High-Performance Polarization Remote Sensing With the Modified U-Net Based Deep-Learning Network

The existing scattering medium has a serious impact on the optical remote sensing. It is a challenge and urgent problem to extract the original target's information influenced by scattering systems. Here, we adopt the target's polarization information as the original data and load it into the modified U-net-based deep-learning network (MU-DLN) to retrieve the original target's information influenced by the scattering medium. The dense blocks in the MU-DLN can extract the features of the target information contained in the polarization images. Meanwhile, the peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) have been used to evaluate the quality of target reconstruction. The experimental results show that our strategy can reconstruct the target's information very well, and the model trained for a fixed optical thickness (OT) environment can also be used for remote sensing in the larger or smaller OT environment within a certain range. In addition, due to the polarization imaging characteristics, compared with traditional methods, our strategy can improve the quality of target reconstruction effectively. Our work provides a new direction for DL technique in remote sensing of targets' information from the complex scattering systems.