We observed a phenomenon that different scattering components have different decorrelation time. Based on decorrelation time difference, we proposed a method to image an object hidden behind a turbid medium in a reflection mode. In order to suppress the big disturbance caused by reflection and back scattering, two frames of speckles are recorded in sequence, and their difference is used for image reconstruction. Our method is immune to both medium motions and object movements.

针对非相干光照明下的非视域成像问题,提出一种基于深度学习的解决方法。结合计算机视觉领域中经典的语义分割及残差模型,构造一种URNet网络结构,并改进了经典瓶颈层结构。实验结果表明,改进的网络可以恢复更多的图像细节,并具有一定泛化性,相比于基于非相干光照明的散斑自相关成像技术,该网络恢复性能有较大提升。