解调
计算机科学
人工神经网络
模式(计算机接口)
控制理论(社会学)
电子工程
控制工程
工程类
实时计算
人工智能
跳跃的
作者
Shun Wan,Shuaimin Wu,Kaibo Gao,Jia Liu,Qianyu Ren,Gensen Yang,Pinggang Jia
出处
期刊:Measurement
[Elsevier BV]
日期:2026-08-10
卷期号:290: 122697-122697
标识
DOI:10.1016/j.measurement.2026.122697
摘要
Fiber-optic Fabry-Pérot sensors often suffer from the mode-jumping problem due to spectral degradation in practical environments. To address this, a physics-informed dual-head convolutional neural network is proposed. This method utilizes the free spectral range to generate a segmented weighting mask, avoiding misjudgments caused by local spectral overlapping, and decouples the absolute cavity length prediction into integer order classification and fractional phase regression. Under simulated degraded conditions, the model maintains a zero mode-jumping rate when the signal-to-noise ratio drops to 0 dB. In dynamic micro-displacement and practical static pressure tests across a wide cavity length range, the algorithm exhibits robust tracking without discrete jumps. Furthermore, system performance evaluations reveal a high measurement resolution of 0.764 nm and a relative repeatability error of ±0.0654 % F.S., indicating its reliability for high-precision applications.
科研通智能强力驱动
Strongly Powered by AbleSci AI