仿生视觉多信息融合水下偏振成像研究

Underwater optical imaging in marine resources exploration, military reconnaissance and underwater marine life-saving and other fields has a very important application. However, in complex underwater scenes, there are some problems of underwater images such as image blurring, reduced contrast and others. The mantis shrimp has a sophisticated structure and a unique imaging mechanism that can simultaneously sense linearly polarized light and circularly polarized light and have a "spectral tuning" effect, greatly improving its underwater target detection capability and adaptability. It opens up new ideas for underwater imaging. This project aims to study on a bionic visual information fusion based underwater polarization imaging method to overcome the scattering efficiency of water, and the effect of suspended particles on the spectrum and light intensity. Firstly, the real-time polarization imaging system combined with the sub-aperture and the dichroic prism is designed for the four-direction linear polarization images and circular polarization and color images with the aberration analysis and system optimization. Secondly, the collected images are preprocessed and image segmentation is performed at the pixel level. The image registration is done based on the SURF feature extraction algorithm. Finally, a polarized, spectral and light intensity image fusion model is established, and to simulate the imaging mechanism of mantis shrimp machine learning method is used for guiding the fusion of multi-information images. This study is of certain significance for the underwater imaging analysis, system design, and so on.

水下光学成像在海洋资源勘探、海洋军事侦查及水下救生等领域具有非常重要的应用。在复杂的水下场景中，水下图像往往存在诸如图像模糊，对比度低等问题，影响成像距离和深度。螳螂虾复眼具有精密的结构和独特的成像机制，可以同时感知线偏振光和圆偏振光，并具有“光谱调谐”效果，大大提高其水下目标探测能力和适应能力，为水下光学成像研究拓宽了思路。本项目旨在研究一种基于仿螳螂虾复眼的偏振成像方法。首先，仿螳螂虾复眼结构，研究分孔径与分光棱镜相结合的实时偏振成像系统，同时提取4个方向线偏振图像及圆偏振和彩色图像，完成像差分析及系统优化；其次，对采集的图像进行预处理，并在像素水平上进行图像分割，基于SURF特征提取算法完成图像的配准。最后建立偏振、光谱、光强特征图像融合模型，以机器学习模拟螳螂虾成像机制，指导多信息图像融合。项目的研究对水下目标成像分析、水下成像系统设计等方面都具一定参考价值。

螳螂虾复眼具有精密的结构和独特的成像机制，使其具有强大的水下目标探测能力。本项目围绕仿螳螂虾复眼结构及其偏振感知视觉进行研究。首先，根据仿生结构要求，提出一种分孔径式多通道阵列同时偏振成像系统，该系统前置共孔径组和后置四个离轴偏心子孔径系统组成，结构紧凑，可实现对水下目标进行实时偏振图像的提取，并对光学系统的像差、公差进行分析，所设计的光学系统能够满足使用要求。其次，完成采集系统的硬件设计与实验平台的搭建工作，在自主研制的图像采集平台上完成多通道图像采集，并将图像进行灰度化处理，基于形态学图像处理原理提取各探测器上探测到的各图像边缘信息，进行图像分割，基于仿射变换模型实现图像配准等图像预处理。而后，基于Stokes矩阵对偏振图像进行分析处理，提取偏振度信息，对强度图像采用改进的K-means算法与暗通道先验算法相结合的方法进行去雾处理，并采用加权策略对强度、颜色、偏振图像信息进行融合，实现了水下图像对比度增强的效果。通过本项目开展，搭建了水下分孔径多通道图像采集平台，可用于多种水下图像信息实时采集；提供了一种偏振、颜色、强度信息融合的方法，对水下目标成像分析研究具有一定参考价值。