Adaptive iterative optimization method for spectral calibration based on deep learning

Abstract

The miniature fiber optic spectrometer is smaller, cheaper and has a wide range of applications. However, the measurement error is larger. In order to solve this problem, the adaptive iterative optimization method for spectral calibration is proposed. In this study, a trinity neural network model is built based on spectral wavelength segmentation to improve the calibration degree. Based on the ‘pseudo-label’, a self-optimization method for spectral calibration is proposed to reduce the amount of data required. This study optimizes the measurement accuracy without changing the structure of the spectrometer. And the self-optimization of calibration model in practical application is realized. After experiment, the calibration degree of the calibration model can reach 75.72%. After a self-optimization, it can be increased to 87.45%. The calibration time of 401 spectral values (380 nm–780 nm) is less than 0.01 s. The results show that the operator can use this method to calibrate spectral data without having optical knowledge. This method has low cost, high calibration speed, good reliability and application value.