Abstract: In order to study the real-time identification of the strength and structural state of the roof strata during the drilling process of the anchor drilling rig, and to provide a technical way for the detection of roof hidden dangers in the process of roadway excavation, a comprehensive research method was used to reveal the drilling signal response law of different strength rock mass and combined rock mass. The rock mass strength characterization and prediction model based on drilling data was studied, and its suitability was compared and evaluated. The results show that the signals such as thrust, torque and rotational speed have strong responsiveness to the change of surrounding rock strength, which is consistent with the change of rock structure. The response of triaxial vibration to rock mass strength is not obvious, but the signal mutation at the rock interface is significant. The trend of each characterization parameter and its strength characterization value is basically consistent with the trend of drilling signal. The uniaxial compressive strength characterization model based on the unit volume grinding energy of the drill bit has the smallest deviation from the real value (optimal adaptability). The thrust and torque have obvious linear positive correlation with other parameters except the three-axis vibration, and the rotational speed also has obvious linear correlation, but it has a negative correlation trend with other characteristics except the drill-ability index. In the three-axis vibration, the X-direction and Y-direction vibration have little correlation with the other characteristics, but there is still a certain correlation, and the correlation in the Z-direction vibration is very small. The Attention-CNN model performs best in robustness, accuracy and adaptability. The coefficient of determination R² between the predicted value and the real value obtained from the uniaxial compression test is 0.94, and the fitting degree is the highest. The prediction accuracy of the uniaxial compressive strength of rock mass is at least 12 % higher than that of other models. The research results provide theoretical and technical support for real-time perception and support decision-making of roadway roof rock strength.