Space Target Recognition Based on 4-D Range–Frequency–Time–Power Radar Data Cube

Radar-based space target recognition is crucial for the space target defense system. Employing micromotion features as a means to differentiate space targets has proven to be effective. However, most existing space target recognition methods based on deep neural networks make it difficult to analyze the underlying dependence between the four radar signal variables of time, range, frequency, and power. In this study, a framework for space target recognition is proposed, which utilizes a multidomain radar tool called the four-dimensional (4-D) range–frequency–time–power radar data cube to capture micromotion features. The radar echoes are first transformed into a series of high-resolution range–Doppler sequences. Next, the estimation method for scattering point information is applied to acquire four types of information related to the targets, which are subsequently used to generate the 4-D radar data cube. The resulting 4-D radar data cube is then inputted into a recently developed coordinate-temporal attention network (CTA-Net) to extract features and perform micromotion classification. Finally, an electromagnetic computation dataset is collected to validate the performance of CTA-Net. This research thoroughly investigates multiple crucial parameters of the dataset on recognition performance. In addition, the robustness of the proposed framework is demonstrated through a wide range of experimental results.