Three-dimensional reconstruction of complex lithological ore districts based on a divide-and-conquer strategy: A case study of the Jiaojia gold deposit

Three-dimensional geological reconstruction is a core technical means for analyzing deep mineralization patterns and guiding prospecting predictions. However, reconstructing complex mining areas remains challenging. Traditional methods are often limited by sparse sampling, low efficiency, and difficulties in characterizing the dynamic evolution of complex geological attributes in deep environments. While machine learning shows potential, limitations persist in complex lithological systems like the Jiaojia gold deposit, characterized by multi-phase tectonic and alteration overlays. Therefore, this paper proposes a 3D reconstruction framework based on a divide-and-conquer strategy. The innovation of this framework lies in: utilizing an expert knowledge-driven semantic decoupling algorithm to decompose the complex lithological system into independent sub-features such as protolith, deformation, and alteration; constructing a cascaded KANRF (Kolmogorov-Arnold Network and Random Forest) prediction model to achieve collaborative governance of multiple attributes; and constraining the merging of composite lithologies and the reconstruction of lithological sequences based on expert knowledge. Experimental results quantitatively demonstrate the spatial pattern evolving from the fault core to the periphery, characterizing the ”structure-alteration-mineralization” trinity distribution. Comparative tests indicate that the divide-and-conquer strategy reduces local categorical complexity, increasing the weight of physical features in discrimination tasks. The KANRF model achieved 99.61% accuracy, significantly outperforming the full-scale direct classification model (87.32%), and showed excellent robustness in suppressing false positives and identifying small-sample features like silicification. In summary, this framework provides an effective solution for reconstructing deep complex lithological sequences, significantly supporting the analysis of metallogenic structures and guiding prospecting practices.