Imaging consecutive targets through scattering medium and around corners beyond the optical memory effect using untrained network
Yingjie Shi, Enlai Guo, Ming Sun, Lianfa Bai, Dan Mu, Jing Han
Abstract
Limited by the existence of the optical Memory Effect(ME), the speckle correlation technique can only restore the target in a small angle of Field of View(FOV). In this paper, a method based on untrained network is proposed for imaging consecutive targets beyond the ME range. The proposed method only needs collected speckle and unpaired real data, and it can reconstruct the consecutive target non-intrusively. By combining the data-driven method with the speckle correlation theory, the advanced training process of neural network is replaced by online training. Compared with the traditional methods, the proposed method no longer restricts the sub-targets to be independent and the distance between sub-targets is beyond the ME range. The proposed method realizes the imaging of moving targets beyond 1.78 times the ME range that is hidden behind dynamic media. These advantages enhance the possibility of the proposed method to restore the target in a non-laboratory environment.