Feature Fusion CGAN Based HRRP Denoising and Reconstruction Method

This paper addresses the problem of radar automatic target recognition using high resolution range profile (HRRP) data under noise interference by putting forward a denoising and reconstruction method based on a feature fusion conditional generative adversarial network (CGAN). Compared to existing methods based on auto-encoder models that capture only local data characteristics, the proposed CGAN effectively learns the global distribution of HRRP data through adversarial training, where the generator follows an encoder-decoder structure and the discriminator is implemented as a multilayer perceptron. Additionally, to realize precise HRRP denoising and reconstruction, inspired by the application of radial length for coarse target classification, we introduce two simple yet innovative modules designed to extract high-dimensional representations of geometric information and identity information. These representations are then fused with the high-dimensional representation of HRRP extracted by the encoder and serve as input to the decoder. In our experiments, we employ a one-dimensional convolutional neural network to classify the denoised and reconstructed HRRPs and evaluate the effectiveness of the proposed method. The results demonstrate that under peak signal-to-noise ratio (PSNR) conditions of 20 dB, 10 dB, and 5 dB, the proposed method achieves superior performance in recognition accuracy, PSNR, and structural similarity compared to other methods on both simulated and measured datasets.