Currently, the most robot arm controllers rely solely on either force information or visual information. However, controllers utilizing only one type of information often fail to meet the requirements of most tasks. A visual impedance control method is proposed, which integrates information from vision and end-effector force sensors. This method utilizes a Multi-modal synergistic approach to combine force and visual feedback. The impedance parameters in the visual impedance system are adaptive, and the PPO (Proximal Policy Optimization) algorithm in deep reinforcement learning is used to find the optimal impedance parameters. The experiment results showed that the adaptive visual impedance control method adjusts impedance parameters through deep reinforcement learning, resulting in a 200% increase in force control accuracy compared to visual servo control. The Adaptive control method improves the reliability of the robotic system in interacting with task environments.