In order to improve the efficiency and accuracy of safety risk management in machinery manufacturing enterprises，the Bayesian network and machine vision technology were combined. Based on improved YOLOv5，Intersection over Union(IoU) values of safety hazard events occurring at the operation site were calculated. By leveraging the audit risk assessment in conjunction with AHP to derive the danger weights，the prior probabilities of the root nodes of Bayesian network were determined. Bayesian network model and design management system were established to realize closed-loop control. A safety risk management model of machinery manufacturing enterprises was constructed and verified by examples. The results show that the model has a more accurate identification and evaluation ability，and can find some potential safety hazards，so as to optimize the current management process. At the same time，the model also successfully realizes the effective combination of qualitative and quantitative analysis，integrates the expert experience and data quantification results，and confirms each other，so that the risk assessment results have a certain improvement in scientificity and reliability，which can provide a practical new idea for safety risk management.