Abstract: To address the challenges of real-time construction information acquisition and low historical data utilization in initial support decision-making for drill and blast tunnels,which lead to compromised timeliness and safety,this study proposes an intelligent support decision-making method integrating machine vision and multi-source data-driven approaches.Leveraging construction data from tunnels in the Yunnan Xuanwei-Huize Expressway project,a multi-source database containing 13-dimensional feature parameters was established.These parameters were acquired through convolutional neural network-based visual extraction,on-site recordings,and laboratory tests.An ensemble learning framework was developed to construct dual prediction models for support scheme classification and parameter regression:(1)An AdaBoost-based classification model achieved intelligent support scheme selection with precision,recall,and F1 scores of 0.960,0.978,and 0.966,respectively;(2)A Random Forest-based regression model realized quantitative support parameter prediction with MAE=0.0416 and R²=0.8275.Through grid search optimization and error cost evaluation,both models demonstrated superior performance compared to conventional machine learning methods.The research innovatively combines machine vision with multi-source data mining,providing an intelligent decision-making framework for initial support design in mountain tunnels,with significant practical value for engineering applications.