ID-SAbRUNet: Deep Neural Network for Disturbance Suppression of Drone ISAR Images

Herein, an approach for mitigating disturbances in real-field inverse synthetic aperture radar (ISAR) images of drones is introduced. This is achieved through the utilization of combining UNet, atrous spatial pyramid pooling (ASPP), and residual learning, referred to as ISAR Denoising Simplified ASPP based Residual UNet (ID-SAbRUNet). Starting with raw radar data maps obtained from real-field experiments, we extract data maps of size
${64} {}\times {}{64}$
centered around the target signal. Subsequently, the target and disturbance data are separated from the signal data. Also, only-disturbance data which were also acquired from the experiments are extracted to generate training, validation, and test datasets for ID-SAbRUNet. We extract feature maps of the proposed network after training to evaluate whether the training process is conducted as we expected from the design. Finally, we verify the performance by comparing the output results of ID-SAbRUNet with those of IUNet, a state-of-the-art deep learning algorithm designed for denoising ISAR data. This evaluation is conducted on the two test datasets that are divided into pretrained and nontrained datasets, to assess the performance of the proposed system. The results demonstrate that the proposed method outperforms the IUNet method on both datasets in terms of disturbance suppression and target restoration in most scenarios, regardless of the characteristics of disturbance.