Complex-Valued Optical Convolution Accelerator for Feature Extraction
Author Information
Author(s): Bai Yunping, Xu Yifu, Chen Shifan, Zhu Xiaotian, Wang Shuai, Huang Sirui, Song Yuhang, Zheng Yixuan, Liu Zhihui, Tan Sim, Morandotti Roberto, Chu Sai T., Little Brent E., Moss David J.
Primary Institution: Beijing University of Posts and Telecommunications
Hypothesis
Can a complex-valued optical convolution accelerator improve the recognition of complex-valued radar images?
Conclusion
The complex-valued optical convolution accelerator achieved an accuracy of 83.8% in recognizing complex radar images, demonstrating its potential for real-time data analysis.
Supporting Evidence
- The accelerator operates at over 2 Tera operations per second (TOPS).
- Experimental tests with 500 images yielded an accuracy of 83.8%.
- The approach facilitates feature extraction of phase-sensitive information.
- It represents a pivotal advance in artificial intelligence for high-dimensional data analysis.
- The system can process 13.7 million 100×100 SAR images per second.
Takeaway
This study shows a new machine that can quickly understand complicated pictures from satellites, helping us learn more about our planet.
Methodology
The study involved using a complex-valued optical convolution accelerator to process synthetic aperture radar images, achieving high-speed computations.
Limitations
The recognition accuracy may depend on the experimental system's time/frequency response and other factors like weight control accuracy.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website