基于光学和声学信息融合实现水下目标的智能识别与跟踪的技术研究和应用

Undersea equipment plays an important role in national defense, economy and other fields. The research and development of 'Jiao long' and other important underwater equipment and their successful deep diving, mark our country's significant breakthrough in the field of ocean. Undersea equipment has achieved great success, but there is still broad space for development in reliability, intellectualization and so on. The realization of steady and reliable recognition and tracking of specific object, is an important step in the completion of underwater robots' reliability and intellectualization. Compared to the land environment, the recognition and tracking of underwater targets need to deal with many special problems, such as non-uniform illumination condition, disturbance caused by carrying platforms' unstable pose. In view of the limitations of current methods, this project proposed to use biological cognitivemechanism as a framework, combine with proper mathematical methods and fuse optical image with acoustic information to realize the identification and tracking of underwater targets. We expect that the related research can effectively promote the development of robottechnology in ocean, in the space as well as on the land. The success of this project can play an important role in the basic research on our country's ocean and space robot.

海洋水下装备对国防建设和国民经济都有重要的作用，蛟龙号等水下装备的研发和深潜成功，是我国在海洋领域的重要突破。目前，深海装备虽然取得了巨大的成就，但在可靠性和智能性等方面还有很大的发展空间。对水下目标的智能识别和跟踪，是实现水下装备可靠性和智能化的重要一步，但水下目标识别和跟踪的特殊性（如非均匀光照条件、搭载平台不稳定带来的扰动等）使得一般的模式识别方法难以直接应用。 本项目拟（1）通过信息-数学方法和（2）探索计算神经方法（借鉴人的认知神经机制），实现水下声-光信号的融合，力图获得水下目标的可靠、稳定和可学习的识别和跟踪，并通过申请团队三方（中科院应用数学所、哈工程国防科工局重点实验室和中船重工集团七〇二所）的紧密合作，将部分研究成果应用于深海装备。同时，项目的成功将促进陆地机器人和水下机器人的识别、跟踪技术的相互借鉴，逐步形成海陆空机器人识别

海洋水下装备对我国国防建设和国民经济都有重要的作用，而对水下目标的智能识别和跟踪是实现水下装备可靠性和智能化的重要一步，也是极具挑战性的重大难题。本项目通过信息-数学方法实现水下声-光信号的融合，力图获得水下目标的可靠、稳定和可学习的识别和跟踪。具体研究内容包括：1）有效抑制声呐和极化合成孔径雷达图像斑点噪声的新方法；2） 弹性散射目标的声波近场成像算法；3） 基于流形学习和稀疏编码的声呐和SAR目标识别算法；4） 基于神经网络和认知模型的声纳和SAR目标分类与识别算法；5） 基于图正则化稀疏编码和稀疏正则化的光学图像分类算法；6）基于选取近似凸包顶点和支持向量机在线分类算法；7）基于张量表示和多流形学习的SAR图像和声纳目标分类与识别算法；8） 自适应脉冲耦合神经网络的水下激光图像分割方法。