Pulse Repetition Interval Modulation Recognition and Classification Based on Deep Convolutional Neural Networks Improved with Extreme Learning Machine

PRI modulation recognition and classification is a critical task in ESM and ELINT systems to accurately detect radar threats. However, this practice is challenging in a real environment due to missing pulses and spurious pulses and undesirable antenna scanning effects that lead to noisy PRI modulation sequences. To address this issue in this papper, three methods based on deep convolutional neural network (LeNet5, AlexNet, GooglNet) which have been optimized using extreme learning machine (ELM) have been proposed. In fact, in the first step, a deep convolutional neural network (DCNN) is used as a feature extractor. Then, in the second step extreme learning machine (ELM) is used for real-time recognition and classification of PRI modulation. To evaluate the proposed methods, data corresponding to the real data were designed and simulated, and all the destructive effects on the PRI sequence were considered. The results of simulations on 60,000 images show that the AlexNet-ELM network performs better in most of the evaluation criteria and has achieved a high accuracy of 93%.