Gaussian Machine Learning Model Atmospheric Correction

The patent introduces an intermediary deep learning model that acts as a mediator between the input digital image and the corrected output. This model learns the complex mapping relationship between radiance and reflectance without requiring ground truth data, enabling accurate atmospheric correction even when ground reflectance values are unavailable. The model serves as an intelligent intermediary that captures atmospheric effects and reverses them through learned transformations.

The patent transforms the atmospheric correction problem by changing the parameters used in the correction process. Instead of relying on traditional parameters like ground reflectance values, the system uses learned parameters from the deep learning model that are optimized during training. This parameter transformation allows the system to achieve high accuracy without conventional input parameters that are often unavailable in practice.

The patent applies preliminary action by pre-training the deep learning model on a large dataset of radiance-reflectance pairs before deployment. This offline training phase captures the complex atmospheric correction mappings in advance, so that during actual operation, the correction is performed through efficient forward propagation rather than iterative computation. This preliminary preparation dramatically reduces the computational time required during practical application.

The patent replaces traditional mechanical/computational atmospheric correction methods with a data-driven deep learning approach. Instead of using iterative mathematical optimization or physics-based models that require extensive computation, the system uses a trained neural network that performs correction through efficient matrix operations, significantly improving computational efficiency and reducing processing time.