Abstract: Underground Karst formations pose significant safety risks to the design, construction, and safe operation of land transportation projects. To investigate the development of underground Karst, comprehensive multi-source heterogeneous survey data were utilized, including remote sensing interpretation, engineering geological mapping, drill hole data, and the results from electrical resistivity tomography and transient electromagnetic methods. These data were unified in terms of temporal and spatial coordinates and standards to achieve data-level fusion of adverse Karst geological bodies. Based on this foundation, a recognition framework for underground Karst geological bodies was established. A method for assigning initial basic probability assignment(BPA) functions to underground spatial point domain was developed, and a modified Dempster-Shafer (DS) evidence theory algorithm based on Kendall’s correlation coefficient was employed for multi-source data fusion to obtain Karst evaluation indices. After 3D spatial interpolation, 3D imaging of Karst features was plotted. The results indicate that the improved DS algorithm effectively solves the high conflicts among comprehensive survey evidence, enabling intelligent decision-making regarding Karst targets. This improved DS algorithm facilitated the integration of Karst geological information features from comprehensive survey results. The 3D imaging of the integrated results enhanced the reliability and accuracy of underground Karst geological body investigations, increasing work efficiency by 30% or more, and providing methodological guidance and evaluation tools for engineering design, construction, and operation.