Translational Motion Compensation of Space Micromotion Targets Using Regression Network

The high-speed translational motion of space targets will cause the micro-Doppler to shift, tilt, and fold, which brings great difficulty to the extraction of micromotion features. Translational motion must be compensated in advance to extract the authentic characteristics of micro-Doppler curves. To solve the problem of translational motion compensation, an estimation method of translational parameters based on deep learning theory is proposed. A polynomial model describing translational motion is constructed firstly by analyzing its dynamic principle. Meanwhile, the parametric characterization of micromotion signal is deduced by taking the cone target as an example. On this basis, two regression networks that can estimate acceleration and velocity respectively from the time-frequency graph are built using transfer learning. The translational motion compensation is accomplished with high accuracy and low computation complexity. The proposed method can also achieve satisfactory results in the presence of high intensity noise and discontinuous scattering points. Finally, the effectiveness and robustness of the proposed method are validated by simulations.