PTrans: Transformer-Based HRRP Target Recognition Method with Patching
Yichuan Jiang, Kejiang Chen, Weiming Zhang · 2024
Due to the characteristics of high-resolution range profiles (HRRP), which are easy to obtain, have low computational cost, and contain rich structural information about the target, HRRP-based target recognition has become a popular research direction in the field of radar automatic target recognition (RATR). To address the issues of traditional HRRP target recognition methods, which require time-consuming and labor-intensive, researchers, inspired by the powerful feature extraction capabilities of convolutional neural networks (CNN), have widely applied them to HRRP target recognition, significantly improving the accuracy and generalization of recognition methods. However, the inherent vulnerability of CNNs makes them susceptible to adversarial attacks, which significantly impact the recognition performance. Nevertheless, such adversarial attacks are typically constrained by real-world factors and can usually only generate adversarial samples by adding large, localized perturbations. Therefore, to defend against theses adversarial attacks, we propose a Transformer-based HRRP target recognition method, as Transformer enables target recognition through global feature. In order for the recognition model to simultaneously understand local feature, we propose a data patching strategy as a supplement. Experimental results show that our method has better recognition accuracy and robustness compared to existing methods.