Siamese Network based Pulse and Signal Attribute Identification

Target identification and signal type differentiation is an essential factor in radar systems. Traditional methods to identify targets and signal types require deterministic approaches, which may need a lot of computational power and development time. With the use of the Siamese Network proposed in this work, the time needed to train and simulate the results is significantly reduced. This work presents a method to identify signals and pulses and the number of interleaved target signals in them. First, the received signal is processed through various windowing functions to achieve an appropriate signal-to-noise ratio with a low error percent of data loss. Next, the processed signal's Continuous Wavelet Transform is taken in order to simultaneously capture both the slowly varying fluctuations and the transient phenomena. Finally, the wavelet transform is inputted into the Siamese Network for identification and prediction. The Siamese Network used in this work was developed in MATLAB language for training, testing, and validation. The network simulation shows promising results and robust performance for signal type and number of signals identification.