资源环境科技发展态势分析平台

The stochastic models and deep-learning models are the two most commonly used methods for subsurface sedimentary structures identification. The results from the stochastic models typically involve uncertainty due to their nature. For the deep-learning models, sufficient structure samples are necessary for training, but they are practically difficult to obtain. This study develops an inversion framework to combine the strength of these two models to overcome the limitations. The stochastic model is first adopted to generate the structure samples required by the deep-learning models by integrating available observations. Then the trained deep-learning model is utilized to reduce the uncertainty of the structures generated by the stochastic models. This integrated framework can successfully estimate the structures using available observations. Importantly, no additional structure training samples are required in the identification process. To summarize, the combination of the stochastic and the deep-learning models shows great advantages in identifying subsurface sedimentary structures.