Abstract: By collecting the multi-lithology and multi-index drilling data, an automatic interpretation technology of tunnel rock mass integrity based on integrated algorithm and multi-scale model fusion is proposed considering comprehensive interpretation accuracy and practical effect. First, the pre-processing such as noise reduction and equidistant segmentations (0.5, 1 and 2 m) is carried out of the raw data to form a multi-scale, high-quality machine learning dataset. Then the operations such as automatic parameter optimization, training, evaluation and interpretability of model are performed to verify the accuracy and reliability. Finally, the weighted average method is used to fuse the multi-scale interpretation results to enhance the engineering practical effect. In addition, in order to facilitate practical engineering applications, a lightweight automatic interpretation platform is developed. The application results of several limestone and sandstone tunnels show that compared with the conventional interpretation, the multi-scale model fusion interpretation has the overall excellent performance in interpretation efficiency and prediction effect. It can provide reliable rock mass integrity information for the excavation and support of tunnel construction.