| Polarization imaging technology is an imaging technology that obtains information about the polarization state of the target through a light detector.This technology uses the uniqueness and difference of polarization information of scattered light field to extract the polarization information of target and background in turbid environment,analyze the polarization information of target image and background image,accurately estimate the light intensity change of target information light and background scattered light according to the polarization characteristics of target information light and background scattered light,effectively remove or suppress the background scattered light,enhance the contrast of target and background,and realize clear imaging underwater.Due to the influence of light absorption and light scattering,traditional polarization imaging methods have problems such as uneven light field and loss of target detail information.The traditional underwater active polarization imaging selects polarized light as the active light source and reconstructs the image by the physical model of underwater active polarization imaging,which solves the problem of the low contrast of imaging.However,this method is still affected by light absorption and light scattering.Based on the analysis of the current status of domestic and foreign research on active polarization imaging,this paper carries out systematic theoretical analysis,simulation analysis and experimental verification for light absorption,light scattering and polarization direction distribution.The specific research contents and main innovation points are as follows:Because of the problems of background interference,uneven optical field and limited transmission distance on the image clarity and short imaging distance of the traditional underwater active polarization imaging method,an underwater active polarization imaging method based on vector beam is proposed.The theoretical analysis of the effects of background interference,uneven optical field,and limited transmission distance on image clarity is first carried out.Then the polarization imaging method using a vector beam as the active light source irradiation is studied and proposed.The method is based on the traditional physical model of underwater active polarization imaging and introduces the vector beam as the active light source to take advantage of the structure,light field,and polarization characteristics of the vector beam compared with the linear polarization Gaussian beam for reconstructing target images in turbid environments.In addition,an underwater polarization Monte Carlo inversion imaging method is designed in this paper to simulate and analyze the underwater photon motion and imaging.Finally,optical imaging experiments are conducted to verify that the proposed method of underwater active polarization imaging based on vector beams is feasible and effective.The experimental results show that the proposed method can achieve better removal or suppression of scattering effects,solve the imaging distance problem,and improve the imaging clarity in different turbidity environments than the traditional underwater active polarization imaging method.To address the problems of uneven distribution of polarization directions and similarity of the intensity of scattered light components and the imaging information they contain in any polarization state based on the traditional active polarization imaging method,the twin neural network method is introduced to solve the problem of accurate selection of the orthogonal polarization intensity map in the experimental process of underwater active polarization imaging with linearly polarized Gaussian beams by using the complementary characteristics of light intensity information and polarization information.Firstly,the theoretical analysis of the deep learning-based polarization imaging method is carried out.Then the polarization imaging method based on twin neural networks is studied and proposed to solve the problem of polarization imaging using metric algorithms.In the neural network-based polarization imaging,for the problem of an accurate selection of orthogonal polarization that exists in the polarization imaging method,image similarity comparison is performed by a twin neural networks,and feature fusion of similar features of the target is performed by using the polarization image fusion method.Finally,the proposed deep-learning polarization imaging method is analytically demonstrated to be able to select orthogonal polarization more accurately and improve the imaging clarity through simulation analysis and optical experiments. |
