ISAR Resolution Enhancement Using Residual Network
Dan Qin, Diyang Liu, Xunzhang Gao, Gao Jingkun · 2019
Following the breakthroughs in natural image super-resolution task using faster and deeper neural networks, a method for inverse synthetic aperture radar (ISAR) resolution enhancement using residual network is proposed in this paper. Our method learns an end-to-end mapping between low/high-resolution ISAR images. The mapping is represented as a residual network that takes the low-resolution ISAR image as its input and outputs the high-resolution one. In our residual network, residual blocks without batch normalization (BN) layers are applied to retain the range of magnitude for ISAR images, parametric rectified linear unit (PReLU) is used as an activation function to adaptively learn negative coefficient. In addition, we propose a new evaluation index for ISAR resolution enhancement performance called IMV replacing 3dB bandwidth. Simulation and experimental results indicate that this method can achieve higher quality high-resolution ISAR image than traditional sparsity-driven method, which can improve the estimation accuracy of scattering points and recover weak scattering points.