Terahertz Image Super-Resolution Based on a Complex Network With Wavelet and Attention Mechanisms

Owing to the distinctive optical and electromagnetic properties of terahertz waves, terahertz imaging technology finds widespread application in security inspections, non-destructive testing, and various other domains. Nevertheless, its fixed longwavelength and scattering properties contribute to the degradation of image resolution. This letter introduces a complex-valued network augmented with wavelet and attention mechanisms for terahertz image super resolution. In the context of multiscale feature extraction within the network, we introduce a wavelet transformation module. Additionally, we have incorporated a multihead attention mechanism into the network, enabling the model to focus on more salient feature information. To further enhance the computational efficiency of complex convolutional neural networks, utilizing the Gaussian complex multiplication principle has led to the simplification of intricate convolutional network layers, resulting in a reduction of computational load by 25%. During the model training process, we have incorporated a simulated annealing algorithm to dynamically adjust the network parameters, aiming to elucidate an optimal parameter configuration. The results from both simulated experiments and laboratory experiments demonstrate that the proposed method substantially enhances the quality of THz images. The average resolution reaches 36.7 dB on the peak signal-to-noise ratio (PSNR) index and 0.98 on the structural similarity (SSIM) index.