摘要
【目的】为应对传统边缘检测存在的运算速度瓶颈与高能耗问题,文章提出了一种基于四波混频(FWM)效应的边缘检测光计算方法。【方法】该方法利用FWM效应完成边缘检测中梯度计算的乘法运算,随后利用数字信号处理技术完成边缘检测。为验证其处理速度与检测准确性,文章利用FWM效应完成边缘检测并构建了可实现10 Gbit/s高速图像处理的仿真平台,深入研究了载波频偏与卷积核类型对检测性能的影响。【结果】结果表明,在系统性能方面,该方法的能耗和时延分别为0.25 pJ/bit和0.98μs,优于传统计算方案。此外,文章深入研究了载波频偏对边缘检测结果的影响。当承载边缘检测结果的信号与承载卷积核的载波信号中心频率间隔10 GHz时,二者频谱显著重叠,相关性系数仅为0.15164。当承载卷积核的载波频偏为图像数据载波频偏的两倍时,采用水平或垂直方向卷积核进行边缘检测会受到显著干扰,相关性系数最低降至0.2499;而在相同条件下,采用全方向卷积核则能够显著提升检测性能,相关性系数最高可达0.94722。【结论】研究表明,该方法能进行高速乘法运算,而且通过合理调整载波频偏和优化卷积核选择,可进一步提升检测精度。
【Objective】To address the bottlenecks of computational speed and high energy consumption in traditional edge detection,this study proposes an optical computing method for edge detection based on the Four Wave Mixing(FWM)effect.【Methods】This method uses the FWM effect to complete the multiplication operation of gradient calculation in edge detection,and then uses digital signal processing technology to complete the edge detection.To validate the method's processing speed and detection accuracy,a simulation platform capable of 10 Gbit/s high-speed image processing was established.The influence of carrier frequency offsets and convolution kernel types on detection performance was systematically investigated.【Results】The results demonstrate superior system performance,with an energy consumption of 0.25 pJ/bit and a latency of 0.98μs,outperforming conventional computing schemes.Regarding interference,a 10 GHz frequency separation between the resulting signal and the kernel-carrying carrier led to significant spectral overlap,reducing the correlation coefficient to just 0.15164.Furthermore,under specific interference conditions where the kernel carrier's frequency offset was twice that of the data carrier,using single-direction(horizontal or vertical)kernels caused severe disruption(correlation coefficient dropping to 0.2499).In contrast,employing an omnidirectional kernel under the same conditions significantly enhanced the detection performance,achieving a correlation coefficient as high as 0.94722.【Conclusion】This study demonstrates that the proposed method not only enables high-speed multiplication but also achieves higher detection accuracy through the rational configuration of carrier frequency offsets and optimization of convolution kernel selection.
作者
张洪波
赵思豪
蔡炬
张敏
万峰
王森
尚德浜
ZHANG Hongbo;ZHAO Sihao;CAI Ju;ZHANG Min;WAN Feng;WANG Sen;SHANG Debang(College of Communication Engineering,Chengdu University of Information Technology,Chengdu 610225,China;TianfuJiangxi Laboratory,Chengdu 641419,China)
出处
《光通信研究》
北大核心
2025年第5期27-33,共7页
Study on Optical Communications
基金
四川省科技计划资助项目(2021YFG0149)
重庆市自然科学基金创新发展联合基金资助项目(CSTB2023NSCQ-LMX0033)
四川省高等教育学会资助项目(GJXH2024YB-025)
成都信息工程大学引进人才科研启动资助项目(KYTZ202007)。
关键词
边缘检测
四波混频
光计算
高速乘法运算
edge detection
FWM
optical computing
high-speed multiplication
