For absolute value terms related to yaw velocity generated by cross-flow effect and the unmodeled factors, a nonlinear parameter-varying (NPV) model is established with absolute value disturbances to regulate the heading of unmanned surface vehicle (USV) by state feedback (SF) control with a NPV disturbance observer (NPVDO). Firstly, the model of NPVDO is designed with gains to estimate the non-differentiable absolute disturbance. Secondly, a Lyapunov function is constructed with full states and varying parameter to solve the gains with NPVDO stability conditions by Euler homogeneity. Thirdly, since the NPVDO stability conditions contain the coupling term between the observer gain and the Lyapunov matrix, it is decoupled into NPVDO-sum of squares (SOS) conditions by the projection theorem and matrix transformation to solve NPVDO gains. Finally, a robust SF with NPVDO is designed for heading regulation with NPV model. Simulations and experiments indicate that NPVDO has a superior performance on parameter variation suppression and nonlinear non-differentiable disturbance estimation by comparing with other methods to reduce heading errors.