HRRP data augmentation based on Conditional Generative adversarial networks

In the field of radar target recognition, the scarcity of non-cooperative target data and the limitations of traditional HRRP data enhancement methods have severely restricted the improvement of recognition accuracy. This paper aims to break through this predicament, propose an HRRP data augmentation method based on the self-attention mechanism to improve the Conditional Generative Adversarial Network (CGAN), and apply it to radar target recognition. The research first analyzes the basic principle of the Generative Adversarial Network (GAN) and clarifies its generation and discrimination adversarial process. Subsequently, the improved method based on CGAN is elaborated in detail. By introducing conditional information, the generator can generate samples of specific categories. Integrate the self-attention mechanism to effectively extract the correlation information among the features of HRRP data. Experiments were carried out using the existing HRRP dataset. After data preprocessing, the improved CGAN model was trained, and the target recognition effects of the datasets before and after enhancement were compared with the aid of the convolutional neural network (CNN). The experimental results show that the CGAN enhanced dataset based on the self-attention mechanism improves the recognition accuracy of the three targets by 6.9%, 5.8% and 2.2% respectively, which is significantly superior to the traditional time-shift, mirror image enhancement and ordinary CGAN methods. This indicates that this method can effectively expand the data, significantly improve the accuracy of radar target recognition, and has a good performance on different datasets. This method has good robustness and provides new ideas and effective means for the research of radar target recognition.