Ship HRRP target recognition against decoy jamming based on CNN-BiLSTM-SE model

Abstract

Due to the thinner resolution range of broadband radar, ship recognition issues arise such that minor fluctuations within the targeted area significantly affect the high-resolution range profile (HRRP) of ships. Especially in the presence of reflector decoys around the surroundings of a ship, the HRRP of mixed targets might take a vastly different shape than of single ship, which makes it difficult to capture the effective features for ship identification. This article proposes a novel radar target recognition model based on parallel neural networks. The framework of this model consists of two stages: the data preprocessing and the parallel neural network. The data preprocessing stage effectively solves the sensitivity issue of HRRP and maps one-dimensional HRRP into a two-dimensional image. The second stage employs CNN and bidirectional LSTM to extract overall envelope features and temporal features, respectively. The parallel features are then processed by the Squeeze Excitation (SE) block to enhance critical information. The experimental results, based on HRRP data from mixed targets of ships and reflector decoys, demonstrate that the proposed model outperforms other methods in recognition performance and is quite robust against small sample sets, high noise, and large amounts of decoy jamming.