A Study on Near-field Target Localization Algorithm Based on Residual Network

The wavefront shape of near-field target has a nonlinear variation with respect to array position, and the position of the target must be determined by the distance as well as the direction of arrival, so the traditional methods often have limitations in dealing with near-field target localization, and it is difficult to satisfy the demand for high-precision localization. To address the above problem, a near-field multi-dimensional target localization algorithm based on a residual neural network with a convolutional attention module(ResNet-CBAM) is proposed in this paper to realize high-precision localization on multi-dimensional signal processing. First, a cross-shaped multiple-input multiple-output radar model is constructed to realize the decoupling of angle and distance parameters by using the symmetry of the transmit and receive antennas. Then, the network dataset of angle and distance is constructed by extracting the real and imaginary parts of the upper triangular array using the Hermitian property of the array output covariance matrix. Finally, an improved ResNet-CBAM network is proposed, optimizing the network structure and introducing the CBAM attention mechanism in the residual block to obtain more critical complex features, which significantly improves the estimation accuracy. The experimental results show that the proposed algorithm still achieves accurate localization with low signal-to-noise ratio and fewer number of snapshots, and significantly outperforms the traditional method such as sparse Bayesian learning and convolutional neural network.