SAR-GMTI for Slow Moving Target Based on Neural Network

Abstract

As an important applications of synthetic aperture radar (SAR), slow moving target detection is causing more concern from people. Commonly, with the increase of computer computational efficiency and utilization of GPU, convolutional neural network has been becoming an efficient approach for target detection and classification. Here we propose a method using Faster R-CNN to detect the slow moving target in SAR images. When using existing datasets to detect moving targets, the detection accuracy is low due to the small defocusing of slow moving targets. So we use the bidirectional imaging mode to create the dataset. By increasing the displacement, it is more conducive to detect slow moving targets. At the same time, neural network provides a feasible way for target detection in this mode. In order to close to the reality echo, we use FEKO to simulate the target echo and use measured ground data to generate the ground echo. Deep learning combined with the forward and backward beams can detect slow moving target more effectively. The simulation results validate the effectiveness of the proposed method.