Hao Fugui, Tang Chen, Xu Min, Lei Zhenkun
Appl Opt. 2019 May 1;58(13):3338-3346. doi: 10.1364/AO.58.003338.
The denoising of electronic speckle pattern interferometry (ESPI) fringe patterns is a key step in the application of ESPI. In this paper, we propose a method for batch denoising of ESPI fringe patterns based on a convolution neural network (CNN). In the proposed method, the network is first trained by our training dataset, which consists of the noisy ESPI fringe patterns and the corresponding noise-free images. We propose a new computer-simulated method of ESPI fringe patterns to create our training dataset. After training, the other multi-frame ESPI fringe patterns to be processed are fed to the trained network simultaneously, and the corresponding denoising images can be obtained in batches. We demonstrate the performance of the proposed method via application to 50 computer-simulated ESPI fringe patterns and three groups of experimentally obtained ESPI fringe patterns. The experimental results show that our method can obtain desired results even when the quality of ESPI fringe images is considerably low because of variable density, high noise, and low contrast, and our method can denoise multi-frame fringe patterns simultaneously. Moreover, we use the computer-simulated ESPI fringe patterns to train the network; after training, the trained network can be used to denoise either computer-simulated ESPI fringe patterns or the experimentally obtained ESPI fringe patterns. The proposed method is especially suitable for processing a large number of ESPI fringe patterns.
电子散斑干涉（ESPI）条纹图的去噪是ESPI应用中的关键步骤。本文提出了一种基于卷积神经网络（CNN）的ESPI条纹图批量去噪方法。在所提方法中，网络首先由我们的训练数据集进行训练，该数据集由含噪声的ESPI条纹图和相应的无噪声图像组成。我们提出了一种新的ESPI条纹图计算机模拟方法来创建我们的训练数据集。训练后，将其他待处理的多帧ESPI条纹图同时输入到训练好的网络中，即可批量获得相应的去噪图像。我们通过将所提方法应用于50个计算机模拟的ESPI条纹图和三组实验获得的ESPI条纹图来展示其性能。实验结果表明，即使ESPI条纹图像由于密度变化、噪声高和对比度低而质量相当低时，我们的方法也能获得理想的结果，并且我们的方法可以同时对多帧条纹图进行去噪。此外，我们使用计算机模拟的ESPI条纹图来训练网络；训练后，训练好的网络可用于对计算机模拟的ESPI条纹图或实验获得的ESPI条纹图进行去噪。所提方法特别适合处理大量的ESPI条纹图。
