Leader–follower formation tracking is surveyed for underactuated unmanned surface vehicles (USVs) with velocity constraints in this paper. The formation is expected to track a planned trajectory, and the planned trajectory is only accessible to the root leader. Each follower carries sensors for obtaining reference information. To tackle collision and connectivity constraints, error transformation is carried out on the dynamic surface with prescribed performance functions, and a concise dynamic surface that satisfies finite-time stability is created with K-class functions and barrier Lyapunov functions. Then, neural networks are utilized to approximate unknown disturbances, and a bounded dynamic controller is proposed based on the velocity estimation. Given that the planned trajectory might cause velocity saturation due to external disturbances and maneuvers for formation acquisition, a virtual leader is introduced to conduct trajectory improvement for anti-saturation control. Based on the saturated information, a feedback mechanism is proposed to adjust the velocity of the virtual leader and velocities of USVs simultaneously. Mathematical and numerical analyses are conducted to prove the feasibility of the controller and the mechanism.