Abstract: It is of significant engineering value to implement differential support strategies for roadways based on precise evaluations of heterogeneous rock mass quality, as this approach can reduce support costs while ensuring roadway safety. In this study, deep learning is employed to automatically calculate Rock Quality Designation (RQD) from borehole core images, while discontinuities are identified from rock mass exposures using photogrammetry and a shortest-path search algorithm. This enables fast and accurate rock mass quality assessment. Additionally, geostatistical methods are applied to analyze the spatial correlation of measured rock mass quality, deriving an anisotropic semivariogram. A block model of heterogeneous rock mass quality is then established using spatial interpolation. Using the Jiama Copper Mine as a case study, differential roadway support is designed based on the heterogeneous rock mass quality (Q-system). The results demonstrate that rock mass quality is both heterogeneous and continuous within the same lithology, but exhibits strong variability at lithological boundaries. By considering heterogeneous rock mass quality, optimal roadway positioning, support methods, and parameters can be determined. This study aims to establish a systematic framework for precise rock mass quality evaluation and differentiated roadway support, ultimately achieving cost-effective and efficient mine support solutions.