Anton V. Saetchnikov1、2、*, Elina A. Tcherniavskaia3, Vladimir A. Saetchnikov2, and Andreas Ostendorf1
Author Affiliations
1Applied Laser Technologies, Ruhr University Bochum, Bochum 44801, Germany2Radio Physics Department, Belarusian State University, Minsk 220064, Belarus3Physics Department, Belarusian State University, Minsk 220030, Belarusshow less
Fig. 1. Overview of the instrument configuration: 1-laser diode; 2-collimation lens; 3-camera; 4-beam dump; 5-right angle optical prism; 6-adhesive thin layer; 7-microresonator.
Fig. 2. Overview of the WGM signal.
Fig. 3. Comparison of the experimental data collected in the laser frequency sweeping and the fixed frequency schemes for the same sensor sample under changing ambient refractive index.
Fig. 4. Sensor response on the refractive index changes.
Fig. 5. (a) Distribution of the absolute error values between the measured refractive indexes and the values predicted with different processing approaches: deep neural network (dNN), linear regression (LR), random forest (RF), general regression neural network (GRNN), gradient boosting (GB), and support vector regression (SVR). (b) Statistics on the performance of the refractive index prediction with dNN approach with different combinations of weights optimization methods (Adam, RMSprop, Nadam, Adagrad, and Adadelta) and activation functions (tanh, sigmoid, relu, selu, linear, and softplus).
Fig. 6. Statistics on the refractive index prediction accuracy represented as absolute error values for different dNN configurations with varying number of neurons (N) in the input layer (n = 16, 32, 48, 56, 64, 80, 96, 112, 128, 256, 512, 1024) and hidden layers (L) number (3, 4, 5, 6) with 2n neurons.