Aiming at the problems of slow response and low efficiency faced by traditional control methods in gait switching of quadruped robots, a CPG (Central Pattern Generator) network based on fully symmetric topology is constructed by using Matsuoka neural oscillator, and the CPG signals are converted into joint angle signals of target foot trajectories by using composite trigonometric functions. While gait matrix is used to define the gait of quadruped robot, TIS (Transient Interference Signal) is introduced into CPG network to adjust the phase transfer of CPG signal at any time, improving the switching speed of robot from walk gait to trot gait. The computer simulation and the experiment based on FlyDog quadruped robot platform verify that the method is effective and the robot runs smoothly during gait switching.