基于U-Net网络的二维MIMO阵型稀疏成像

U-Net Network-Based Two-Dimensional MIMO Array Sparse Imaging

摘要: 传统雷达成像算法在处理不规则阵型的MIMO雷达成像时存在着诸多挑战。这些传统算法通常计算复杂且耗时长，难以满足MIMO雷达对实时成像的要求。因此，迫切需要一种适用于任意阵型且具备实时成像能力的新型成像算法。考虑到神经网络具有自学习、自组织和自适应等特点，并已广泛应用于雷达成像领域，因此提出了一种基于神经网络的二维MIMO不规则阵型稀疏成像算法。神经网络的训练通常需要大量的数据集，而构建回波数据集的过程耗时较长。为了解决这一问题，选择了U-Net网络结构，该结构在少量数据集下也能进行有效训练。我们将神经网络算法得到的成像结果与传统的BP算法进行了对比。结果表明，在相同数据集的条件下，神经网络算法表现出更优异的成像性能和更短的成像时间，从而实现了二维MIMO不规则阵型的高分辨率实时成像。

Abstract: Traditional radar imaging algorithms face numerous challenges when dealing with irregular MIMO radar imaging. These traditional algorithms are often complex and time-consuming, making it difficult to meet the real-time imaging requirements of MIMO radar. Therefore, there is an urgent need for a novel imaging algorithm that is applicable to arbitrary array configurations and possesses real-time imaging capabilities. Considering the self-learning, self-organizing, and adaptive characteristics of neural networks, which have been widely applied in the field of radar imaging, a two-dimensional MIMO irregular array sparse imaging algorithm based on neural networks is proposed. Neural network training typically requires large datasets, and the process of constructing echo datasets is time-consuming. To address this issue, the U-Net network structure was chosen, which can effectively train with small datasets. We compared the imaging results obtained from the neural network algorithm with those from the traditional BP algorithm. The results demonstrate that under the same dataset conditions, the neural network algorithm exhibits superior imaging performance and shorter imaging time, thereby achieving high-resolution real-time imaging of two-dimensional MIMO irregular arrays.