Two-axis fast steering mirror based on flexure hinge support and voice coil motor drive is a strong coupling system with two inputs and two outputs. The coupling between X-axis and Y-axis greatly reduces the positioning accuracy of the fast steering mirror. It is difficult to achieve high precision decoupling control by using traditional PID control algorithm. Based on the centrosymmetric and axisymmetric two-axis fast steering mirror, the coupling sources of the two-axis fast steering mirror—DC coupling component and non-DC coupling component were analyzed theoretically, and the coupling physical model of between X-axis and Y-axis was established. A dual feedforward + dual neural network adaptive decoupling control algorithm was proposed to respectively compensate DC coupling components and non-DC coupling components. Experimental results show that, compared with the traditional PID control algorithm, the coupling degree of the proposed algorithm is reduced from about 5% to less than 1.0‰, which significantly improves the positioning accuracy from about 2.5% to less than 0.5‰.