表面缺陷类型的判定和尺寸识别对于精密光学元件表面质量评价至关重要.针对利用角分辨散射信号反演表面缺陷结构特征参数时,传统反演算法在反演维度和尺度上存在的局限,提出一种基于决策树模型的级联机器学习反演算法.为了构建训练模型所需的数据集,基于时域有限差分(FDTD)法建立了角分辨散射系统的电磁仿真模型,通过仿真计算获得数据集.测试集数据的反演结果表明,所建立的算法分别以0.99的area under the curve(AUC)和平均0.932的R2实现缺陷类型和深度的预测,拓宽了反演维度,并且能够以平均0.997的R2在各缺陷深度尺度范围下实现缺陷宽度的准确预测,拓展了反演尺度.所提算法为精确定量分析光学元件表面的微小尺寸缺陷提供了新思路.
Inversion of Light Scattering for Optical Component Defects Using a Cascaded Machine Learning Algorithm
Determining the types and sizes of surface defects is crucial for an evaluation of the surface quality of precision optical components.We propose a cascaded inversion algorithm based on a decision tree model to address the limitations of traditional inversion algorithms in terms of inversion dimension and scale when angle-resolved scattering signals are used to invert the structural characteristic parameters of surface defects.To construct the dataset needed to train the model,an electromagnetic simulation of the angle-resolved scattering system was established using the finite difference time domain method,and the dataset was obtained through simulation calculations.The inversion results for the test set data show that the proposed algorithm is able to predict the defect type and depth with a precision having an area under the curve of 0.99 and an average R2 of 0.932,expanding the inversion dimension.The algorithm also accurately predicts the width of defects at different defect depths with an average R2 of 0.997,increasing the scale of inversion.The proposed algorithm offers a new approach to the precise quantitative analysis of small defects on the surface of optical components.
NETL
NSTL
万方数据
