A Case Study of Attention Mechanism in ML-Recognition of Highly Similar Low-RCS Targets

With the development of radar and stealth technology, target recognition based on machine learning (ML) has been widely used in radar field. However, for highly similar objects with low radar cross sections (RCS), the task of classifying ML objects is very difficult. In this paper, nine highly similar low RCS targets are taken as examples to construct a simulation dataset to study the role of attention mechanism in recognition of deep neural networks. In the experiment, a deep neural network method integrating the custom attention mechanism is used to strengthen the model's attention to the key scattering features of the target in the low frequency segment, which effectively improves the recognition accuracy. The experimental results show that compared with the baseline model without attention mechanism, the neural network model with attention mechanism can greatly improve the accuracy of low radar scattering cross section target recognition, which has important scientific significance and engineering application value.