基于仿生视觉感知建模的水下构筑物裂缝检测方法研究

Cracks not only generally exist in the surface of underwater hydraulic concrete structures such as the dams and piers, but also extend to the inner. Timely and accurate crack detection is an important technology for routine maintenance, repairing, strengthening of underwater structures and guarantee for its safe operation. Due to the complexity of underwater environment and the randomness of cracks presentation, the measured signals about cracks are easily overwhelmed by noise, and the crack information is scarce and hard to be characterized, which will lead to low recognition rate and high false alarm rate based on existing methods. Therefore, inspired by the biological mechanisms of biological vision such as effective compression fusion mechanism, super-sensitive visual perception, accurate inference and cognition, a new crack detection method for underwater structures in complex environment is proposed, which includes establishing a lower order bionic model of crack detection, designing a novel detection algorithm, and evaluating the detection performance. This study is focused on solving the questions as follow. 1) Modeling the mapping of multi-source heterogeneous sensors based on mimicking compound eye mechanism; 2) The detection information fusion based on mechanism of lamina cassette; 3) The crack feature characterization inspired by the information processing mechanism of vision neurons. This study will enrich and develop the unstructured weak target detection theories and methods in complex environment, and it will also provide a theoretical basis and practice guidance for the application of this method.

裂缝普遍存在于大坝、桥墩等水下混凝土构筑物表面，且向内部延伸。对裂缝及时准确的检测，是水下构筑物日常维护、修复加固和保障其安全运行的必要技术途径。因水下环境的复杂性及裂缝呈现的随机性，裂缝量测信号极易被噪声淹没，裂缝信息匮乏且难以表征，致使现有检测方法识别率低、虚警率高。为此，借鉴生物视觉高效压缩融合、超敏锐度感知及准确推理认知未知世界的生物学机理，将声、光成像传感器融合映射到一个虚拟的仿复眼平台上，建立仿生机制的“低阶”裂缝检测模型，设计检测算法，进行检测性能评估，进而构建一套复杂水环境下构筑物裂缝检测方法。重点解决（1）仿复眼机理的多源异构传感器映射建模；（2）仿“薄板”暗盒机理的探测信息融合；（3）仿视觉神经元信息处理机制的裂缝特征表征。本研究将丰富和发展复杂环境下非结构化弱目标检测理论和方法，并为该方法的应用提供理论依据和实践指导。

本项目针对大坝、桥墩等水下混凝土构筑物表面裂缝的检测问题进行研究，面对水下环境的复杂性及裂缝呈现的随机性，借鉴生物视觉高效压缩融合、超敏锐度感知及准确推理认知未知世界的生物学机理，将声、光成像传感器融合映射到一个虚拟的仿复眼平台上，建立仿生机制的“低阶”裂缝检测模型，设计检测算法，构建了一套复杂水环境下构筑物裂缝检测方法。. 完成了声-光异构仿复眼的水下构筑物裂缝检测流程和载体设计，实现了仿生复眼视觉信息处理模式工程化模拟；构建了仿复眼机理的多源异构传感器映射模型，设计出LF系统和SF系统协同工作的检测模型；仿复眼视觉系统信息处理机理，围绕水下大坝裂缝检测提出了多种大坝裂缝检测算法；针对多个“小眼”图像信息的相关性及互补性关系，提出了基于模糊证据的图像融合方法；仿复眼视觉系统神经通路，提出水下构筑物裂缝图像背景抑制和目标增强算法；最后提出了仿视觉神经元信息处理机制的裂缝特征表征算法。. 本项目通过构建一套复杂水环境下构筑物裂缝检测方法，大大提高了检测率和识别率、降低了虚警率和漏检率。项目组成员在国内外期刊和会议发表论文32篇，其中SCI检索论文14篇，EI检索论文11篇。参加国际学术会议4次，申请专利15项，授权8项，申请软件著作权3项，撰写专著1部（成稿待刊）。本研究成果将丰富和发展复杂环境下非结构化弱目标检测理论和方法，为该方法的应用提供理论依据和实践指导，并有望用于水下构筑物日常维护、修复加固和健康诊断过程。