A multi-feature joint method for radar pulse repetition interval （PRI） modulation type recognition is proposed to address the challenges of low recognition accuracy， significant impacts from pulse loss and false pulses， and sensitivity to decision threshold settings in existing types. First， the concept of density clustering is employed to remove irrelevant pulses， enabling the preprocessing of the pulse train and mitigating the effects of pulse loss and spurious pulses on pulse train patterns. Statistical methods are then applied to extract the stability features， second-order differential variation features， continuity features and pulse pair matching features of signal time difference. Finally， a random forest algorithm is used to train and test the extracted features. The simulation results demonstrate that the proposed method achieves high recognition accuracy for PRI modulation types and exhibits robust resistance to pulse loss and spurious pulses.
