RCS Reduction for Multiple-Sparsity-Rate Arrays With a Feature Multitask Network

Abstract

Deep learning networks have been widely used in sparse array optimization in recent years. However, these networks are designed for a single sparsity rate, which makes them unsuitable for reducing the radar cross section (RCS) of sparse arrays with multiple sparsity rates. In this article, multitask learning is first applied to address this limitation. Subsequently, we carefully design a feature multitask network (FMTN). The proposed FMTN offers two main improvements over current multitask networks. First, we present a deep sharing (DS) strategy that increases the generalizability of the network and compresses the network. Second, fully connected layers are replaced with multiple convolutional layers to reduce the complexity and increase the network's nonlinearity. Finally, the simulation results demonstrate that the proposed FMTN achieves higher classification accuracy, greater stealth capabilities, and lower complexity than existing multitask networks do.