Enhancement-detection cascade network for SAR ground target detection
In synthetic aperture radar ground targets detection tasks,the conventional methods usually suffer from a severe performance loss for taking a fixed model assumption in processing.As a type of data-based method,the convolutional neural network can significantly improve the detection accuracy with a sufficient training set,but its performance is still instability when detecting tiny and blurred targets in strong ground background.To address these challenges,this article proposed a cascade network.A multi-function enhanced network based on a Transformer backbone is designed to focus on ground target characteristics.Specially,to heighten the discrimination of the target features,a cross-feature spatial attention module is then proposed following the enhancement.In the second stage,a detection network with multi-scale features is designed for detecting ground targets.A joint two-stage loss function is formed for achieving better reasoning performance.In extensive experiments on the Moving and Stationary Target Acquisition and Recognition(MSTAR)benchmark dataset,the results show that the proposed method can outperform other existing methods,concretely,the precision can reach 93.6% .
NETL
NSTL
万方数据
