result, the team successfully captured edge images at a rate of 6,248 frames per second using only 12.5 percent of the original data, a data acquisition performance hundreds of times faster than that of conventional cameras, demonstrating the feasibility of an ultra-low-power, high-speed edge detection imaging system.

The research team expects the technology to find application across a range of fields, including real-time object recognition in autonomous vehicles, low-power vision sensors for mobile and wearable devices, robotic vision, and industrial defect inspection. The team also anticipates that, when combined with reconfigurable metasurface technology capable of freely modulating the phase of light, the system could be extended into an adaptive camera whose functions can be reconfigured according to context.
The study involved Yoon Sejin (first author) and Kim Haseong (co-author), students in the Department of AI Semiconductor Engineering at Korea University Sejong Campus, with Prof. Hyunkeun Lee of the Department of AI Semiconductor Engineering and Prof. Trevon Badloe of the Department of Electronics and Information Engineering serving as corresponding authors.
Yoon Sejin stated, "By integrating and optimizing the optical computing capability of the metalens with the data efficiency of the compressive sensing image sensor into a single system, we were able to simultaneously achieve ultra-low power consumption alongside high speed and high performance, two qualities that were previously difficult to reconcile."
Supervising professors Badloe and Lee remarked, "It is a deeply meaningful achievement that research begun at the undergraduate level has culminated in a cover article in a world-class journal. This hybrid imaging platform presents a new form of vision system that combines optics and semiconductor technology, and will serve as an important technological foundation toward the commercialization of next-generation on-device artificial intelligence vision systems."
The research was supported by Korea University's internal research fund (K2515331); the Human Resources Development Program for Industrial Innovation (RS-2025-02215617) funded by the Korea Institute for Advancement of Technology (KIAT) under the Ministry of Trade, Industry and Energy (MOTIE); the Alchemist Project (1415185027, 20019169) funded by the Korea Evaluation Institute of Industrial Technology (KEIT) and MOTIE; the Regional Innovation System & Education (RISE) program through the Sejong RISE Center, funded by the Ministry of Education and Sejong Metropolitan Autonomous City (2025-RISE-08-001); and the Information Technology Research Center (ITRC) program (RS-2026-25520273) of the Institute of Information & Communications Technology Planning & Evaluation (IITP) under the Ministry of Science and ICT.