Synchronous locating and imaging behind scattering medium in a large depth based on deep learning
Scattering medium brings great difficulties to locate and image planar objects especially when the object has a large depth. In this letter, a novel learning-based method is presented to locate and image the object hidden behind a thin scattering diffuser. A multi-task network, named DINet, is constructed to predict the depth and the image of the hidden object from the captured speckle patterns. The provided experiments verify that the proposed method enables to locate the object with a depth mean error less than 0.05 mm, and image the object with an average PSNR above 24 dB, in a large depth ranging from 350 mm to 1150 mm. The constructed DINet can obtain multiple physical information via a single speckle pattern, including both the depth and image. Comparing with the traditional methods, it paves the way to the practical applications requiring large imaging depth of field behind scattering media.