The detection accuracy of a yarn tension sensor using surface acoustic wave (SAW) devices has become increasingly important. We investigate a nonlinear compensation scheme based on SSA and SVR models to improve its detection accuracy, and the principle of SSA--SVR model and training method are also explored. We take the output frequency of the two sensors as input, the yarn tension applied to the working sensor as output, train an SSA--SVR model and use it for nonlinear compensation. We analyze and calculate the linearity, compensation accuracy and robustness of the SSA-SVR model, and compared it with the multiple regression model and BP neural network. The comparison results show that the SSA--SVR model has the best linearity, the highest compensation accuracy and the most robust. Finally, a novel nonlinear compensation scheme is proposed.