基于序列深度显微图像的非织造滤材三维结构重建

Nonwoven is a filtering material essential for many industrial and medical applications. The filtering performance of nonwoven is mainly dictated by its structure. However, fibers in a nonwoven structure are spatially entangled, twisted and overlapped, forming an extremely complex porous assembly which makes it difficult to express and measure three-dimensional (3D) porosity. This project will utilize the sequential microscopic nonwoven images captured at the same (x, y) position but varying depth positions to detect well-focused surface pixels at all registered depths, and then to reconstruct the 3D model of the nonwoven revealing its “antpile” like porous structure for porosity analysis. We will propose a new “fiber-slicing” algorithm for 3D reconstruction, which takes four major steps: (1) image fusion to generate a fully focused image from the microscopic image sequence; (2) image segmentation to separate fibers in the fused image; (3) One-to-Many projection to generate data clouds by recovering pixels’ depth information; and (4) fiber particles to reconstruct a 3D surface image (model) of all data clouds. The 3D model will precisely represent the complete structure—both fibers and holes—of a nonwoven, and enable the comprehensive analysis of nonwoven’s porosity and filtering performance. The expected outcomes of this research will lay a theoretical foundation for high-fidelity reconstruction of 3D nonwoven structures, and provide a new approach for structural analysis of complex porous materials, and supply data support for better understanding of how a new air-filtering nonwoven captures and retains harmful particles.

非织造过滤材料的结构决定其性能，然而材料中纤维相互缠绕、层叠形成的三维孔隙非常复杂，难以准确表征。本项目利用同视野序列深度光学显微图像，采用表层可探测区真实三维重建与内部模型重建相结合的方法，探索非织造过滤材料完整“蚁穴”孔隙三维重建解决方案。项目提出了纤维“切片”再重组的构思：将多层序列图像融合成二维，对纤维目标进行切片后再投影回各自的深度，得到纤维中轴线完成纤维三维重建。研究将从融合图像纤维分割、One-to-Many投影、纤维重建模型、过滤性能分析等方面展开，重点解决细长弯曲纤维不同位置在空间的唯一深度测算，以及纤维质点生长模型的数值解等问题，拟形成直接可用的非织造过滤材料三维孔隙完整结构进行滤材过滤性能分析。预期研究成果将为非织造材料三维结构真实重建奠定理论基础，为多孔隙复杂材料的结构分析提供新思路，为有害颗粒物精确拦截，空气过滤材料设计、生产、使用等应用领域提供数据支持。

本项目利用全自动光学显微镜获取的多聚焦序列图像为基础对非织造过滤材料进行三维重建。采用全自动显微镜、工业相机与多焦面序列图像采集软件，建立了一套显微序列图像采集系统。在此基础上通过清晰度判定与聚焦测距技术(DFF)，获取序列图像上相应的清晰聚焦纤维，并提取其对应深度信息。随着项目研究的深入，将这一思路进一步优化，引入了卷积神经网络来进行对序列图像上清晰纤维进行提取。由于纤维材料与卷积神经网络常见的识别场景大有不同，为了获得对于非织造纤维更好的适应能力，制作了由非织造纤维构成的训练集来对网络进行训练，并不断改进网络架构，由此实现了更为精准的三位重建模型。.本研究对非织造材料的三维重建进行了研究，并且结合卷积神经网络技术与聚集测距技术（DFF）得到了较为不错的结果，且最终测得的纤网结构信息准确，为有害物的拦截、滤材的制备以及对过滤机理的研究都有非常重要的意义。

内容获取失败，请点击重试