One to multiple mapping dual learning: Learning multiple signals from one mixture

Liu, Ting; Wang, Wenwu; Zhang, Xiaofei; Guo, Yina

doi:10.1016/j.dsp.2022.103686

Cited by 2 publications

(3 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We compare quantitatively and qualitatively the proposed Transformer-guided GAN with the frontier algorithms FastICA [ 2 ], NMF [ 4 ], Neural egg separation (NES) [ 24 ], AGAN [ 12 ], and PDualGAN [ 11 ] to indicate its superiority on SCBIS tasks. To ensure a fair comparison, the AGAN has the same UNet-GAN structure parameters as configured in this paper, except that the channel reduction factor k is set to 8 in the self-attentive module; the PDualGAN consists of 2 DualGANs, each using 2 UNet-GANs in the same configuration as this paper to implement mix to source mapping.…”

Section: Resultsmentioning

confidence: 99%

“…where denotes the dot product operation, √ d k is used to scale the dot into the result to make the gradient of the model more stable, and softmax is used calculate the attention score by normalization, as shown in Equation (10). Finally, the output of MHA is obtained by concatenating the attention results of the obtained individual head vectors, as shown in Equation (11).…”

Section: Transformer Schemementioning

confidence: 99%

“…Ref. [ 11 ] treated the source and observed signals as information in different areas and solved the multisource single-channel blind split problem by combining multiple UNet-model-based DualGAN networks to train PDualGAN, but the detail representation of the source signal was not taken into account in the generation process. Ref.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Single-Channel Blind Image Separation Based on Transformer-Guided GAN

Jia

Shen

et al. 2023

Sensors

View full text Add to dashboard Cite

Blind source separation (BSS) has been a great challenge in the field of signal processing due to the unknown distribution of the source signal and the mixing matrix. Traditional methods based on statistics and information theory use prior information such as source distribution independence, non-Gaussianity, sparsity, etc. to solve this problem. Generative adversarial networks (GANs) learn source distributions through games without being constrained by statistical properties. However, the current blind image separation methods based on GANs ignores the reconstruction of the structure and details of the separated image, resulting in residual interference source information in the generated results. This paper proposes a Transformer-guided GAN guided by an attention mechanism. Through the adversarial training of the generator and the discriminator, U-shaped Network (UNet) is used to fuse the convolutional layer features to reconstruct the structure of the separated image, and Transformer is used to calculate the position attention and guide the detailed information. We validate our method with quantitative experiments, showing that it outperforms previous blind image separation algorithms in terms of PSNR and SSIM.

show abstract