基于FHRNet的快速离轴数字全息重建算法
首发时间:2021-03-19
摘要:针对传统离轴数字全息重建技术无法兼顾重建质量和重建速度,该文以经典的深度神经网络模型U-Net为基础,使用非对称卷积获取底层特征信息,使用多组残差卷积提取不同维度的特征信息,使用空洞卷积重建输出图像的分辨率,使用混合损失函数兼顾相似度与峰值信噪比,建立全息图与重建图间的端到端像素级映射关系。经过实验验证,相较于PhUn-Net和Holo-UNet,该结构对实验振幅数据和相位数据都有较好的表现。
For information in English, please click here
Fast off axis digital holographic reconstruction algorithm based on FHRNet
Abstract:In response to the traditional off-axis digital holographic reconstruction techniques that cannot combine reconstruction quality and reconstruction speed, this paper is based on the classical deep neural network model U-Net, using asymmetric convolution to obtain the underlying feature information, using multiple sets of residual convolution to extract feature information of different dimensions, using null convolution to reconstruct the resolution of the output image, using hybrid loss function to combine similarity and peak signal-to-noise ratio, establishing the end-to-end mapping relationship between hologram and The end-to-end pixel-level mapping relationship between the hologram and the reconstructed map is established. After experimental verification, the structure performs better for both experimental amplitude data and phase data compared to PhUn-Net and Holo-UNet.
Keywords: Off-axis digital holography Deep learning U-Net Holographic reconstruction
基金:
引用
No.****
动态公开评议
共计0人参与
勘误表
基于FHRNet的快速离轴数字全息重建算法
评论
全部评论0/1000