The underground observation series are signals with non-Gauss noise which can’t be done by traditional filtering such as Kalman filtering (KF). Particle filtering(PF) does well in denoising the non-linear disturbed signals, but as the observing time extend, the PF will have problems with sample degeneration weight degeneracy. The paper presents new particle filter schemes which can be widely used in the field of chaos signal denoise and target identification, especially for underground signals. The approach improves the estimation accuracy without decreasing computing speed.