Research on Surface Plasmon Polariton Interferometer Sensing Method Based on BP Neural Network

本研究提出了一种融合微纳光学仿真与深度学习算法的折射率传感预测方法。针对传统狭缝–凹槽干涉结构，通过理论建模与数值仿真，生成不同折射率环境下的干涉条纹图像。利用主成分分析对图像数据进行特征提取与降维，构建低维表征空间以简化输入复杂度。在此基础上，设计基于贝叶斯正则化反向传播的BP神经网络模型，通过引入正则化约束优化网络参数，有效抑制过拟合现象，实现了干涉条纹至折射率的高精度映射。仿真结果表明，该模型预测误差低于10−4量级，展现出优异的稳定性与泛化潜力。然而，受限于理想化仿真数据的有限样本量，未来需通过实验数据扩充与噪声嵌入进一步提升模型对实际复杂场景的适应性。本研究为光学传感系统的智能化分析与设计提供了新的参考方法。This study proposes a refractive index sensing prediction method integrating micro-nano optical simulations with deep learning algorithms. For traditional slit-groove interferometric structures, interference fringe images under varying refractive index environments were generated through theoretical modeling and numerical simulations. Principal component analysis was employed for feature extraction and dimensionality reduction of the image data, constructing a low-dimensional feature space to simplify input complexity. On this basis, a BP neural network model based on Bayesian regularized backpropagation was designed, where regularization constraints were introduced to optimize network parameters, effectively mitigating overfitting and achieving high-precision mapping from interference fringes to refractive indices. Simulation results demonstrated that the model exhibits a prediction error below the order of 10−4, showcasing exceptional stability and generalization potential. However, limited by the idealized simulation data with finite sample size, future work should enhance the model’s adaptability to practical complex scenarios through experimental data augmentation and noise embedding. This research provides a novel methodological reference for intelligent analysis and design of optical sensing systems.