It is one of the key technologies of underwater pulsed laser detection to suppress the backscattered noise caused by water transmission in signal processing. Based on the correlation between two adjacent detections of backscattering in the same water area, an improved Normal Least Mean Square (NLMS) adaptive filtering algorithm is proposed. The step factor is controlled according to the intensity of backscattering and the size of the filtered target signal to make it achieve good filtering effect for target signals at different distances and intensities. The algorithm is verified by simulation and swimming pool experiment. The results show that the filter algorithm can effectively filter the backscattered noise with a lower order for targets with different reflection strengths at different distances, and retain the target echo signal. Compared with the existing adaptive filtering method, it greatly reduces the computing resources required for filtering, and has certain application value in underwater pulsed laser detection systems.