Aimed at the problems that in extracting multipletarget state by PPHD filtering, the traditional K-Means clustering method exists in peak extraction, extended clustering time and incorrect clustering for clusters with different sizes, a new measurement extraction method is proposed based on K-neighborhood membership degree. In the category of measurement, estimation of a target is interrelated with the measurements and the particles, and the distance is used to discard false alarm measurements. The particle distributes to every actual measurement category of each estimation by K neighboring membership degree. On this basis, a new particle set is formulated, and target state can be extracted directly from the set, and there is no need to execute the peak extraction operation. The simulation results show that the proposed method is high in stable retrieval precision, and short in operation time compared with the K means method and free clustering method.