<p>The rate of exhalation of radon from porous materials is a key parameter used to study the migration of radon and to assess environmental radiation hazards. Measurements of radon concentrations are often affected by statistical fluctuations and instrument errors, which lead to reduced measurement accuracy. To address this issue, we propose a deep learning method based on a hybrid model that combines a convolutional neural network with a transformer (CNN-Transformer). The proposed method is designed to optimize radon concentration data and improve the accuracy of radon exhalation rate fitting. The proposed method combines the advantages of one-dimensional convolutional neural networks in local feature extraction with the capabilities of the transformer model to represent time-series data. After training the model on 6 million samples of simulated radon concentration data, it exhibited strong generalization capability and high prediction accuracy across different radon concentration ranges. The optimized results from fitting the curves were close to the theoretical true values, with the coefficient of determination (<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(R^{2}\)</EquationSource> <EquationSource Format="MATHML"><math> <msup> <mi>R</mi> <mn>2</mn> </msup> </math></EquationSource> </InlineEquation>) remaining stable between 0.97 and 0.99. These results significantly outperformed simulated instrument measurements and reduced the uncertainty of the data. In an experimental evaluation of the performance of the proposed approach for practical applications, optimization predictions were generated by combining the proposed solid radon source exhalation reference model with measurement data collected from the RAD7 radon detector. The fitting results for each experimental group showed a clear improvement. The optimized radon exhalation rate was much closer to the reference value (<InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(43 \pm 0.48~\mathrm {mBq \, m^{-2} \, s^{-1}}\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>43</mn> <mo>±</mo> <mn>0.48</mn> <mspace width="3.33333pt" /> <mrow> <mi mathvariant="normal">mBq</mi> <mspace width="0.166667em" /> <msup> <mi mathvariant="normal">m</mi> <mrow> <mo>-</mo> <mn>2</mn> </mrow> </msup> <mspace width="0.166667em" /> <msup> <mi mathvariant="normal">s</mi> <mrow> <mo>-</mo> <mn>1</mn> </mrow> </msup> </mrow> </mrow> </math></EquationSource> </InlineEquation>), with a significant reduction in deviation. The optimized fitting coefficient of determination (<InlineEquation ID="IEq3"> <EquationSource Format="TEX">\(R^{2}\)</EquationSource> <EquationSource Format="MATHML"><math> <msup> <mi>R</mi> <mn>2</mn> </msup> </math></EquationSource> </InlineEquation>) remained stable between 0.988 and 0.996, which was significantly higher than the results obtained from measurement instruments (<InlineEquation ID="IEq4"> <EquationSource Format="TEX">\(R^{2}\)</EquationSource> <EquationSource Format="MATHML"><math> <msup> <mi>R</mi> <mn>2</mn> </msup> </math></EquationSource> </InlineEquation>, which ranged from 0.851 to 0.973). The proposed CNN-Transformer hybrid model provides an innovative approach to optimize data on the rate of radon exhalation from various materials. Thus, the results show that the proposed approach improved measurement accuracy while reducing errors and showed significant potential for practical applications.</p>