Research On Multipoint Light Focusing Based On Enhanced Multiobjective Optimized Wavefront Shaping

The multiple scattering phenomenon of light passing through the scattering medium cause light to lose its original propagation direction and form optical speckle.This is due to the spatially non-uniformly distributed refractive index inside the scattering medium.This phenomenon of multiple scattering limits the development and application of many optical technologies,especially in the field of biomedicine.Wavefront shaping is a method of overcoming multiple scattering phenomena by constructing specific incident light waves to focus light at any desired location.Currently,wavefront shaping technology has been greatly developed,and researchers have proposed a variety of wavefront shaping methods that can achieve focusing of light after multiple scattering.However,compared to the widely studied single point focusing,research into multi-point focusing is still in its preliminary development stage.Using a single mask to generate multiple focal points simultaneously is more meaningful for biomedical research such as brain research and disease diagnosis.Unlike single point focusing,multi point focusing requires simultaneous attention to two objectives: enhancement factor and uniformity,and is therefore equivalent to a multi target problem.The most widely used feedback wavefront shaping method has received the favor of researchers due to its strong noise resistance and excellent focusing effect.However,its biggest disadvantage is that the optimization time is too long.Feedback optimization algorithms are limited by the feedback mechanism and the frame rate of hardware devices,it usually require up to tens of minutes to complete.In addition,the performance of feedback algorithms also greatly affects the focusing results.The artificial intelligence method that has emerged in recent years has also been used for wavefront shaping technology.The trained neural network can achieve rapid focus,which is very beneficial for saving the time of wavefront shaping.However,the performance of the network largely depends on the quality and quantity of training samples.The enhancement factor of the final focus is low,and the uniformity of multiple focus points cannot be guaranteed.To address the aforesaid challenges,this thesis combines neural networks and multi-objective optimization algorithms and utilizes their complementarity to propose an enhanced multi-objective algorithm: neural network enhanced non dominated sorting genetic algorithm(NN-NSGA2).In this hybrid algorithm,the mask predicted by the neural network is used as the initial mask of the NN-NSGA2 algorithm,which is optimized based on this initial mask.This method allows NN-NSGA2 algorithms to iterate on the basis of a relatively good initial solution,thereby avoiding optimization failures caused by the algorithm’s sensitivity to the initial solution.It can also help algorithm save optimization time and accelerate the optimization process.In this paper,NN-NSGA2 algorithm is simulated to verify the feasibility of the algorithm.The simulation compares the effects of the NNNSGA2 algorithm proposed in this paper with the NSGA2 algorithm when the number of focal points is different.The results indicate that the method in this paper can obtain higher enhancement factor values after optimization,and can help reduce algorithm optimization time: when the two algorithms reach the same enhancement factor value,the NN-NAGA2 algorithm uses fewer iterations.The superior performance of this algorithm is further demonstrated in experiments.In the experiment,a single layer neural network was successfully trained using 30000 sets of data,and multiple focus masks were predicted.The predicted masks had preliminary focus effects.The NN-NSGA2 algorithm using neural networks to predict masks successfully achieves single,three,and nine point focusing.The three cases are compared with NSGA2 algorithm,which is consistent with data simulation.This algorithm ultimately has higher enhancement and less optimization time,which increases the enhancement factor value by about 10% compared to the unenhanced multi-objective optimization algorithm and saves about 30% of optimization time,proving its strong focusing ability in multi-point light focusing.