Light interferometric ranging technology,with its nanometer-level resolution and non-contact measurement advantages,has become a core tool in precision manufacturing and fundamental scientific research.However,its per...Light interferometric ranging technology,with its nanometer-level resolution and non-contact measurement advantages,has become a core tool in precision manufacturing and fundamental scientific research.However,its performance is still constrained by multiple factors,such as environmental interference,light signal attenuation,and detection noise.The advent of artificial intelligence(AI)technology has enabled competitive solutions.By recognizing data features through neural networks,it can promote further upgrades in measurement accuracy,demodulation speed,and environmental robustness.Here we report a deep-learning-based interferometric(DLI)ranging method utilizing a dispersion-controlled dual-swept laser(DCDSL),which forms asymmetric interference signals.We demonstrate that deep neural networks can robustly identify signal features even amid strong interference via training and incorporating underlying physical constraints.This circumvents the reliance of conventional phase demodulation techniques on high signal-to-noise ratios and enables absolute distance measurements with repeatability starting from 449 nm in a single shot to 8.41 nm after averaging 1.5 ms,with a corresponding network inference time of 1.579 ms.Furthermore,our proposed method supports dynamic distance measurement with a tunable temporal resolution.Such a combination of AI and interferometry ranging technology may pave the way for the design of future LiDAR systems with robust interference immunity.展开更多
基金National Natural Science Foundation of China(NSFC)(62405036,62441508)Graduate Research and Innovation Foundation of Chongqing,China(CYB240013)+1 种基金National Key Research and Development Program of China(2024YFF0617000)Research Grants Council,University Grants Committee of Hong Kong SAR(PolyU15206023)。
文摘Light interferometric ranging technology,with its nanometer-level resolution and non-contact measurement advantages,has become a core tool in precision manufacturing and fundamental scientific research.However,its performance is still constrained by multiple factors,such as environmental interference,light signal attenuation,and detection noise.The advent of artificial intelligence(AI)technology has enabled competitive solutions.By recognizing data features through neural networks,it can promote further upgrades in measurement accuracy,demodulation speed,and environmental robustness.Here we report a deep-learning-based interferometric(DLI)ranging method utilizing a dispersion-controlled dual-swept laser(DCDSL),which forms asymmetric interference signals.We demonstrate that deep neural networks can robustly identify signal features even amid strong interference via training and incorporating underlying physical constraints.This circumvents the reliance of conventional phase demodulation techniques on high signal-to-noise ratios and enables absolute distance measurements with repeatability starting from 449 nm in a single shot to 8.41 nm after averaging 1.5 ms,with a corresponding network inference time of 1.579 ms.Furthermore,our proposed method supports dynamic distance measurement with a tunable temporal resolution.Such a combination of AI and interferometry ranging technology may pave the way for the design of future LiDAR systems with robust interference immunity.
