Neural Network Processing for Refractive Index Mapping from Light Paths
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods are limited in obtaining object information based on light beam paths that cannot be formulated as a mathematical formula, restricting the calculation of refractive index distribution.
Innovation Solution
A processing apparatus utilizing a neural network model to calculate refractive index distribution by discretizing light beam data and optimizing the model using an evaluation index, independent of time, to solve the ray equation for spatial coordinates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If analytical methods are used to calculate refractive index distribution from light beam paths, then calculation accuracy is improved, but the method is limited to cases where light beam paths can be formulated as mathematical formulas
Solution Approach 1:
The patent replaces traditional analytical mathematical methods with a neural network-based computational approach. The neural network is trained to predict refractive index distributions from light beam path data, enabling the system to handle complex cases where analytical formulations are not feasible while maintaining high calculation accuracy.
Solution Approach 2:
The patent transforms the problem from solving differential equations analytically to a parameter prediction problem. By representing the refractive index distribution as a set of parameters that the neural network learns to predict, the method extends applicability to cases where traditional analytical solutions cannot be obtained.
2Adaptability or versatility
If neural network models are used to calculate refractive index distribution, then adaptability to various light beam paths is improved, but computational complexity increases
Solution Approach 1:
The patent performs preliminary training of the neural network model using synthetic light beam path data generated from known refractive index distributions. This pre-training phase enables the model to learn the mapping relationship between light beam paths and refractive index distributions, so that during actual operation, the model can quickly predict results without requiring complex real-time calculations.
Solution Approach 2:
The patent uses synthetic data generated by ray tracing simulations to train the neural network. These synthetic light beam paths serve as training examples that replicate real measurement scenarios, allowing the model to learn from simulated data and generalize to real measurements without requiring complex computational resources during actual operation.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Enables the numerical calculation of refractive index distribution along light beam paths, even when analytical methods fail, providing accurate spatial information of objects.
Implementation Method 1
an evaluation index of the estimation model calculated from a ray equation independent of a time which the light beam path follows
Data Source
AI summary
According to the embodiment, a processing apparatus includes an arithmetic section. The arithmetic section is configured to calculate a refractive index distribution forming a light beam path based on an estimated output calculated by inputting light beam data indicating the light beam path to an estimation model, an updated output calculated based on the light beam data and the estimated output, and an evaluation index of the estimation model calculated from a ray equation independent of a time which the light beam path follows.