High-precision microcavity pressure sensing aided by MLP and GRNN

In response to the requirements for high-precision detection and diverse data scenarios in the field of intelligent optical sensing, this research combines whispering gallery mode (WGM) microcavity sensing with machine learning to solve the problems of low spectral information utilization, large random errors, and poor adaptability to data scales in the traditional microcavity sensing. Firstly, the WGM microcavity sensing system is used to collect transmission spectral datasets of different scales. Secondly, a multi-layer perceptron (MLP) deep learning algorithm based on full-spectrum feature mapping is adopted to train and test the datasets through hierarchical feature extraction and nonlinear fitting. The results show that the MLP achieves the test accuracy of 99.95% on large datasets. However, it exhibits poor performance on small datasets. Subsequently, the generalized regression neural network (GRNN) is introduced, leveraging its non-iterative training and strong local feature fitting advantages to optimize the small sample scenarios. The results indicate that the GRNN can achieve a test accuracy of 98.85% on small data sample datasets, improving by 10.29% compared to MLP. Finally, this study quantitatively compares and analyzes the test performance of MLP and GRNN models for five datasets of different scales, clarifying the performance advantages of the two models under different data conditions. This study fully utilizes the characteristics of MLP and GRNN models to achieve high-precision detection under different data scales, providing strong technical support for the application of intelligent optical microcavity sensing technology in various scenarios.