<p>The application of high thermal conductivity materials to asphalt mixtures represents an effective solution to enhance the temperature adaptability and durability of pavements. However, the incorporation of these materials significantly alters the heat transfer mechanisms within asphalt mixtures making it difficult to measure their thermal conductivity coefficient accurately. This study aims to enhance the prediction accuracy of thermal conductivity models for asphalt mixtures with high thermal conductivity materials through fractal theory, employing laboratory experiments and theoretical derivation. The asphalt mixtures were identified into four components by the micro-computed tomography (CT) scanner and deep learning method. Next, the fractal dimensions of each material were calculated by box counting method. Through the fractal dimensions, the relationship between the content and gradation of high thermal conductivity materials and thermal conductivity was investigated. Finally, the fractal dimensions were incorporated to refine the series–parallel model, establishing an improved series–parallel fractal model whose accuracy was verified by experiments. The results show that the CT scanner and deep learning method were utilized to differentiate each material with high identification accuracy. The fractal dimension can characterize the content and gradation of high thermal conductivity materials, which can be used as an indicator to characterize the fine-scale distribution of the materials. Additionally, high thermal conductivity materials demonstrate the highest correlation with thermal conductivity of asphalt mixture, with a correlation coefficient of 0.6956. The accuracy of the improved series–parallel fractal model is higher than that of other composite thermal conductivity models, and the error is only 4.55% compared with the experimental values.</p> Graphical abstract <p></p>