Estimate The Initial Iterative Value Of Nonlinear Optimization Calculation Of Laser Phase

We generally use nonlinear optimization algorithms to iteratively calculate the laser phase in image-based laser phase detection technology.We need to set the initial iteration value in advance when using nonlinear optimization algorithms and the initial iteration value is usually set to zero or a random value,it may cause the nonlinear optimization algorithm to fall into a local minimum when iteratively calculates the laser phase and make the error of result of the algorithm and the true value larger.So,it is of great significance and practical value to study the influence of the setting of the initial iteration value of the nonlinear optimization algorithm on the laser phase detection accuracy.The phase diversity(PD)algorithm is widely used because of its simple system and the ability to detect continuous and discontinuous wavefront distortions.Therefore,this paper proposes geometric optics and deep learning methods to calculate the initial iterative value based on the PD algorithm optical model,and we make simulation experiments and actual experiments.The main research contents are as follows:(1)We proposed the geometric optics method to calculate the initial iteration value.First,we analyzed the optical model of the PD algorithm,proposed the principle of the geometric optics method,and verified it with simulation experiments.Second,we applied the geometric optics method to the PD laser phase detection for simulation experiments.The only variable in the experiment is different initial values(zero,random value,calculated value by geometric optics).The experimental results show that when the initial iterative value is the value calculated by the geometric optics method,the nonlinear optimization algorithm has a better effect in calculating the laser phase,and the geometric optics method can better avoid the nonlinear optimization algorithm from falling into the local minimum.(2)We propose a deep learning method to estimate the initial iteration value.First,we analyzed the framework of the deep learning method and introduced the method of generating data sets.Second,we selected the most representative convolutional neural network models(Alex Net,Vgg Net16,Res Net18,UNet),conducted a detailed analysis of the model's structural parameters,training results,and test results,and compared each model.Finally,we applied the optimal model to PD laser phase detection,and carried out simulation experiments with the initial value of zero,random value,and model predicted value as experimental variables.The experiment showed that when the initial iteration value was the model predicted value,the nonlinear optimization algorithm performs better and can better avoid falling into the local optimal situation.(3)We conducted practical experiments on geometric optics method and deep learning method.First,we set up the experimental light path and collected the experimental data according to the experimental steps.Second,we analyzed the true value of the experiment,and finally we processed and analyzed the experimental data.The experimental results verified that the geometric optics method and the deep learning method can better prevent the nonlinear optimization algorithm from falling into the local optimal situation,and verified the effectiveness of the geometric optics method and the deep learning method proposed in this paper.