Transmitted wavefront testing of window glass with confocal photon-sieves radial-shearing interferometry based on deep learning

Interference wavefront detection technology has been widely used due to its non-contact, high sensitivity and high precision. Among many wavefront detection techniques, radial shearing interferometry is one of the most effective techniques because it does not need a standard reference beam, and it is easy to be designed as a common optical path without information loss and little affected by system error. However, the current radial shear interference technology still has the disadvantages that the optical path structure is not simple enough and cannot be applied to large-aperture optical systems. Focusing on the main problems of shear interference technology, we propose a radial shear interference technology based on a pair of photon sieves. A system consisting of two photon sieves can achieve radial shear interference, which greatly simplifies the optical setup and improves the stability of the optical system. In addition, the photon sieve has low manufacturing difficulty, flexible design, low cost and can meet the detection requirements of different optical systems. Therefore, the application of photon sieve in radial shear interference technology has great research significance.

As an excellent holographic information carrier, metasurface holograms possess the advantages of subwavelength footprint, leading to large field of view and high resolution. Meanwhile, intelligent algorithms should be developed to fully exploit the flexible modulation properties provided by metasurfaces. Here, two amplitude holograms with quantitatively correlation are realized using the photon sieve as the amplitude modulation device. A new iterative algorithm is developed to obtain the amplitude profiles of the two holograms with correlation, which can make the two profiles mutually connect by simply adding/erasing dynamic pixels. Two photon sieves composed of nanoholes arrays are fabricated to demonstrate the feasibility of the scheme. This amplitude holography method can operate throughout the entire visible and near‐infrared wavelength regions with polarization independence, and it is expected to achieve switchable metasurface holography with the development of materials science. Such scheme promises to be applied in compact data storage, dual pattern recognition, optical encryption, and so on.

The large slope of the aspheric departure presents great difficulty for optical testing researchers to test aspheric surfaces and wavefronts. The instrument dynamic range of traditional interferometers does not support the high number of fringes due to the steep slopes commonly found in aspherics. The radial shearing interferometer, which can greatly reduce the slope of the aspheric wavefront under test, is always adopted to test aspherics. In this paper, the two evaluating parameters of a radial shearing interferometer, the wavefront slope tolerance and the wave phase sensitivity, are proposed. The effective radial shear plays important role in reduction of the wavefront slope and that of the wave phase sensitivity. How the two parameters effect the radial shearing interferometer is discussed in detail. Computer simulation shows that by properly choosing the effective radial shear, the radial shearing interferometer can obtain the best performance when testing an arbitrary aspheric wavefront. This research is of greatly importance to the design of the radial shearing interferometer for aspheric testing.

Optical metrology for elements with large dynamic range is very important, especially in aspheric components testing field. Cyclic radial shearing interferometer(CRSI) with a small radial shearing ratio of the expanded wavefront's radii to the contracted version's radii can obtain a small magnitude optical path difference(OPD) because the expanded wavefront and its contracted version nearly have the same shape and magnitude, especially for an optical component to be tested with rotationally symmetric surface. The number of fringe pattern can be decreased dramatically and recorded by CCD. This paper demonstrates the feasibility that CRSI could be used to measure an aspheric surface with large dynamic range and large aperture.

Multidisciplinary single-cell optical cytometry.

The approximate Strehl Ratio expressions that need only the wave phase distribution is not appropriate in the laser fusion system when the phase has great aberration. The accurate Strehl Ratio value need both the phase and the amplitude distribution information. The paper provide a method of reconstructing wave phase and amplitude distribution by a single radial shearing interferogram. The radial shearing interferometer based on spatial phase modulation is simple and stable, needs only one interferogram. The principle and structure of cyclic radial shearing interferometer are briefly discussed and the process of spatial phase modulation, Fourier transform method and wavefront reconstruction are analyzed. It also illluminates the physical meaning and establish the mathematical model. The process of getting SR value is simulated by computer. The simulation result indicates that the RMS error of the reconstructing wavefront can achieve λ/100.

Artificial Intelligence in the American Healthcare Industry: Looking Forward to 2030

An automated method for detecting and classifying submicron surface defects on mirrors used in high-precision optical systems has been developed, utilizing interferometric image analysis and deep learning. This approach replaces manual inspection with a neural network, delivering faster and more objective defect diagnostics, eliminating human bias, and accelerating quality control in serial production of optical components. A synthetic dataset of interferometric images was generated to train the model, simulating scratch-type defects on mirror surfaces through specialized software based on the Linnik interferometer model. The dataset encompasses three surface classes: flat surfaces, single scratches, and multiple scratches. The neural network is built upon MobileNetV2, pre-trained on ImageNet, with fine-tuning of its final blocks to adapt to the task’s specifics. The architecture incorporates GlobalAveragePooling2D for feature compression, Dense layers with ReLU activation and BatchNormalization, Dropout to mitigate overfitting, and a Softmax output layer for classifying the three categories. Data augmentation and soft voting techniques were employed to enhance the model’s generalization ability. Classification accuracy, assessed using the accuracy metric, achieves 96% on the synthetic validation set and 82.7% on real images acquired from a Linnik interferometer. The highest accuracy is observed for flat surfaces, while the lowest occurs for multiple scratches, highlighting challenges posed by real-world conditions such as noise and artifacts. The method proves its practical value for automated diagnostics, with future enhancements tied to improving synthetic data realism-potentially by incorporating modeled noise-and extending the model’s adaptability to additional defect types like indentations or protrusions, thereby broadening its applicability in optical system manufacturing

A radial shearing interferometer with two equispaced circular gratings is described for obtaining a constant radial displacement of the test wavefront. The procedure for the fringe analysis is shown by applying the Fourier transform method to the circular fringe pattern for precisely obtaining the differentiation of the test wavefront. A novel method is also shown for holographically making the equispaced circular gratings by using a cylindrical mirror and a collimating lens. It is shown that the deflection and misalignment of the gratings will not appear in the interferogram by using the holographic production method.

Image denoising is a vital step in many image processing and computer vision professions that aims to improve picture quality by decreasing noise while maintaining important image information. In this article, a detailed overview is presented of traditional and deep learning based picture denoising approaches. Classical methods, such as linear filtering, transform domain techniques, and patch-based approaches like Non-Local Means (NLM), are commonly employed because they are simple and effective in removing Gaussian noise. However, these approaches frequently struggle with complicated noise patterns and blur fine features. Recent advances in deep learning, including Convolutional Neural Networks (CNNs) and other designs such as Generative Adversarial Networks (GANs) and autoencoders residual networks, have considerably improved picture denoising performance. These data-driven techniques excel in learning complicated noise patterns from big datasets, providing superior generalization across various noise types, including non-Gaussian noise, and dealing with a larger range of image degradation conditions.

A wavefront reconstruction technique which obtains wavefront from a radial shearing interferometer with high precision based on two-dimensional windowed Fourier transform (WFT) is presented. First, the radial shearing fringe is processed by WFT, and by inverse WFT after setting threshold and frequency range of integration. Then wavefront phase difference distribution is obtained by the treatment of removal-carrier frequency and phase unwrapping. Finally, the actual wavefront can be reconstructed from phase difference by wavefront iterative algorithm. The simulations show that the maximum phase reconstruction error is 0.82%, RMS is 0.020 9 rad. Also corresponding experiments are conducted for comparison. In addition, window function selection, window size and threshold setting are briefly discussed. Compared with the traditional Fourier transform method (FFT), shearing interferometer wavefront measurement based on windowed Fourier transform has higher precision and stability, which provides a new processing method for wavefront measurement.

This research addresses the challenge of camera calibration and distortion parameter prediction from a single image using deep learning models. The main contributions of this work are: (1) demonstrating that a deep learning model, trained on a mix of real and synthetic images, can accurately predict camera and lens parameters from a single image, and (2) developing a comprehensive synthetic dataset using the AILiveSim simulation platform. This dataset includes variations in focal length and lens distortion parameters, providing a robust foundation for model training and testing. The training process predominantly relied on these synthetic images, complemented by a small subset of real images, to explore how well models trained on synthetic data can perform calibration tasks on real-world images. Traditional calibration methods require multiple images of a calibration object from various orientations, which is often not feasible due to the lack of such images in publicly available datasets. A deep learning network based on the ResNet architecture was trained on this synthetic dataset to predict camera calibration parameters following the Brown-Conrady lens model. The ResNet architecture, adapted for regression tasks, is capable of predicting continuous values essential for accurate camera calibration in applications such as autonomous driving, robotics, and augmented reality.

Electrical resistance tomography (ERT) is an efficient technology for rapid, accurate, and real-time monitoring of the dynamic of industrial process. However, due to the inherent nonlinearity and ill-posed, image reconstruction of ERT remains a challenging problem of significant importance for industrial visualization. A novel Landweber iterative reconstruction network (LIRN) that combines the mathematical structure of Landweber iterative reconstruction method with deep learning is proposed. As an iterative algorithm, the proposed method solves the problem of parameters selection, and as a deep learning method, LIRN has less dependence on incomplete databases and better generalization capabilities compared with the black box models. Each layer of the LIRN is composed of three parts: fully connected subnet, convolution subnet, and the output of the former layer. The output of the former layer is mapped to the outputs of the two subnets, which constitutes the residual module. The fully connected subnet is used to represent the gradient of the data fidelity, in which the relaxation factor vector is a learnable parameter corresponding to the weights of the subnet. The convolution subnet as a self-learning regularizer learns the image prior information. During the training process of LIRN, the relaxation factor vector and the image prior information are jointly trained and learned. The experimental and analytical results demonstrate that the proposed image reconstruction network with four layers can achieve a better reconstruction distribution than the traditional imaging methods and the existing image reconstruction algorithms based on deep learning.

It has generally been assumed that a radial shearing interferometer cannot be used to test a mirror with a central hole. We show that this can be done by a proper choice of the shear ratio and the use of digital techniques of measurement and analysis.

Hand gesture recognition has great applications in human-computer interaction (HCI), human-robot interaction (HRI), and supporting the deaf and mute. To build a hand gesture recognition model using deep learning (DL) with high results then needs to be trained on many data and in many different conditions and contexts. In this paper, we publish the TQU-HG dataset of large RGB images with low resolution (640×480) pixels, low light conditions, and fast speed (16 fps). TQU-HG dataset includes 60,000 images collected from 20 people (10 male, 10 female) with 15 gestures of both left and right hands. A comparative study with two branches: i) based on Mediapipe TML and ii) Based on convolutional neural networks (CNNs) (you only look once (YOLO); YOLOv5, YOLOv6, YOLOv7, YOLOv8, YOLO-Nas, single shot multiBox detector (SSD) VGG16, residual network (ResNet)18, ResNext50, ResNet152, ResNext50, MobileNet V3 small, and MobileNet V3 large), the architecture and operation of CNNs models are also introduced in detail. We especially fine-tune the model and evaluate it on TQU-HG and HaGRID datasets. The quantitative results of the training and testing are presented (F1-score of YOLOv8, YOLO-Nas, MobileNet V3 small, ResNet50 is 98.99%, 98.98%, 99.27%, 99.36%, respectively on the TQU-HG dataset and is 99.21%, 99.37%, 99.36%, 86.4%, 98.3%, respectively on the HaGRID dataset). The computation time of YOLOv8 is 6.19 fps on the CPU and 18.28 fps on the GPU.