Ship Formation Identification Method for HFSWR based on Deep Learning

The multi-ship formation (MSF) identification method on the large-scale RD image has been proven to separate formations, single-target, and clutter backgrounds. Due to the low-resolution, the imaging bias caused by measure errors for HFSWR is the principal problem of adopting the deep learning network to identify MSF. In this paper, we construct the deep learning dataset with parameters
$R_{0}$
,
${\theta }_0$
and
$v_{ship}$
variations. Then, we verify the identification performance and transfer learning performance of the CNN-ELM on the new dataset. To some extent, the CNN-ELM network can ignore the impact of radar measurement parameter fluctuations and accurately identify MSF. Compared with the published paper, there is a decrease in the accuracy of the CNN-ELM, Alexnet, and Resnet 18. The subtle differences caused by measurement error cannot be completely overcome through transfer learning on the above three networks.