Large-Scale Sparse Antenna Array Optimization for RCS Reduction With an AM-FCSN

The existing deep learning methods for radar cross section (RCS) reduction are unsuitable for large-scale sparse arrays because their large scales result in large numbers of classes and high network complexity levels. This article proposes a new deep learning method to solve these problems. First, a hybrid-interval particle swarm optimization (HIPSO) algorithm is presented. The number of classes is reduced using the presented HIPSO algorithm, which adaptively adjusts the sampling interval. Then, a fully convolutional shortcut network based on an attention mechanism (AM-FCSN) is designed. The optimal spatial arrangement is selected by the designed AM-FCSN. Finally, simulations show that the proposed HIPSO algorithm reduces the number of classes from <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${O}\text {(} {{10}^{{4}}} \text {)}$ </tex-math></inline-formula> to <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${O}\text {(} {{10}^{{2}}} \text {)}$ </tex-math></inline-formula>. Moreover, compared with different neural networks, the proposed AM-FCSN achieves computational complexity and parameter complexity reductions of 38.46% and 80.2%, respectively, while attaining higher accuracy. The proposed method can effectively reduce the large-scale sparse array RCS in real time.