基于物理信息神经网络的光斑质心计算

Calculation of spot entroid based on physical informed neural networks

为了实现强噪声干扰下的远场光斑质心高精度计算,研究了一种基于物理信息神经网络的质心定位方法—质心物理信息神经网络(centroid-PINN),该方法利用U-Net神经网络优化质心计算误差损失.为了验证该方法,利用模拟产生不同强度的两种类型噪声(斜坡噪声和白噪声)干扰下的高斯光斑训练网络.通过两种类型的光斑(高斯光斑和类Sinc函数光斑)测试神经网络,均得到了较高的质心定位精度.相比传统质心定位计算方法,centroid-PINN无需根据噪声水平设置参数,特别是能够处理斜坡噪声的干扰,获得高精度定位结果.成果可用于高性能激光光斑质心参数测量设备的研制,对于夏克-哈特曼波前测量装置的研制也有一定的借鉴意义.

To determine the centroid of far-field laser beam spot with high precision and accuracy under intense noise contamination,a positioning algorithm named centroid-PINN is proposed,which is based on physical information neural network.A U-Net neural network is utilized to optimize the centroid estimation error.In order to demonstrate this new method,Gaussian spots polluted by two kinds of noises,i.e.ramp noise and white noise,are generated by simulation to train the neural network.The neural network is tested by two kinds of spots,i.e.Gaussian spot and Sinc-like spot.Both are predicted with high accuracy.Compared with traditional centroid method,the centroid-PINN needs no parameter tuning,especially can cope with ramp noise interference with high accuracy.This work will be conducive to developing the far-field laser beam spot measurement device,and can also serve as a reference for developing the Shack-Hartmann wavefront sensor.