Abstract: Reverberation is one of the most important factors that degrade the performance of active sonar detection.Traditional methods such as HOS prewhitening and modified-covariance prewhitening method have difficulties in dealing with static targets.One of the most dominant reasons is that such methods have poor ability to estimate parameters of an AR model from the nonlinear and varying statistics reverberation data.A novel method using support vector machine(SVM) to estimate the parameters of an AR model is presented.Experimental results show that the approach improves the detection performance significantly even if the target is static.