仿复眼视觉系统光谱-偏振敏感机制的水下目标检测方法

Underwater imaging detection technology is very important for the underwater engineering and operating maintenance, which underlies the tasks as the underwater structures defect detection, high slope piping and turbulence flow protection. The up-to-date methods only have the ability to cater for the static and clear water. However, in the high-turbid and complicated underwater engineering environments it is critical to represent and detect objects, due to the fuzzy, distorted and nondeterministic object information. This nature results in the low object detection accuracy. Inspired by the visual system of the compound eye in mantis shrimps, the spectrum-polarization sensitivity mechanism is imitated to propose a new bionic underwater object detection method. This method is established by combining the bionic spectral-specialized polarization imaging and virtual view information multilevel fusion processing methanisms, it is able to be applied in the high-turbid flowing waters. The advantage of the spectrum-polarization sensitivity in the visual system of the mantis shrimps is transformed and explored. The research specially focuses on the bionic compound eye establishment method to overcome the intrinsic difficulties in underwater object information acquisition and modeling. A novel breakthrough would be made for the critical technology in underwater polarization imaging and detecting, establishing the basement for applications.

在现代水下工程作业和运行维护中，对于水下构筑物缺陷及损毁探测、高边坡管涌及工程病害防护等任务均迫切需要水下成像探测技术的支持。然而目前现有方法仅能部分实现静水流况和清洁水体中的目标检测。而在水工作业和水下探测的高浑浊流动水体条件下，由于水下光学环境的复杂性，目标信息呈模糊、畸变和不确定等特性，使得精确表征并检测目标属性及特征具有很大难度，导致水下目标检测正确率过低。受水下生物螳螂虾复眼视觉系统光谱-偏振敏感机制的启发，针对上述水下复杂光学环境，拟提出仿螳螂虾复眼视觉系统特定光谱段偏振成像与虚拟视像信息融合处理联合的水下目标检测方法；探索并转化复眼系统光谱-偏振敏感性和追踪目标的优势，尤其重点研究适宜水下复杂光学环境下基于仿复眼机制的目标表征建模方法，如何克服难以准确提取和检测目标特征的本质性困难。通过本项研究，拟在水下偏振成像及目标检测等关键理论方法上取得创新性突破，为实际应用打下基础。

水下光学成像检测是构筑物探测、资源开发以及水权益保护等需求中用于发现兴趣目标的一种关键技术手段。目前所采用的多是面向一般光学环境仅依赖光谱、光强信息实现目标检测方法的推广或优化，仅能部分实现少数清洁水体静态光学环境中的目标表征及检测。而在常见高浑浊水体中，面对高衰减、强散射的水下光学环境，加之目标姿态、尺度，光照变化等动态因素的影响，目标的光谱、光强信息常出现模糊、畸变和随机等特性，使得精确表征并检测目标具有较大的技术难度。鉴于此，本项目提出仿复眼视觉系统特定光谱段偏振成像与视觉神经信息处理联合的水下目标检测方法，通过建模分析并转化了螳螂虾复眼视觉系统的光谱-偏振信息获取和“电脉冲”序列特征神经计算在水下目标感知上的优势，实现了一种新颖并能够克服水下复杂光学环境并准确表征水下目标不变性特征并检测的方法，突破复杂水下光学环境中难以可靠、准确地获取目标特征并检测的瓶颈。项目成果研发了水下多通道偏振成像关键技术设备及一系列水下目标检测算法模型，通过实验验证将水下目标检测的准确率提高至80%以上。