This PDF file contains the front matter associated with SPIE Proceedings Volume 12277, including the Title Page, Copyright information, Table of Contents, and Conference Committee listings.

We previously proposed a grating-assisted phase-shifting method for incoherent digital holography (IDH). This method leverages the diffraction of a binary phase grating, making it possible to create four self-interference holograms with different phase shifts in parallel. By simultaneously capturing the holograms with four image sensors, implementing a four-step phase-shifting algorithm and detecting complex amplitude distribution at a single exposure are possible. In this study, we investigated the effect of the axial misalignment of image sensors on the reconstructed images in the grating-assisted phase-shifting method via numerical simulations. The numerical simulations are based on the scalar diffraction theory and allows us to emulate IDH systems. The numerical results provide basic insight and a guideline to design an optical setup for the method within an acceptable tolerance.

Optical scanning holography (OSH) has been researched as one of single pixel holography. Although incoherent holograms can be obtained with a bucket detector, OSH requires sequentially scanning an illumination pattern on an object with a moving stage and a phase shifter. This fact imposes disadvantages of unstable and bulky setup on OSH. To solve the problem, we have proposed the OSH technique without mechanical scanning by using a polarization property of a spatial light modulator. The proposed technique makes its setup inline and simple due to absence of moving parts. In this presentation, a basic principle of the proposed technique and its improvement based on parallel phase shifting and compressive sensing are introduced.

The wavefront coding technology can improve the depth of focus of traditional optical systems. Thus, the influence of thermal defocus on the imaging quality of the infrared system canbe effectively reduced when applied the wavefront coding technology to the athermalization design process. Till now, the analysis of the athermalization wavefront coding system is all under the uniform temperature distribution condition. Aiming at the environment with uneven temperature and gradient distribution, this paper used the temperature gradient polynomial to simulate the temperature distribution model, and calculated the phase function of the typical phase mask. Modulation transfer function (MTF) and point spread function (PSF) are calculated by simulation methods. And analyzing the effects of different temperature gradient values and different temperature gradient orders by Hilbert angle, MTF area and Fisher information. The simulation results show that, the typical wavefront coding phase mask can work normal under a certain range of temperature gradient. When the simulation had been done at the thickness of the phase mask is 4mm and radius is 6mm, the cubic phase mask can still work properly in the gradient range [−10^oC/mm, 10^oC/mm . And the larger gradient order, the the greater the influence on the imaging of wavefront coding system.

A reference light multiplexing method based on computer-generated hologram is proposed to increase spatial bandwidth and information capacity. When the multiplexing computer-generated hologram is illuminated by different reference light waves, different objects can be reconstructed. In addition, Gerchberg-Saxton algorithm is utilized to optimize the multiplexing computer-generated hologram. Numerical simulations and optical experiments are performed to demonstrate the effectiveness of proposed method. It is expected that proposed method could be widely applied in holographic display in the future.

We proposed optical see-through and reflective see-through augmented reality (AR) displays based on integral imaging. For the optical see-through AR display, we designed and fabricated a lens array holographic optical element and a multiplexed lens array holographic optical element and developed a high-resolution optical see-through three-dimensional (3D) display and a dual-view optical see-through 3D display system, respectively. For the reflective see-through AR display, we designed a mirror-based pinhole array and reflective polarizer as an image combiner and then combined them with the conventional integral imaging display unit. The developed prototypes have potential application in vehicle, navigation, and stomatology.

Bandwidth constraint optimization method is proposed to design phase-only holograms for speckle-suppressed and ringing artifacts-free computational holographic projection. The bandwidth properties of the reconstructed wave field are analyzed, which helps to effectively sample and describe the reconstructed intensity distribution based on the sampling theorem. Iterative calculation with bandwidth constraint strategy and quadratic initial phase can optimize phase-only holograms without vortex-stagnation problem. The proposed method can effectively suppress the speckle noise as well as the ringing artifacts. Numerical and optical experiments have been performed to validate the feasibility of the proposed method for achieving high-quality holographic projection.

An augmented reality (AR) head-up display (HUD) system based on holographic optical elements (HOEs) with multiple depths, large area, high diffraction efficiency and a single picture generation unit (PGU) is proposed. Since HOEs has excellent wavelength selectivity and angle selectivity, as well as the ability to modulate complex light waves, HOEs can image the red, green and blue parts of the color image emitted by PGU on different depth planes. The experimental results show that the three HOEs of red, green, and blue clearly display images at different depths of 150cm, 500cm, and 1000cm, and the diffraction efficiencies are 75.2%, 73.1%, and 67.5%, respectively. The size of HOEs is 20cm×15cm. In addition, the field of view (FOV) and eye-box (EB) of the system are 12°×10° and 9.5cm×11.2cm.

Time-of-flight (ToF) cameras obtain the depth information of the whole scene simultaneously by floodlighting it. In the presence of mutual reflection between the targets, the measurement would suffer from the multipath interference (MPI), because the signal received by the sensor is a mixture of direct component and multipath (or global) component. MPI would lead to significant measurement errors. In this paper is introduced a method of separating the direct and global components by once dot-matrix lighting and twice floodlighting with different frequencies. With dot-matrix lighting, the depth information at the dot matrix position is almost only determined by the direct component. The phase value of the direct component is readily calculated. The global component at the dot position with floodlighting is separated by minimizing the separation error through solving the nonlinear least square problem. Then the global component of the whole scene can be obtained by two-dimensional interpolation from data at dot matrix position. As a result, the direct component can be calculated by subtracting the interpolation result from the floodlighting signal, and the depth were calculated only from direct component. Semi-physical experiments were made for three kinds of scenes, blank corner with uniform reflectivity, the corner with different reflectivity areas and the corner with an object placed in front of it. The results demonstrate that the MPI has been significantly suppressed in all scenes. Especially in the area with strong MPI in the first two kinds of scenes, the measurement errors can decrease to about 10%~20%.

The retina is a light-sensing neural network, converting light energy to bioelectric signals for visual information processing. The retina is often damaged by neurodegeneration, which leads to severe vision loss. The initial symptoms of retinal neurodegenerative diseases are manifested by functional abnormalities. Delayed dark adaptation is one of the earliest functional symptoms of retinal neurodegeneration. Therefore, objective measurement of dark adaptation promises the early diagnosis of various retinal diseases. In this study, we demonstrate the feasibility of functional optical coherence tomography (OCT) imaging of the retina during dark adaptation in vivo. A custom-designed spectral-domain OCT was used in dark adaptation measurement, and two-month-old C57BL/6J mice were used in this study. We observed three image biomarkers. First, the outer retinal thickness became thinner during dark adaption. Second, OCT intensity of the inner segment ellipsoid zone was gradually decreased over time under darkness. Third, during dark adaptation, there was a rearrangement of the interdigitation zone between the outer segment and retinal pigment epithelium. Functional OCT enabled the concurrent measurement of retinal thickness changes and intrinsic optical signal (IOS) changes in the retina during dark adaptation. These functional biomarkers may help assess the early dysfunction of outer retinal neurons.

With the rapid development of stereoscopic 3D （S3D）display technology, some problems have gradually emerged. Among them, the color information of the S3D image has a significant impact on the visual comfort of S3D content. In this paper, in order to obtain more information about the visual discomfort of hue asymmetries 3D images and to ensure the reliability of the experimental results, we adopted a subjective and objective method. In the subjective experiment, the subjects rated the 3D images according to their own visual perception when watching the hue asymmetric 3D images. Finally, we will get the subjective score corresponding to each 3D image to indicate the visual comfort of the stereoscopic image. The objective experiment means that when the subjects watch the 3D images, we will use the eye tracker to record the subjects' blink data. Statistical analysis of visual comfort scores and blink data shows that as the hue asymmetric level of the 3D image increases during the viewing of hue asymmetric 3D images, the more blinks and the longer the blinking time. The stronger the visual discomfort.

The single integrating sphere combined with a spectrometer is often used to measure the diffuse reflectance spectrum of biological tissue. However, the optical fiber focusing of a spectrometer often introduces a certain error, which measures biological tissue parameters inaccurate. In addition, the traditional single integrating sphere is bulky and inconvenient to carry. The single integrating sphere system needs to be assembled with optical fiber, which is inconvenient to operate. Repeated disassembly and assembly of the system also cause the optical fiber interface to loosen, thereby introducing errors. In this paper, we proposed an integrated single integrating sphere system to measure the diffuse reflectance spectrum of biological tissues in vivo. The integrated single integrating sphere system is composed of a broad-spectrum light source, a stepping motor, a single integrating sphere, and an extended photomultiplier tube. The small-scale integrating sphere is convenient to carry, which can accurately and quickly collect the diffuse reflected light of the sample. The photomultiplier tube is used as a detector instead of a spectrometer to measure the diffusely reflected light from the sample passing through this small-scale integrating sphere. In this paper, we conducted an experiment with 7 subjects by this system to measure the diffuse reflectance spectra in vivo. We also used the Monte Carlo algorithm to verify the measurement results of the diffuse reflectance spectra. The Monte Carlo simulation results show that the system can achieve the in vivo non-invasive measurement of the diffuse reflectance spectra of biological tissues.

Intravascular ultrasound(IVUS) technology is one of the main technologies used to diagnose atherosclerosis. The accurate segmentation of the lumen and media area in IVUS images can help doctors perform clinical evaluations well. To overcome the effects of severe ultrasound speckles, various artifacts, and lesions in IVUS images, and improve the accuracy of IVUS image segmentation, an IVUS segmentation network MFU-Net based on multi-task learning is proposed. The algorithm selects the UNet network as the basic structure and introduces edge detection as an auxiliary task to build a multi-branch fusion deep neural network, which can force the network to pay attention to the edge information. The MFU-Net performance was verified on the 20MHz IVUS images data set, which is constructed by clinically IVUS images including a large number of interfering structures, such as calcified lesions, side vessels, vascular bifurcation, and stents. The artificial labels were annotated by two researchers with the guidance of a professional cardiologist. The experiment results show that the MFU-Net achieves 0.86 Jaccard measure(JM) for the media area and 0.91 Jaccard measure(JM) for the lumen area. Compared with the single-task UNet structure, the MFU-Net has higher segmentation accuracy and robustness and has a significant improvement in IVUS images containing vascular bifurcation and calcification.

In many scene classification applications, the variety of surface objects, high within-category diversity and between-category similarity carry challenges for the classification Framework. Most of CNN-based classification methods only extract image features from a single network layer, which may cause the completed image information difficult to extract in complex scenes. We propose a novel transfer deep combined convolutional activations (TDCCA) to integrate both the low-level and high-level features. Extensive comparative experiments are conducted on UC Merced database, Aerial Image database and NWPU-RESISC45 database. The results reveal that our proposed TDCCA achieves higher experimental accuracies than other up-to-date popular methods.

Upconversion nanoparticles (UCNPs) is a series of lanthanoid ions doped nanocrystals that are of great interest for biomedical applications, including nanoscale optical sensing and imaging, benefiting from its bright, stable, multicolour emission. Each of the nanoparticles contains thousands of Lanthanide ions, which works as both sensitizers and activators to absorb the near-infrared photons and transfer the energy from sensitizers to activators through nonlinear energy transferring process for an upconverting emission. A few new super-resolution imaging methods have been developed recently based on UCNPs’ unique nonlinear energy transferring process. Most recently, upon these advances, we have found that the thousands of Lanthanide ions provide a strong dielectric resonance effect in a single UCNP. In this work, we will review using the nonlinear response of lanthanoid ions to improve super-resolution nanoscopy. We will also report the ion resonance effect in UCNPs could substantially increase the permittivity and polarizability of nanocrystals, leading to an enhanced optical force on a single 23.3 nm radius UCNP, more than 30 times stronger than the reported value for gold nanoparticles with the same size. The enhanced optical force also provides a way to bypass the optical trapping requirement of “refractive index mismatch”. We further report that the resonance effect could engineer the Rayleigh scattering of UCNPs. These applications suggest a new potential of UCNPs as force probe, scattering probe and fluorescence probe simultaneously for multiplexed imaging.

Tracking and locating moving targets in real time and analyzing the motion of moving targets are extremely important for sports on ice. In practical applications, the moving target may exceed the field of view of a still camera, which makes it impossible to perform accurate motion capture and three-dimensional position positioning, and there is interference from factors such as illumination and shadows. In order to alleviate the above-mentioned problems, a method for motion capture of dynamic targets based on binocular vision is proposed. First, the binocular camera is calibrated by Zhang Zhengyou's calibration method, the internal parameters, external parameters and distortion parameters of the binocular camera are obtained, and the image distortion is corrected. Then the binocular camera is mounted on the smart car, and the motion capture method based on neural network is used to obtain the two-dimensional coordinates of the 18 key points of the target. The binocular camera obtains the disparity map through stereo matching, and the moving target can be obtained according to the stereo vision three-dimensional measurement method. The three-dimensional coordinates of the key point in the camera coordinate system. The binocular camera estimates its movement through IMU information and visual odometer information during movement, and finally calculates the three-dimensional coordinates of the moving target in the world coordinate system. Experiments show that this method can analyze the movement of moving targets in real time and overcome the limitation of the field of view of a still camera.

We propose a kinematics feedforward and PI controller combination algorithm. The proposed algorithm measures the speed of the target in the image coordinate system to calculate the normalized position and speed of the target, and uses the kinematics calculation of the target speed as a feedforward. This strategy improves the tracking accuracy of the PTZ. We designed a feedforward PI control simulation and hardware system test platform, and finally performed simulation and experimental verification. The results show that this article Compared with the traditional PTZ controller vision feedback control strategy, the proposed method has better dynamic performance and higher tracking accuracy.

Recently, the target tracking technology has become a popular scheme, which can assist skaters to achieve better results in training and competition. However, However, this is still a challenging task for smaller, fast-moving targets, such as skaters. To solve above problems, a target tracking method based on the combination of action-decision networks for visual tracking with deep reinforcement learning and the algorithm for precising bounding box estimation, and this method is used for skaters. The proposed method consists of two phases. In the first phases, the proposed method uses a tracker to obtain rough target tracking results, which search area is the position and size of the previous frame tracking result, and the tracker is controlled by sequentially pursuing actions. In the second phases, the alpha-refine, an accurate bounding box estimation algorithm, is applied to precise tracking of the target via the result of the first phases. That is, expand the prediction of the current tracking result to a concentric search area twice the size, and predict a more accurate bounding box in this area as the final tracking result. In addition, to train and test the proposed method, we produced a dataset for skaters. Compared with the traditional target tracking methods, the results of presented dataset show that the proposed method has higher tracking accuracy.

Non-Line-of-Sight technology is to image objects that are hidden from the camera's view. It has a wide range of application prospects in robotic vision, national defense, remote sensing, medical imaging, and unmanned driving. Active Non-Lineof- Sight imaging mainly relies on time-resolved optical impulse responses. The Non-Line-of-Sight imaging system emits ultra-short light pulses to illuminate the diffuse reflection wall, and uses ultra-fast time-resolved single-photon detectors to collect multiple reflected photon information, thereby obtaining information in the hidden scene. Finally, various reconstruction algorithms are used to reconstruct the hidden scene. However, most of the existing reconstruction algorithms have the problems of slow reconstruction speed and fuzzy reconstruction results, especially in the aspect of human pose estimation. In this article, we describe a method of active Non-Line-of-Sight human pose estimation based on deep learning. In order to solve the problem of lack of deep learning data, we simulate large amounts of pseudo-transient images for the network, including various complex actions: walking, jumping, turning, bending back and forth, rotating, using the confocal Non-Line-of-Sight imaging model. And then we train the simulated transient images using light cones Transformation and U-net coding and decoding network structure. Finally, we examine the performance of our method on synthetic and experimental datasets. The prediction results show that our method can not only estimate the pose of real measured non-view human pose data, but also significantly improve the quality of reconstruction.

OCT is a new imaging technology based on the principle of Michelson interferometer to perform tomography of sample tissue. OCT has the characteristics of high-resolution, non-invasive, real-time detection, and non-destructive imaging. The OCT vascular imaging method can generate high-contrast microvascular images without injecting a contrast agent. This paper presents a new microvascular imaging method (sv-cmOCT) that combines two existing microvascular imaging methods - speckle variance optical coherence tomography (svOCT) and correlation mapping optical coherence tomography (cmOCT). The binarised image of svOCT is used to mask the cmOCT image, which can effectively suppress background noise and static artefacts, and ultimately generate a microvascular image with high contrast. In this paper, blood flow mimic experiments and animal experiments were designed to validate the imaging performance of the microvascular imaging method. It is demonstrated that the proposed method can improve the signal-to-noise ratio of microvascular images well compared to methods such as svOCT and cmOCT and B-scan mask cmOCT. The study of microvascular imaging methods will greatly help to improve the understanding of the pathology of some diseases and thus may improve the clinical interpretation of human diseases!

Because LED has the advantages of energy saving, environmental protection, long service life and fast response time for traditional light sources, LED is gradually replacing the position of traditional light sources in the field of lighting, and LED is more and more widely used in automobile headlights. However, there are some safety problems during vehicle driving. For example, the glare of strong high beam lights on oncoming vehicles and passers-by at night has become the main cause of many traffic accidents. The emergence of ADB (adaptive driving beam) has reduced the occurrence of such traffic accidents to a certain extent, However, compared with pixel car headlights, it still has some limitations. For example, its controllable precision is not high enough, and its ability to deal with some more complex environments is not enough. At the same time, due to its projection ability, pixel headlights can not only remind drivers of driving safety, but also help other traffic participants, and their safety has been greatly improved. This paper introduces a new type of intelligent automobile headlamp based on pixel led. In this automobile headlamp, five LEDs with 1024 pixels are used, which are divided into imaging area and lighting area. The light distribution lens matching each area is designed. At the same time, the brightness and brightness of each pixel LED are controlled through circuit design, so as to realize the projection imaging of traffic signs and the light type transformation of high beam anti glare and low beam. Pixel LED car headlights have more accurate control accuracy and higher safety than general adaptive car headlights.

Optical coherence tomography (OCT) is a high-resolution non-invasive 3D imaging modality, which has been widely used for biomedical research and clinical studies The robotic-arm-assisted OCT system has the advantages of multiple degrees of freedom and high precision and stability during the imaging process, which can be used to assist doctors in quantitative assessment of vascular status. In this work, we propose a method for system calibration and the pose optimization of the OCT probe to ensure that the probe is in the optimal imaging pose. The system calibration is implemented with the corresponding pixel domain to spatial domain conversion coefficients and Tsai-Lenz method. The pose optimization of the OCT probe can be optimized with image processing. The Laplacian random walk algorithm was used to obtain the skin phantom surface contour to calculating the normal vector p^→ of the skin phantom surface. Meanwhile, the contour of vessel phantom was segmented with the convolutional neural network method. The radius vector b^→ of vessel phantom was obtained with the segmented results. There are two vectors that determine the attitude of the probe. Vp denotes the unit vector parallel to the optical axis while Ve denotes the unit vector parallel to the B-scan direction. When the OCT probe is in the optimal imaging pose, the vector Vp should be parallel to the p^→ and the vector Ve should be parallel to the b^→ . We demonstrate an experiment with vessel phantom and skin phantom. It is expected that robotic-arm-assisted OCT imaging can ensure precise data acquisition process assisting the intraoperative decision making.

Since the single-pixel imaging technology based on compressive sensing was proposed, single-pixel three-dimension (3D) imaging has attracted a lot of attention in the scientific research circle, because it also has the advantages of wide waveband and high sensitivity. In this paper, the calibration of fringe projection 3D measurement and point cloud computing method are applied to the field of single-pixel 3D imaging. The deformed fringe images are compressed and reconstructed using greedy algorithms and L1 norm minimization method. Finally, the 3D point cloud information of the object is calculated from the deformed image. We designed a simpler imaging system, compared the accuracy differences of different compression ratios and reconstruction algorithms in 3D measurement, and the results of simulation analysis were given at the end of the paper.

In recent years, the field of image super-resolution has mainly focused on the single-image super-resolution (SISR) task, which is to estimate an HR image from a single LR input. Due to the ill-posed ness of the SISR problem, these methods are limited to increasing the high-frequency details of the image by learning the a priori of the image. And multi-frame super-resolution (MFSR) provides the possibility to reconstruct rich details using the spatial and temporal difference information between images. With the increasing popularity of array camera technology, this key advantage makes MFSR an important issue for practical applications. We propose a new structure to complete the task of multi-frame image super-resolution. Our network takes multiple noisy images as input and generates a denoised, super-resolution RGB image as output. First, we align the multi-frame images by estimating the dense pixel optical flow between the images, and construct an adaptive fusion module to fuse the information of all frames. Then we build a feature fusion network to simultaneously fuse the depth feature information of multiple LR images and the internal features of the initial high-resolution image. In order to evaluate real-world data, We use the BurstSR data set, which includes real images of smartphones and highresolution SLR cameras, to prove the effectiveness of the proposed multiframe image super-resolution algorithm.

In the field of computer vision, template matching technology is an important research direction. This technique compares the template image with the sample image to find out the position of the template image in the sample image. It has the characteristics of simple algorithm, small amount of calculation and high recognition rate, so it is usually used in other computer vision fields such as object detection and target tracking. In addition, with the popularity of infrared sensors, and infrared images can obtain additional information that is not included in visible light images, the integrated processing of visible light and infrared information has always been a research hotspot. The traditional template matching algorithm mainly focuses on the matching between visible light images. For the information difference between visible light and infrared images, the traditional template matching algorithms are difficult to achieve accurate matching between the two types of images, and the amount of calculation is large. In response to this problem, a template matching algorithm based on feature extraction of convolutional neural networks is proposed in this paper. Our method draws on the robust template matching using scale-adaptive deep convolutional features. We use a scaleadaptive method to extract the deep features of visible light and infrared images, and then uses the traditional NCC matching algorithm to obtain the matching position of the template on the feature map. Then the regression and optimization of the template position are performed to obtain the position of the template image on the sample image. The research results show that our method can achieve the matching of the infrared template on the visible light image, and the position error is not large.

World Health Organization(WHO) estimates that there will be 300 million diabetic patients worldwide in 2025. The real-time monitoring and detection of blood glucose levels is considered a way to prevent the occurrence of complications.The scattering coefficient(μs), which can be used for noninvasive BGC monitoring, in the dermis layer of human skin can be obtained by optical coherence tomography(OCT), and it has a strong correlation with the blood glucose concentration (BGC). Unfortunately, the non-homogeneity in the skin may cause inaccuracies for the BGC analysis, and the different physical conditions on different people may also cause errors. In this paper, The 3D correlation analysis method is applied to identify the regions in the skin where the μs is sensitive to BGC variations. To reduce the error of the non-homogeneity regions and improve the accuracy of OCT-based BGC monitoring, the proposed method only uses the μs in the high correlation regions for building the blood glucose prediction model. Experiment shows that the proposed method is better than that obtained with the previously reported 2D correlation method. We believe that the method demonstrated in this paper is important for understanding the influence of BGC on μs in human skins and therefore for improving the accuracy of OCTbased noninvasive BGC monitoring, although further studies are required to validate its effectiveness.

The evaluation of port wine stain based on three-dimensional information can overcome the inaccuracy of twodimensional image evaluation methods commonly used in clinic. In this paper, an end-to-end multitasking method is designed for the application of 3D information acquisition of port wine stain. Based on deep learning and position map regression network, the reconstruction from 2D pictures to face 3D point cloud is realized. the facial information of patients with port wine stain is represented by UV position map recording 3D point information of the face, and the dense relationship between 3D points and points with semantic meaning in UV space is characterized with this method. The deep learning network framework based on Encoder-Decoder structure is used to complete unconstrained end-to-end face alignment and 3D face reconstruction, whose parameters are obtained by training the data set with lightweight CNN structure. In the process of neural network training and end-to-end unconstrained image facial reconstruction, each point on the UV position map can be assigned different weights, which can not only be used to improve the network performance in neural network training, but also be used to assign corresponding weights to the focus areas with different disease course in the three-dimensional information reconstruction of the focus area therefore the accuracy of the reconstruction results can be increased. With the help of this method, the three-dimensional reconstruction results can be quickly obtained from a single patient's face image, which can be used for subsequent accurate lesion information analysis and treatment.

With microLED panel technology quickly evolving to smaller pixel size and larger resolution, optical metrology is on-demand to support both design verification and process yield control by providing a solution with high resolving power high through-put and less calibration spectrum dependency. This paper reviews the trade-off between all conflicting factors, and discusses the calibration algorithm to remove the spectrum dependence, describes a novel imaging colorimeter which precisely attacks all above technology inconsistencies with the final goal of: single micron pixel resolving power by combination of optical resolution and digital imaging processing algorithm, large optical FOV to reduce number of frames to be captured for whole panel inspection, calibration algorithm to precisely transfer the true color and brightness information between NIST traceable light source without spectrum matching demand, and inherent final balanced high through-put image capturing.

Digital micro-mirror device (DMD) has considerable influences on the high quality computer-generated holographic display. We propose a DMD transmission model for analyzing the influences of the diffraction characteristics of DMD in this display process. The propagation of the hologram affected by the DMD modulation function is calculated based on the discrete Rayleigh-Sommerfeld diffraction integral. The results demonstrate that the diffraction characteristics of DMD do not essentially change the reconstructed wavefront of the original hologram, but cause the lateral offset of the reconstructed image. This offset can be compensated by changing the angle of the incident beam, and the image can be reconstructed in the correct position. Numerical simulations demonstrate the effectiveness of the proposed model, which provides theoretical support for DMD-based computer-generated holographic display.

In the context of Industry 4.0, a new concept called “smart oilfield” is introduced, which relies on large-scale information exchange among various parts and there is an urgent need for special fiber links for both increased data transmission capacity and high-sensitivity distributed sensing. Naturally, multicore fibers can be expected to play a critical role in smart oilfields, in which part of cores are responsible for data transmission, while other ones are used for sensing. In this paper, we propose a heterogeneously integrated seven-core fiber for interconnection and awareness applications in smart oilfields, which not only could support digital and analog signal transmission, but also could measure temperature and vibration through Raman effect and phase-sensitive optical time-domain reflectometry. The core arranged in the center is used for digital transmission to maintain communication stability, while others are arranged around the center for analog transmission and sensing with equal pitches to increase sensing sensitivity. The core for digital signal transmission has low differential mode group delay of 10 ps/km over C-band and the crosstalk between adjacent cores is lower than -55 dB/km at the pitch of 50 μm. A 25-Gbaud real-time transmission over 50 km is simulated as a proof of concept. Each core for analog signal transmission has a large effective area of 172 μm² to suppress the nonlinear effect due to the watt-scale input power. In all, the proposed heterogeneously integrated multicore fiber exhibits great potential to be applied in smart oilfields, meeting the demand for efficient and cost-effective oil production.

As a new type of zoom element, the liquid lens has the characteristics of fast response, no wear, and small size, which brings new ideas to the design of the zoom system. This paper proposes a liquid-solid composite zoom optical system without mechanical movement. In the system, two liquid lenses are used as the zoom group and the compensation group of the system, so that the image plane position is always stable during the zooming process of the system. Based on the Gaussian theory, the relationship between the focal length of the system and the focal length of a single liquid lens is derived, and a multi-layer traversal algorithm is proposed to solve the initial structure parameters of the system. The optimized design of the zoom optical system and image quality analysis was completed by using ZEMAX software. The results show that the system can achieve continuous zoom in the range of 20-60mm, the zoom ratio of the system is 3, and the image quality is good. The modulation transfer function of each focal length of the zoom system at a spatial frequency of 50lp/mm is greater than 0.2, and the RMS value of wave aberration is greater than 0.3 times the wavelength of the probe light, which meets the design requirements.

Long-distance measurement systems are widely used in vehicle ranging, obstacle avoidance, virtual reality and other fields. This paper proposes a fast and effective binocular stereo vision model method to improve the accuracy of long-distance measurement systems. We explored the experimental constraints of Zhang Zhengyou's calibration method, combined with the epi-polar correction method for image matching, improved ranging model based on binocular parallax, to attain accurate measurement results. We performed on obstacles within 100 meters to verify the effectiveness of the system. Using the method in this paper, the relative error within the overall ranging range is less than 10%，where the measurement error is less than 3% (distance: <60m) and less than 8% (distance: 60m~80m). The results show that ranging error is relatively smaller in the optimal calibrated sampling space. Our work has effectively improved the accuracy of longdistance measurement, which is expected to realize obstacle measurement at a low cost, improving the environmental perception ability based on passive ranging.

The research and development of miniaturized and stable hyperspectral imaging spectrometer has far-reaching significance. In this study, it is designed with grating as the spectroscopic element, in order to reduce the volume of the instrument without sacrificing spectral imaging performance. Comparing the advantages and disadvantages of the spectral spectroscopy structures of three different structures, the imaging spectrometer is designed with improved Schwarzchild "M"-type optical path structure with the planar grating. On the basis of good correction image difference, the process of processing is simplified while the target of high resolution, large aperture and wide band is reached. The MTFs of the optical system of visible optical channel reached 0.7, and the main wavelength distortion was about 0.01%, and relative aperture of is 1/2.5. Using ZEMAX optical design software, the optical elements are designed independently for four typical bands (one visible light band and three infrared bands) and the aberration is corrected, and the light trace and optimization are carried out on each channel, so that continuous imaging detection of wide band multi-channel hyperspectral can be achieved in a small volume.

Fusion of infrared (IR) and visible sensor images is the widely applied. The registration is the basis for sensor fusion and fusion methods are very sensitive to the level of registration accuracy. However, the different imaging systems of the both lead to quite different image characteristics in a same scene and significant misalignment due to differences in field of view, lens distortion and other camera characteristics. So that, the registering accurately the both of sensor images is very challenging. A method to improve the accuracy of image registration is proposed in this paper. The method is based a self-designed long-wave IR/visible dual-band imaging system for capturing simultaneously both of images, which is not only with synchronous focusing and optical registration as much as possible, but also with a sliding potentiometer that record the voltage corresponding to the focusing distance. At first, an affine transformation of the registration at several different distances is acquired with two calibration board images captured self-designed calibration board by the dual-band imaging system. Then, the affine transformation matrices corresponding to the several finite distance are interpolated to obtain more affine transformation matrices within a certain distance, and an accurate look-up table is established. Last, when the dual-band imaging system is working, the current focal length is read out according to the sliding potentiometer in the system, and then the corresponding affine transformation matrix is searched for image registration. The proposed method is evaluated by comparing the deviation of the corresponding feature point coordinates on both of calibration board images before and after registration. Experimental results show that the proposed method can improve the registration accuracy of IR and visible images at different distances.

The probability of pulmonary embolism in lung cancer is more than 20 times that of normal people, and its mortality increases by 2 to 8 times. Once pulmonary embolism occurs in lung cancer patients, it will increase the difficulty of diagnosis and treatment and shorten the survival period. Therefore, early diagnosis and prevention are critical. Imaging is a non-invasive technique used to identify individuals with symptoms of disease. Advantages of plain X-Ray photography: easy to use, economical and inexpensive. However, the use of X-Ray for the diagnosis of lung cancer in combination with pulmonary embolism often requires a very experienced imaging physician. This limits the large-scale diagnosis of the population. In recent years, it turns out that more and more research is based on the application of deep learning to images. And medical imaging is one of the most promising directions, with systems based on deep learning and image processing providing aids to decision-making in the diagnosis and prevention of many diseases. This paper compares several methods three layer neural network, EficientNetB0, VGG16 , KNN, SVM and transfer learning for automatic multi-classification of lung cancer images. We are using a available dataset from Yunnan First People's Hospital with X-Ray images from normal (160), lung cancer (1218) and lung cancer complicated with pulmonary embolism (88). As you can see, our dataset is not perfect and the imbalance between the three categories is more prominent. Therefore, we tried transfer learning using fine-tuning, which resulted in an accuracy rate of over 95%, and is proved to be the best performance. The method can assist doctors in decision making and improve diagnostic efficiency.

Digital cameras have a wide range of applications besides taking pictures and many scholars have focused on how to use a digital camera to recover spectral reflectance of the objects in a scene. However, it is difficult to accurately reconstruct the reflectance from the RGB values with only three dimensions. Several methods have been developed, such as pseudo-inverse (PI), the principal component analysis (PCA) and even deep learning. In order to improve the reconstruction accuracy, research continues to come out. This paper proposes a method of spectral reflectance reconstruction based on the PCA and color objects classification using a color digital camera. The method is based on the three databases previously created, which are the spectral reflectance database of natural objects and their eigenvectors according to of color class of the objects, the spectral power distribution (SPD) of sunlight and its eigenvectors, and the camera's spectral sensitivity functions. The process of proposed method is as follows. Firstly, the camera takes an image of a natural scene containing a color card, and the SPD of the scene light is estimated through spectral reflectance of the color card. Secondly, each pixel of the image is classified to color class of objects according to its CIEXYZ value calculated through the camera's color characteristic data. Thirdly, spectral reflectance of each pixel is reconstructed according to the classes of the objects through eigenvectors of the spectral reflectance and sunlight. Finally, to prove the feasibility of the proposed method, an experiment of image color reproduction is carried out on the scene with plant leaf based on the estimated light sources and reconstructed spectral reflectance, and the results show that the accuracy of spectral reflectance reconstruction based on objects classification has been improved.

An improved target recognition algorithm based on YOLOv5 network was proposed to solve the problem of low accuracy of target recognition due to the complex background of remote sensing image, small difference between target classes and multiple and dense targets which in the target detection task of optical remote sensing image taken from the overhead angle.The algorithm improves the target recognition by changing the visual activation function, loss function and improving the structure of feature pyramid network (FPN). Firstly, the original YOLOv5 network structure is analyzed that the key technologies of input, Backbone, neck and output are introduced. The LeakyReLU activation function has been changed to FReLU which has the ability to adaptively obtain the local context of the image and is simple in form that improving the spatial sensitivity in the activation function. PIOU loss function is used to replace common CIOU and GIOU that rotation parameters are added to better detect rotating and dense objects. Feature Pyramid Grids (FPG) are introduced that the Feature scale space is represented as a regular grid with parallel bottom-up paths which fused by multi-directional horizontal connections. The single path feature pyramid network is improved by significantly improving its performance at similar computational cost. Experimental results show that under the same training conditions, compared with the original YOLOv5 network, the improved YOLOv5 network converges the training results faster, and the average recognition rate of the training model increases by 5%. Through the test set verification that the recognition accuracy of all kinds of images has been improved which the average accuracy has reached 0.702, 7% higher than the original YOLOv5 network.

Control of terahertz (THz) wave polarization state is of great significance for imaging and communication. Dynamically control of THz wave polarization state is achieved by an electronically controlling composite metasurface consisting of the gold cross antennas and a monolayer graphene. The graphene composite metasurface acts as a quarter-wave plate when the external control voltage is 0 V, by which the polarization state of the incident THz wave is converted from linear polarization to circular polarization. When voltage is increased, the chemical potential of graphene is increased gradually, transmitted polarization state of the THz wave is changed from right circular polarization to right elliptical polarization, and to linear polarization. Furthermore, the polarization state of the THz wave is able to be changed from left circular polarization to left elliptical polarization, and to linear polarization if the device is clockwise rotated by 90°. Our work will offer a new avenue for tunable THz polarization modulation devices.

How to detect targets under poor imaging conditions is receiving significant attention in recent years. The accuracy of object recognition position and recall rate may decrease for the classical YOLO model under poor imaging conditions because targets and their backgrounds are hard to discriminate. We proposed the improved YOLOv3 model whose basic structure of the detector is based on darknet-53, which is an accurate but efficient network for image feature extraction. Then Squeeze-and-Excitation (SE) structure is integrated after non-linearity of convolution to collect spatial and channel-wise information within local receptive fields. To accelerate inference speed, Nvidia TenorRT 6.0 is deployed into on Nvidia Jetson series low power platform. Experiments results show that the improved model may greatly achieve the inference speed without significantly reducing the detection accuracy comparing with the classic YOLOv3 model and some other up-to-date popular methods.

We propose a type of hybrid structure metalens composed of a graphene-loaded metallic metasurface sandwiched by two mutually orthogonal gratings, which can work in transmission modes for dynamic terahertz wavefront manipulation with tunability and enhanced focusing efficiency. Experimental results show that due to the multi-reflection between the metasurface layer and the grating layers, the focusing efficiency is enhanced by 1.8 times, and the focal length of the metalens is increased by 0.61mm by increasing the applied gate voltage on the graphene from 0V to 1.4V. We hope the proposed structure may open a new avenue for reconfigurable THz metasurfaces with high efficiencies.

With the wide application of infrared searching and tracking system, variable field of view or zoom optical system is needed to realize the function of searching targets in wide field of view and identifying targets in narrow field of view. The traditional fixed F-number optical system often ensures that the aperture utilization is 100% in narrow field of view. When switching to wide field of view, the aperture is greatly reduced and the energy acquisition efficiency is reduced. Therefore, optical systems with variable F-number highlight the advantages gradually for gaining more target energy to improve detection capability. This paper offers an optical design of cooled mid-infrared system with variable F-number. The operating band is 3.7μm -4.8μm, and the focal length of NFOV is 300mm, F-number is 4, the focal length of WFOV is 150mm, F-number is 2, both of the aperture isФ75mm. The energy acquisition efficiency of variable F-number optical system is four times that of fixed F-number optical system. Therefore, the target search in wide field of view can detect weaker infrared targets or achieve longer detection distance, which greatly improves the system capability.

In this paper, an EOH based multi-spectral image registration algorithm is proposed, which is robust to rotation and scale changes. The key points of EOH descriptor have no main direction, and the neighborhood size of key points is fixed, so it is not robust to rotation change and scale change. The existing multi-spectral image registration methods mainly use the gradient features of the neighborhood of key points, but the gradient information between multi-spectral images is not stable, resulting in the limited improvement effect of these methods. The method proposed in this paper uses mutual information measure to calculate the relative rotation angle between images, so as to determine the main direction of key points, calculate the key point descriptor according to the main direction of key points, and make the size of key point neighborhood change with the scale of key points. Experimental results show that the proposed method in this paper is more robust to rotation and scale change.

The head mounted display (HMD) provides people with a new visual acquisition method that users can obtain the visual information provided by the external scene and the display device at the same time. However, the new display method also increases the difficulty of optical system design. The display system should be compact and lightweight, and provide sufficient field of view (FOV) in a large eye-box. The holographic optical waveguide display has the characteristics of light weight and good perspective, so it is very suitable for HMD. The holographic optical element (HOE) has a high degree of design freedom, which provides more possibilities for the optical system design. To enlarge the FOV, we use two HOEs and stack them, each HOE provides a small FOV, and the two FOVs are spliced together to form a large FOV. To extend the eye-box, we duplicate a single viewpoint several times and distribute the viewpoints evenly in space. Specifically, first, a single viewpoint of the system is copied into two viewpoints separated from each other using a multiplied HOE. Then two polarization gratings placed in parallel are used, which can convert the incident linearly polarized beam into two circularly polarized beams that separate in the opposite direction. The two polarization gratings double the two viewpoints obtained in the first step to four viewpoints. Around these viewpoints, the user can see a clear and complete image.

An integral (II) imaging method using quarter-overlapped microlens arrays (QOMLA) is proposed to improve the display performance. The principle and the simulation of POMLA is analyzed by geometric optics and wave optics, and the optical experiments verify the enhancement of performance. POMLA can double the angular sampling density of II systems in each dimension to increase the spatial resolution, and is able to construct multiple central depth planes by adjusting the gap between the two layers to expand the depth of field. Moreover, POMLA is easier to be processed compared with the same lenslets size single-layer microlens array and reduces the processing cost.

In traditional optical microscopy imaging system, the resolution of time mainly depends on the detector’s detection speed, usually in millisecond or microsecond magnitude. While the spatial resolution is limited by the optical diffraction limit, the lateral resolution of ordinary microscopies generally only reaches 200nm.Just as biological structure has a wide spatial scale, biological living processes also have a broad time scale. When observing biological subcellular organelles, resolution and the speed of life activities should be considered. For the dynamic process, it’s meaningless to simply improve the spatial resolution without correspondingly increasing the imaging speed which should be no less than the movement speed of the observed object.

Existing super-resolution or high-speed optical imaging is limited by the mutual constraints of spatial and time resolution, making it difficult to obtain both super-resolution and high-speed optical imaging. In order to break the constraints of this game and gain high-speed super-resolution images, the CS-SIM system combines SIM (structured illumination microscopy) and CSP (compressed sensing photography). Since they both are wide filed imaging and CSP is a passive receiving imaging technology, SIM and CSP have potential to combine closely and achieve super-resolution and highspeed imaging.

Based on the diffraction neural network, a small-volume diffractive optical element for optical logic calculation in the terahertz band is designed. The optical logic calculation ability and calculation accuracy of the diffraction neural network are calculated. The results show that after training, the diffraction neural network can accurately calculate eight kinds of optical logic and has 100% recognition accuracy when the number of network layers is greater than 5. This article mainly introduces the design principle and calculation process of the diffraction neural network, and compares the recognition accuracy of different network layers.

Laser-induced graphene (LIG) has received extensive attention due to its excellent properties such as high electrical conductivity, high thermal stability and electrical conductivity, simple synthesis, and low manufacturing cost of patterned structures. However, most research on LIG has focused on electrical applications. In this work, we first examine the influence of the substrate on the LIG generated on polyimide, and then fabricate patterned LIG structures, including gratings and Fresnel zone plates for terahertz (THz) wave modulation. The function of the structure is proved through the experiment of the terahertz focal plane imaging system. It is expected that the LIG-based structure can widen the application of THz technology.

Battlefield situational awareness is the core condition that determines the success or failure of the battlefield, and it is also an important application direction of photodetectors. The rapid development of AI technology in recent years is about to cause major changes in future wars. The new AI battlefield will also put forward new urgent needs for situational awareness. This article summarizes the current main modes of collaborative detection of battlefield situation awareness and its research status, including radar / infrared composite detection, multi-source data fusion of radar / infrared detection, cooperative target recognition, target tracking, etc. On this basis, combined with the current development trend of the intelligence level of the main battlefield equipment, we get the development needs of future intelligent battlefield situational awareness for new types of collaborative detection, including requirements for its style, angle, speed, and detection targets of distributed collaborative detection. Based on this, the key development directions and core issues to be solved for intelligent battlefield situational awareness in the future are proposed.