Abstract: ‍ ‍The pulse repetition interval （PRI） is a key parameter of a radar signal and plays an important role in radar performance. The research on radar PRI modulation types can provide important support for radar recognition and performance analysis， and is a key research subject in the field of electronic countermeasures. An intelligent recognition algorithm based on pulse-sequence reconstruction is proposed to improve the accuracy of PRI modulation-type recognition， which is greatly affected by missing pulses. First， an initial estimation is made of the PRI modulation period based on the multi-order difference sequence of the time of arrival data. Second， samples of different pulse sequences are reconstructed based on the PRI modulation period. Furthermore， the pre-processing of the reconstructed samples is achieved by deducting the mean value. A convolutional neural network is then constructed to identify the PRI modulation type. The proposed method can reduce the limitation of manual feature extraction. In addition to determining the deep features of a PRI sequence， it has greater environmental applicability. Finally， simulation results are presented to verify the efficiency of the proposed method. When the pulse loss rate is less than 60%， the recognition accuracy of the proposed method for all PRI modulation types is greater than 92%， and the average recognition accuracy is greater than 98%. When the pulse loss rate is less than 80%， the average recognition accuracy is above 81%.