Strong scattering media brings great difficulties to optical imaging, which is also a problem that medical imaging and many other fields must face. Optical memory effect makes it possible to image through strong scattering medium. However, this method also has the limitation of limited angle field-of-view (FOV), which prevents it from being applied in practice. In this paper, a kind of practical convolutional neural network called PDSNet is proposed, which effectively breaks through the limitation of optical memory effect on image field angle. Experiments is conducted to prove that the scattered pattern can be reconstructed quickly and accurately by PDSNet, and it is widely applicable to restoring speckle pattern generated from objects of different scales and scattering medias with different statistical characteristics.

Learning-based real-time method to looking through scattering medium oriented to practical application