Design Of The Program Controlling And Signal Processing Method Based On Optical Pump-probe Detect

The optical pump-probe detection technology is a high-resolution,high-penetration ultrafast spectroscopy technology based on the photoacoustic effect.By adjusting the relative optical path difference between the pump light and the probe light,it can well reveal that the photoacoustic effect is on the surface of the material or the transient process of internal excitation.This thesis focuses on the joint modulation and data communication problems of the measuring equipment in the optical pump detection system,the noise reduction scheme that is interfered in the propagation and detection process,and the classification and segmentation method of the complex membrane structure.The focus is on the lock-in amplifier and optical displacement.Taiwan’s joint communication and highspeed data acquisition algorithms have studied empirical mode decomposition theory,wavelet threshold reconstruction and other methods;as well as deep learning theory and UNet model framework,it has realized single-layer film structure echo recognition and segmentation calculation,and Based on the COMSOL simulation of the photoacoustic signal,the model denoising level and the accuracy of the neural network are evaluated.main tasks as follows:(1)Study the principle of optical pump-probe detection technology,and determine the measurement process and scheme of pump-probe detection.Use Lab VIEW to jointly control key optical measurement equipment such as lock-in amplifiers,photodetectors,and optical delay lines,including interface definition and settings,system parameter selection,phase-lock node tree call and high-speed communication algorithm writing,DLL call and movement of the translation stage Parameter control,and UI interface design,etc.It realizes the automatic collection and storage of data during the experiment,dynamic system parameter setting and data storage functions,which lays the foundation for subsequent data analysis and signal processing.(2)Study the principle of empirical mode decomposition and related derivative algorithms,use MATLAB to compile EMD algorithm to decompose the photoacoustic signal,filter the decomposed eigenmode function by calculating mutual information,and use the wavelet threshold function idea to filter the result The noise component is reduced by threshold value.Finally,the noise reduction effect is verified by comparing band-pass filtering,wavelet analysis and other methods to improve the adaptability of the subsequent neural network to the input signal.(3)Research on deep learning theory,semantic segmentation model,measured photoacoustic signals for single-layer film and multilayer film samples,based on Tensor Flow framework,using Python environment to write a one-dimensional U-Net model,including the modification of convolutional layer dimensions,Adjust the size of the convolution kernel,label the photoacoustic signal and generate MASK;the model training part includes training set,verification set division,data standardization,production of iterative generator,etc.,combined with the actual photoacoustic signal appearance,focus on adjusting the volume Accumulation core size,step length and other parameters,and finally evaluate the pros and cons of the model through loss and accuracy,which provide a wide range of ideas for studying the echo recognition and film thickness calculation of multilayer films or complex film structures.