Figure - available from: Biomedical Optics Express
This content is subject to copyright. Terms and conditions apply.
(a) shows one quantitative phase image of multiple lung cancer cells. The images are focused manually and then unwrapped by the quality-guided unwrapping algorithm. The unwrapped focused-phase images are used for labeled training in the model. The cross-section and 3D representation of one cell with wrapped and unwrapped signals are shown. (b) show training of model where the UnwrapGAN model consists of a discriminator and a U-net generator. (c) show the results for untrained cells. It tests whether the trained model can generate unwrapped focused-phase images from unseen images that have not been used for training and whether it is possible to recover phase values for other types of cells to evaluate the model generalization. We obtain that the proposed model enables the correction of problems on abrupt phase change. The proposed model’s result is also compared with that of the U-net.
Source publication
Digital holography can provide quantitative phase images related to the morphology and content of biological samples. After the numerical image reconstruction, the phase values are limited between −π and π; thus, discontinuity may occur due to the modulo 2π operation. We propose a new deep learning model that can automatically reconstruct unwrapped...
Similar publications
In this paper, we analyze and evaluate suitable preconditioning techniques to improve the performance of the $L^p$-norm phase unwrapping method. We consider five preconditioning techniques commonly found in the literature, and analyze their performance with different sizes of wrapped-phase maps. Keywords.- Phase unwrapping, $L^p$-norm based method,...
The fisheye camera has a field of view (FOV) of over 180°, which has advantages in the fields of medicine and precision measurement. Ordinary pinhole models have difficulty in fitting the severe barrel distortion of the fisheye camera. Therefore, it is necessary to apply a nonlinear geometric model to model this distortion in measurement applicatio...
The Fourth Industrial Revolution, or Industry 4.0, aims at automating traditional manufacturing and industrial practices using modern smart technology. Autonomous mobile robots equipped with different sensors and employing different techniques have been proposed in the literature in the logistics and inventory warehouse management contexts. Efficie...
Terahertz time-domain spectroscopy (THz-TDS) is a proven technique whereby the complex refractive indices of materials can be obtained without requiring the use of the Kramers-Kronig relations, as phase and amplitude information can be extracted from the measurement. However, manual pre-processing of the data is still required and the material para...
With the development of rail transit in terms of speed and carrying capacity, train safety problems caused by wheel tread defects and wear have become more prominent. The wheel is an important part of the train, and the wear and defects of the wheel tread are directly related to the safety of the train; therefore, wheel tread testing is a key eleme...
Citations
... This can be solved using phase unwrapping algorithms. However, this works well only for a gradual and continuous changing sample surface [15][16][17][18]. Dual-wavelength DH (DWDH) [19][20][21] alleviates the phase wrapping problem by employing measurements at two wavelengths and subtracting the phases of their fields to get the beat phase corresponding to the synthetic wavelength. ...
We demonstrate lensless single-shot dual-wavelength digital holography for high-speed 3D imaging in industrial inspection. Single-shot measurement is realized by combining off-axis digital holography and spatial frequency multiplexing of the two wavelengths on the detector. The system has 9.1 µm lateral resolution and a 50 µm unambiguous depth range. We determine the theoretical accuracy of off-axis dual-wavelength phase reconstruction for the case of shot-noise-limited detection. Experimental results show good agreement with the proposed model. The system is applied to industrial metrology of calibrated test samples and chip manufacturing.
... Since its introduction in 2015, deep learning [19] has rapidly advanced owing to its high computational power, rendering it effective for extracting features from complex multilayered data. Deep learning is widely used in optics for applications such as phase recovery [20][21][22], super-resolution techniques [23][24][25], computer-generated holography [26,27], phase aberration compensation [28], and phase unwarping [29,30]. ...
The speckle noise generated during digital holographic interferometry (DHI) is unavoidable and difficult to eliminate, thus reducing its accuracy. We propose a self-supervised deep-learning speckle denoising method using a cycle-consistent generative adversarial network to mitigate the effect of speckle noise. The proposed method integrates a 4-f optical speckle noise simulation module with a parameter generator. In addition, it uses an unpaired dataset for training to overcome the difficulty in obtaining noise-free images and paired data from experiments. The proposed method was tested on both simulated and experimental data, with results showing a 6.9% performance improvement compared with a conventional method and a 2.6% performance improvement compared with unsupervised deep learning in terms of the peak signal-to-noise ratio. Thus, the proposed method exhibits superior denoising performance and potential for DHI, being particularly suitable for processing large datasets.
... Since its introduction in 2015, deep learning [16] has rapidly advanced owing to its high computational power, rendering it effective for extracting features from complex multilayered data. Deep learning is widely used in optics for applications such as phase recovery [17][18][19], super-resolution techniques [20][21][22], computer-generated holography [23,24], phase aberration compensation [25] and phase unwarping [26,27]. ...
The speckle noise generated during digital holographic interferometry (DHI) is unavoidable and difficult to eliminate, thus reducing its accuracy. We propose a self-supervised deep learning speckle denoising method using a cycle-consistent generative adversarial network to mitigate the effect of speckle noise. The proposed method integrates a 4-f optical speckle noise simulation module with a parameter generator. In addition, it uses an unpaired dataset for training to overcome the difficulty in obtaining noise-free images and paired data from experiments. The proposed method was tested on both simulated and experimental data, with results showing a 6.9% performance improvement compared with a conventional method and a 2.6% performance improvement compared with unsupervised deep learning in terms of the peak signal-to-noise ratio. Thus, the proposed method exhibits superior denoising performance and potential for DHI, being particularly suitable for processing large datasets.
... Among these, convolutional neural networks (CNNs), as representatives of deep learning, can extract various image features through convolution techniques, effectively addressing various computer vision problems [2]. In addition, in the field of optical information processing, deep learning has provided new solutions for fringe analysis [3], phase unwrapping [4], and digital holographic technologies [5], and has achieved significant success [6,7]. Phase unwrapping is a common problem encountered in digital speckle interferometry [8], phase-shifting interferometry [9], and holographic interferometry [10]. ...
Phase unwrapping plays a pivotal role in optics and is a key step in obtaining phase information. Recently, owing to the rapid development of artificial intelligence, a series of deep-learning-based phase-unwrapping methods has garnered considerable attention. Among these, a representative deep-learning model called ${{\rm U}^2}$ U 2 -net has shown potential for various phase-unwrapping applications. This study proposes a ${{\rm U}^2}$ U 2 -net-based phase-unwrapping model to explore the performance differences between the ${{\rm U}^2}$ U 2 -net and U-net. To this end, first, the U-net, ${{\rm U}^2}$ U 2 -net, and ${{\rm U}^2}$ U 2 -net-lite models are trained simultaneously, then their prediction accuracy, noise resistance, generalization capability, and model weight size are compared. The results show that the ${{\rm U}^2}$ U 2 -net model outperformed the U-net model. In particular, the ${{\rm U}^2}$ U 2 -net-lite model achieved the same performance as that of the ${{\rm U}^2}$ U 2 -net model while reducing the model weight size to 6.8% of the original ${{\rm U}^2}$ U 2 -net model, thereby realizing a lightweight model.
... Deep-learning-based hologram reconstruction methods have received widespread attention [27][28][29], among which the Y-Net is a particularly significant class [30]. The Mimo-Net proposed in this paper consists of the Y-Net, a multiscale feature-extraction module, and multiscale reconstruction fusion. ...
Addressing the issue of the simultaneous reconstruction of intensity and phase information in multiscale digital holography, an improved deep-learning model, Mimo-Net, is proposed. For holograms with uneven distribution of useful information, local feature extraction is performed to generate holograms of different scales, branch input training is used to realize multiscale feature learning, and feature information of different receptive fields is obtained. The up-sampling path outputs multiscale intensity and phase information simultaneously through dual channels. The experimental results show that compared to Y-Net, which is a network capable of reconstructing intensity and phase information simultaneously, Mimo-Net can perform intensity and phase reconstruction simultaneously on three different scales of holograms with only one training, improving reconstruction efficiency. The peak signal-to-noise ratio and structural similarity of the Mimo-Net reconstruction for three different scales of intensity and phase information are higher than those of the Y-Net reconstruction, improving the reconstruction performance.
... According to the different use of the neural network, these deep-learning-based phase unwrapping methods can be divided into the following three categories (Fig. 26) 48 . Deep-learning-performed regression method (dRG) estimates the absolute phase directly from the wrapped phase by a neural network (Fig. 26a) [209][210][211][212][213][214][215][216][217][218][219][220][221][222] . Deeplearning-performed wrap count method (dWC) first estimates the wrap count from the wrapped phase by a neural network, and then calculates the absolute phase from the wrapped phase and the estimate wrap count (Fig. 26b) 185,[223][224][225][226][227][228][229][230][231][232][233] . ...
Phase recovery (PR) refers to calculating the phase of the light field from its intensity measurements. As exemplified from quantitative phase imaging and coherent diffraction imaging to adaptive optics, PR is essential for reconstructing the refractive index distribution or topography of an object and correcting the aberration of an imaging system. In recent years, deep learning (DL), often implemented through deep neural networks, has provided unprecedented support for computational imaging, leading to more efficient solutions for various PR problems. In this review, we first briefly introduce conventional methods for PR. Then, we review how DL provides support for PR from the following three stages, namely, pre-processing, in-processing, and post-processing. We also review how DL is used in phase image processing. Finally, we summarize the work in DL for PR and outlook on how to better use DL to improve the reliability and efficiency in PR. Furthermore, we present a live-updating resource (https://github.com/kqwang/phase-recovery) for readers to learn more about PR.
... By adjusting the weights of a convolution kernel in the networks, the model studies the image's features and their relations. This ability means that, nowadays, DL is a State-of-the-Art solution of image classification [11][12][13], segmentation [14][15][16], object detection [17,18], image denoising [19][20][21][22] and PU [23][24][25][26][27] problems. In particular, PU has been solved in two ways: semantic segmentation (SS) and image translation (IT). ...
... Other notable DL-based PU solutions include using a Recurrent Neural Network in [45] and a Generative Adversarial Networks (GANs) [46] in [47] and Pix2Pix GANs in [23]. Yang et al. developed a Deep Image Prior based PU approach in [24], which is a unique DL-based approach, because it does not require a training set. ...
Holographic tomography (HT) is a measurement technique that generates phase images, often containing high noise levels and irregularities. Due to the nature of phase retrieval algorithms within the HT data processing, the phase has to be unwrapped before tomographic reconstruction. Conventional algorithms lack noise robustness, reliability, speed, and possible automation. In order to address these problems, this work proposes a convolutional neural network based pipeline consisting of two steps: denoising and unwrapping. Both steps are carried out under the umbrella of a U-Net architecture; however, unwrapping is aided by introducing Attention Gates (AG) and Residual Blocks (RB) to the architecture. Through the experiments, the proposed pipeline makes possible the phase unwrapping of highly irregular, noisy, and complex experimental phase images captured in HT. This work proposes phase unwrapping carried out by segmentation with a U-Net network, that is aided by a pre-processing denoising step. It also discusses the implementation of the AGs and RBs in an ablation study. What is more, this is the first deep learning based solution that is trained solely on real images acquired with HT.
... In recent years, deep learning (DL) [8] has penetrated various computational imaging industries because of its powerful nonlinear fitting ability. The applicability of deep learning has been demonstrated in optical tomography [9], computer-generated holography [10], digital holography [11], quantitative phase imaging [12], and phase unwrapping [13]. Currently, the mainstream deep learning algorithm is supervised learning, which has superior performance because it is data-driven and resultoriented. ...
Learning-based phase imaging balances high fidelity and speed. However, supervised training requires unmistakable and large-scale datasets, which are often hard or impossible to obtain. Here, we propose an architecture for real-time phase imaging based on physics-enhanced network and equivariance (PEPI). The measurement consistency and equivariant consistency of physical diffraction images are used to optimize the network parameters and invert the process from a single diffraction pattern. In addition, we propose a regularization method based total variation kernel (TV-K) function constraint to output more texture details and high-frequency information. The results show that PEPI can produce the object phase quickly and accurately, and the proposed learning strategy performs closely to the fully supervised method in the evaluation function. Moreover, the PEPI solution can handle high-frequency details better than the fully supervised method. The reconstruction results validate the robustness and generalization ability of the proposed method. Specially, our results show that PEPI leads to considerable performance improvement on the imaging inverse problem, thereby paving the way for high-precision unsupervised phase imaging.
... This method appeared relatively late. In this context, PU could be viewed as a regression problem (He et al., 2019;Wang et al., 2019;Dardikman-Yoffe et al., 2020;Qin et al., 2020;Park et al., 2021;Perera and De Silva, 2021), where the neural network directly learned the mapping between the wrapped phase and the absolute phase. Such a mapping relationship was the most direct Frontiers in Environmental Science frontiersin.org ...
... Although the computational cost of this method was higher, they reported significant benefits. Some studies (Park et al., 2021;Perera and De Silva, 2021) tested the PU performance of long short-term memory (LSTM) networks and Generative Adversarial Networks (GANs), demonstrating their effectiveness. In general, the deep regression method did not fully match the principle of PU when compared with the DLPWC method; however, a more complete network structure could be used to improve accuracy. ...
Synthetic Aperture Radar Interferometry (InSAR) has grown significantly over the past few decades, which were mainly used in remote sensing applications. Most InSAR applications (e.g., terrain mapping and monitoring) utilized a key technique called phase unwrapping Phase unwrapping obtained the absolute phase from the wrapped phase for the subsequent application. However, the collected wrapped phase inevitably contains noise due to the influence of factors such as atmosphere and temperature in the InSAR acquisition stage. This noise made it challenging to obtain the absolute phase from the wrapped phase. This study proposed a deep learning framework (PUnet) for phase unwrapping form InSAR data. pUnet was a robust framework using U-net as the basic structure combined with an attention mechanism and positional encoding, facilitating accurate phase unwrapping from the wrapped phase. Through comparative experiments with typical phase unwrapping algorithms, we demonstrated that pUnet could obtain absolute phases with high accuracy and robustness than from the wrapped phase under various levels of noise.
... In Figure 2C, the Ckpool emphasizes renal tubules and Bowman's capsules, whose values are higher in the FP phase map. However, the strong absorption of the stain provokes phase jumps (Judge and Bryanston-Cross, 1994;Montrésor et al., 2019;Park S. et al., 2021) in the FP map (see the red arrows in Figure 2C, zoomed on the right), which should be corrected by phase unwrapping methods (Hwang W-J et al., 2011). ...
Kidney microscopy is a mainstay in studying the morphological structure, physiology and pathology of kidney tissues, as histology provides important results for a reliable diagnosis. A microscopy modality providing at same time high-resolution images and a wide field of view could be very useful for analyzing the whole architecture and the functioning of the renal tissue. Recently, Fourier Ptychography (FP) has been proofed to yield images of biology samples such as tissues and in vitro cells while providing high resolution and large field of view, thus making it a unique and attractive opportunity for histopathology. Moreover, FP offers tissue imaging with high contrast assuring visualization of small desirable features, although with a stain-free mode that avoids any chemical process in histopathology. Here we report an experimental measuring campaign for creating the first comprehensive and extensive collection of images of kidney tissues captured by this FP microscope. We show that FP microscopy unlocks a new opportunity for the physicians to observe and judge renal tissue slides through the novel FP quantitative phase-contrast microscopy. Phase-contrast images of kidney tissue are analyzed by comparing them with the corresponding renal images taken under a conventional bright-field microscope both for stained and unstained tissue samples of different thicknesses. In depth discussion on the advantages and limitations of this new stain-free microscopy modality is reported, showing its usefulness over the classical light microscopy and opening a potential route for using FP in clinical practice for histopathology of kidney.
