Wheeled cable-climbing robots that can detect cable damages are widely used to liberate human resources and reduce bridge hazards. Most wheeled cable-climbing robots have poor stability and will deflect along the cable under uneven loading. In this paper, a visual servoing method for wheeled cable-climbing robots is proposed with applications in antideflection. The proposed method integrates visual localization and trajectory tracking control, enabling robots to precisely estimate the deflection angles and stably correct deflection in dynamic and complex environments. For visual localization, a deflection angle estimation algorithm that combines object detection and coplanar feature points matching is developed. Then, a reference image update strategy is used to overcome the visual servoing errors caused by images distortion. On the controlling side, the trajectory tracking of the wheeled cable-climbing robots is implemented under the framework of model predictive control (MPC) with constraint functions to achieve expected position while avoiding wheel slipping. Based on the Multibody and V-Realm simulation platforms, the visual servoing method of wheeled cable-climbing robots is demonstrated, and its effectiveness and accuracy are verified by experiments.