학술논문
基于傅里叶成像网络的全息粒子成像技术研究 / A Study on Holographic Particle Imaging Based on Fourier Imaging Networks
Document Type
Academic Journal
Author
Source
影像科学与光化学 / Imaging Science and Photochemistry. 42(1):1-8
Subject
Language
Chinese
ISSN
1674-0475
Abstract
数字全息图重建可以获得物体的三维信息,在医疗、环保、化工等需要测量颗粒大小和三维信息的相关研究有着重要的科学意义和使用价值.粒子全息定位通常有计算复杂和重建精度不足等问题,为了解决相关问题,我们采用一种基于傅里叶成像网络的方法,利用傅里叶变换的特性,能够有效提取全息图的频率特征,结合深度学习技术,通过可学习的滤波器和全局感受野处理输入数据的空间频率,实现对全息图快速和准确的处理,不仅提高了数据处理效率,还具备了更强大的特征提取能力,从而实现对粒子的精确定位.本文通过实验和仿真两方面验证,并与Dense_U_net网络对比.结果显示,基于傅里叶成像网络的方法在定位精度和速度上都有显著的提升.并通过实验验证了该方法可行性.
Digital holographic reconstruction can obtain three-dimensional information of objects,and it has significant scientific significance and practical value in medical,environmental protection,chemical engineering,and other fields that require measurement of particle size and three-dimensional information.The particle holographic localization often faces issues of computational complexity and insufficient reconstruction accuracy.To address these issues,we adopted a method based on the Fourier imaging network.By leveraging the characteristics of the Fourier transform,we were able to effectively extract frequency features from the hologram.Combined with deep learning technology,the network utilized learnable filters and a global receptive field to process the spatial frequencies of input data.This approach not only enhanced data processing efficiency but also endowed the system with superior feature extraction capabilities,thereby achieving precise particle localization.In this paper,we validated our approach through both experimental and simulation studies and compared it with the Dense_U_net network.The results show that the method based on the Fourier Imaging Network significantly improved both localization accuracy and speed.
Digital holographic reconstruction can obtain three-dimensional information of objects,and it has significant scientific significance and practical value in medical,environmental protection,chemical engineering,and other fields that require measurement of particle size and three-dimensional information.The particle holographic localization often faces issues of computational complexity and insufficient reconstruction accuracy.To address these issues,we adopted a method based on the Fourier imaging network.By leveraging the characteristics of the Fourier transform,we were able to effectively extract frequency features from the hologram.Combined with deep learning technology,the network utilized learnable filters and a global receptive field to process the spatial frequencies of input data.This approach not only enhanced data processing efficiency but also endowed the system with superior feature extraction capabilities,thereby achieving precise particle localization.In this paper,we validated our approach through both experimental and simulation studies and compared it with the Dense_U_net network.The results show that the method based on the Fourier Imaging Network significantly improved both localization accuracy and speed.