6 million sets of radon concentration simulation dataset: Based on the radon exhalation mechanism and statistical fluctuation characteristics of the medium surface

This dataset is designed to provide high-fidelity, reproducible training and test samples that support deep learning tasks for optimizing the measurement of radon exhalation rates on medium surfaces. All data are generated and processed in the MATLAB R2021b environment, relying solely on its built-in functions, and can be reproduced without the aid of additional hardware.The generation process is divided into two steps: Firstly, a 12-period theoretical truth sequence is generated based on the radon precipitation model on the medium surface; Subsequently, Poisson noise that conforms to statistical laws and instrument errors within the ±1 counting range were injected into it to form simulated observations with errors. Finally, pairs of samples with one-to-one corresponding theoretical true values and simulated observed values are obtained.The dataset is composed of 6 million pairs of samples, half of which are theoretical true values and the other half are simulated observations with added errors. Each sample contains 12 consecutive measurement periods, with radon concentrations covering a wide range from zero to over a hundred thousand, which can effectively support the training and validation of related deep learning models.