LSTM-based spectral confocal signal processing method

Traditional spectral confocal signal processing methods have problems such as difficulty in accurately extracting the peak wavelength, nonlinear error in the polynomial fitting calibration of the peak wavelength and position, and high dependence on hardware equipment. In this paper, the method of the LSTM neural network is introduced to achieve the direct characterization from the full spectrum signal to the position information by using its nonlinear mapping capability. The experimental results show that the measurement accuracy and measurement resolution of the new, to the best of our knowledge, method are improved, and it can still maintain a good measurement effect when using a low-performance spectrometer.