Enhancing signal extraction and image reconstruction through scattering media using semi-supervised learning methods

Laser imaging systems in scattering environments are typically affected by the effects of medium scattering, resulting in systems that cannot effectively detect objects hidden behind the scattering medium. Supervised learning-based signal extraction and image reconstruction methods can reconstruct the target image, but these methods require a large amount of manually labeled data, and manually labeling signals under different conditions is both laborious and impractical. For this problem, this study proposes a semi-supervised learning-based signal extraction and image reconstruction method. This method is based on the discrepancy between the time profiles of the target reflected signal and the backscattered noise. The proposed method exhibited superior signal extraction and image reconstruction capabilities in strong dynamic scattering environments, and proved that the generalization capability of the model can be improved by utilizing a large amount of unlabeled data. This study can significantly reduce the dependence of the signal extraction method on labeled dataset, which is beneficial for practical applications.