High Precision Laser Arbitrary Beam Splitting Technology Based On Machine Learning And Spatial Light Modulator

As one of the greatest inventions in the 20 th century,laser has been utilized in many fields,such as fiber-optic communication,fiber-optic sensors,all-optical information processing,optical storage,laser processing,laser surgery,lithography technology,laser radar and laser weapons,due to the technical advantages that traditional light sources cannot achieve including monochromaticity,coherence and directivity.High precision laser beam splitting technology is a common key technology in advanced manufacturing and medical fields such as laser processing,lithography,optical tweezers and laser surgery,it is also the development frontier of laser application technology.The existing high precision laser beam splitting technology is mainly used to achieve dual beams with uniform light intensity distribution.The laser beam splitting technology with arbitrary spatial distribution and arbitrary light intensity distribution is not mature enough to meet the requirements of simultaneous precision operation of multi-beams in laser precision processing,lithography,optical tweezers and laser medicine.At present,the bottleneck problem of arbitrary beam laser beam splitting technology is that the accuracy of laser beam splitting is affected by many factors,which makes it difficult to meet the requirements of laser beam splitting accuracy in practical applications.In this work,the machine learning is introduced into the laser beam splitting technology based on spatial light modulator(SLM),and the high precision arbitrary multi-beams laser beam splitting is realized based on the regression task of supervised learning.The SLM loads the phase hologram calculated by the iterative Fourier transform algorithm(IFTA)to realize the modulation of the incident light beam,so as to achieve output multi-beams with arbitrary spatial distribution and light intensity distribution.The numerical simulation and experimental research on the laser beam splitting technology based on SLM are carried out,The image detected by camera after laser beam splitting is employed as the sample,and the target image of beam splitting is utilized as the label.The data set is constructed by simulated and experimental data.Based on the convolution neural network(CNN)algorithm,the mapping relationship between the image of beam splitting and result is established by the regression task of supervised learning,so as to realize the accurate prediction of the IFTA amplitude distribution and effectively improve the accuracy of arbitrary multi-beams beam splitting.The main work is as follows:(1)Based on SLM and IFTA,the numerical simulation and experimental study of multi-beam laser beam splitting are carried out.A numerical simulation model of laser beam splitting based on mathematical transformation process is built to realize the numerical simulation of laser beam splitting.The experimental research on laser beam splitting technology is carried out.The phase hologram is used as the control signal to load to SLM,and a large number of beam-splitting target images and detected results images are automatically collected based on the graphical user interface(GUI).(2)By the method of summation and peak-seeking,the beam section data in the image is extracted from simulation and experiment,the data set and test set required for supervised learning are constructed,and a total of 1000 groups of samples-labels are collected,and the training set and test set are randomly generated,Among them,800 groups construct training set and 200 groups construct test set.(3)Build a reverse CNN,the image detected by camera is utilized as a sample,the image of beam target is utilized as a label,the mapping relationship between the target image of beam splitting and the image detected by camera is established through the supervised learning regression task,so as to realize the prediction of the IFTA input amplitude distribution and effectively improve the accuracy of arbitrary multi-beams splitting.Using the multi-layers convolution layer network structure,MAPE(Mean Absolute Percentage Error)and MSE(Mean Square Error)are used as the loss function of the training process to complete the performance measurement of the learning task.(4)The prediction results are verified.Based on the forward CNN supervised learning regression task and the actual experimental research,the new detected results corresponding to the predicted beam target image are collected.The experimental results show that the beam splitting error based on machine learning is significantly reduced,and the MSE is reduced by 37 %,which effectively improves the laser beam splitting accuracy.The work in this paper has important theoretical guidance and technical reference value for the development of arbitrary multi-beams laser beam splitting technology and the cross-study of artificial intelligence in the field of laser.