This paper presents an enhanced approach to automatic target recognition (ATR) in optical communication systems, leveraging generative adversarial networks (GANs) for data augmentation in few-shot learning scenarios. The proposed method integrates wavelet transform techniques with a GAN-based framework, aiming to improve the recognition of spot images under limited data availability. The system processes spot images using a multi-layer generator architecture, which applies wavelet decomposition to separate high- and low-frequency components of the input images, capturing both structural and detailed information. This enhancement allows the generator to produce high-quality, diverse synthetic samples, even in data-scarce environments. The discriminator network uses a multi-objective loss function that balances adversarial loss, classification loss, and perceptual consistency to ensure the realism and category alignment of generated images. Simulation results show that the proposed method significantly outperforms the traditional ones in terms of peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), and learned perceptual image patch similarity (LPIPS). In particular, the proposed method achieves a PSNR of 34.11 dB, SSIM of 0.74, and LPIPS of 0.3324, demonstrating its superior image quality and perceptual alignment, due to the combining GANs with wavelet decomposition for improving ATR systems in resource-constrained environments.
Jin Jin (Mon,) studied this question.