This PDF file contains the front matter associated with SPIE Proceedings Volume 11565, including the Title Page, Copyright information, and Table of Contents.

The direct pickup of integral imaging typically needs to overcome the limitation on the restricted depth of field under a lenslet array. We use the portable commercial Lytro standard camera as the integral imaging system and optically display these 4D light field data by the proposed parallax boosting algorithms with enhancing parallax. Using depth slicing, retargeting and parallax boosting, the depth of the integral imaging display is presented in a uniform compression with the same proportion as the real scene. The simulation results show the feasibility of the proposed method and provide an efficient way for the acquisition of elemental image array.

In this paper, we discussed the quality of image and the difficulty of the two models, multi-projector panorama system and fish-eye-projector system, according to principles and experimental operations. The multi-projector stitching model mainly analyzes simplified operations and the cooperation of softwares and hardwares in the course of the projection and optimization. Based on fish-eye-lens, we also designd a fish-eye-projector made of fish-eys-lens and ordinary projector to analyze the quality of lens from the perspective of a normal user.

We have promoted and demonstrated a relationship between speckle contrast and the fluctuation of the screen surface in Laser Display Technology (LDT). Based on the speckle theory, we analyzed the speckle characteristics of diffuse scattering screen, and obtained the relationship between speckle contrast C and the standard deviation of surface fluctuation height σ_h. Meanwhile, we used the precision probe surface profilometer to measure the height and morphology of 12 kinds of screens. Finally, we built a speckle testing platform to simulate human eye, and measured the speckle contrast of 12 kinds of screens. By analyzing the gain and speckle contrast parameters of glass bead screen and metal screen, we could conclude that glass bead screen was more suitable for LDT.

A novel design for speckle reduction based on a rotating ball lens is investigated. In addition, a standardized speckle measurement method is also developed and the parameters mainly accord with characteristics of human eye. This method could open up an avenue to the practical speckle measurement application.

Retinal projection display (RPD) is a research hotspot in the field of near-eye display (NED), which has a long depth of field and is able to overcome the vergence-accommodation conflict (VAC) because of the principle of Maxwellian view. However, the existing problems of RPD, such as small field of view (FOV) and large volume heavy weight, restrict its application and development. In this paper, an RPD system based on micro-electro-mechanical system (MEMS) and conicoid curved semi-reflective mirror is proposed in order to realize a large FOV with compact structure and light weight. The MEMS used is a kind of biaxial scanning mirror, which works in coordination with RGB laser diodes for the sake of scanning a two-dimensional image in a specific direction with high illuminance and high resolution. The conicoid curved semi-reflective mirror is used to project the image onto the retina and form a large FOV with a suitable eye relief distance (ERF), meanwhile to keep the external object visible. The combination of MEMS, RGB laser diodes and conicoid curved semi-reflective mirror retains compact size factor suitable for wearable function. The performance of the proposed RPD is quantitatively analyzed. And the key performance is set forth in detail as follows. The FOV is 70° horizontal (H) × 40° vertical (V), ERF is 30mm. And the MTFs are discussed as well. The proposed RPD system realizes a large FOV, a long depth cue and high performance with compact size, which can be further applied in the next-generation optical see-through NED.

It has been a challenge to design an automotive head-up display (HUD) that has large eyebox while maintain a compact structure. In this paper, we present a combiner-type HUD design using freeform mirrors. The HUD is an off-axis fourmirror system consisting of 4 mirrors, which are enclosed in a box to achieve a compact form. The tilt angle of the combiner is adjustable, which can be adapted to different heights of exit pupil. We initially propose an off-axis, fourmirror system which only uses spherical surfaces, then we optimize it by replacing with freeform surfaces, which enables us to balance the residual asymmetrical aberrations. Finally, we obtain an HUD system with good imaging performance. The virtual image is projected at 5 meters away from observer, and the eye box is 106 × 66 mm² .

With the rapidly development of optoelectronic technology and computer technology, the projection display technology has also been unprecedentedly developed. In this paper, a zoom projection objective lens is designed based on Digital Light Processing (DLP) projection display technology, the 0.47in DMD chip of Texas Instruments is selected in this paper, its resolution is 1920×1080 and its micromirror pitch is 5.4 μm. The optical system uses a 100% offset transmissive projection structure. The working wavelength of the whole zoom projection lens is visible light, and 14 lenses including protective glass are used for the projection lens, each lens in the designed system is spherical, the total length of the system is 220mm, the aperture of the optical system is about 58mm , focal length of the zoom projection lens ranges from 25mm to 32 mm, viewing angle ranges from 41 degree to 51 degree , and F number is 1.7 , the distortion of the designed system is less than 3%, the projected image plane relative illumination of the zoom projection lens above 90% , the telecentricity of the system is less than 1 degree. The MTF value is close to 0.3 at the Nyquist frequency 93 lp/mm for the full field of view, the designed system meet the requirements of projection. In order to meet the current processing level, the radius of curvature tolerance, the spacing tolerance, the eccentricity tolerance and the tilt tolerance are analyzed. The results show that the zoom projection lens is easy to process and produce.

A multiview autostereoscopic display system using holographic projection is proposed, which solves the problem of reconstruction damage existing in the traditional display when the image is incompletely loaded. The elemental image (EI) captured from the 3D scene is encoded, and a single reconstructed scene is obtained by adding the phases of the blazed grating and the converging spherical wave. An imaging lens is introduced to control the position and size of the reconstructed scene, so that the reconstructed scene can be matched with the lenticular lens to generate parallax images. The effectiveness of this method was verified by optical experiments. Experimental results show that the system can generate parallax images, and compared with the traditional multiview autostereoscopic display, it can reproduce the complete scene when the image is incompletely loaded, which improves the stability of the imaging system.

Poly-capillary X-ray lens provided a special mechanism to control X-ray radiation. In recent reported works, polycapillary X-ray lens started to be applied in medical imaging fields. The use of poly-capillary X-ray lens in medical imaging systems brought proper benefits in Compton scattering rejection and resolution enhancement. However, the Xray intensity distribution of the output has unsatisfied uniformity when using conventional poly-capillary lens if the primary rays emitted by a common X-ray tube. Uneven intensity distribution will cause problems such as details loss and longer exposure time for better reconstruction results. In this work, a new type of poly-capillary X-ray collimating lens was introduced for medical imaging. Different from the conventional ones, this kind of lens was divided into three regions and subarray units of different diameter channels were placed in their corresponding regions. A program was written based on ray-tracing method for optical design and imaging simulation. The optical properties of this new type collimating lens were tested, using a 50.0μm Cu target X-ray tube with 20.0kV voltage and 0.2mA currents. A plateau region that had good uniformity which is 4.0mm in diameter was obtained. In addition, X-ray transmission imaging experiments for small samples showed improvements in imaging qualities using this new optical component. Although the existence of manufacturing difficulties effects the finished product quality, this new kind of X-ray poly-capillary lens shows its potential in medical imaging.

We quantitatively analyze the effect of the tilted misalignment between the display panel and the reconstruction lenslet array on the reconstructed image quality in integral imaging system. First, the geometric optics and the ray tracing theory are utilized to build the tilted misalignment theoretical model based on the integral imaging principle by considering the optical axis tilt and the pixel information dislocation. Then, the degradation of the image quality in the integral imaging system resulted from the tilted misalignment is simulated on the basis of the model mentioned above. Finally, the peak signal-to-noise ratio (PSNR) is introduced to quantitatively characterize the influence degree of the tilted misalignment on the reconstructed image quality, and the limitations of the tilted misalignment are given corresponding to the human vision system observable threshold for different distance between the display panel and the reconstruction lenslet array. Experimental results show that the smaller the distance between the display panel and the reconstruction lenslet array is, the larger the tilted misalignment limitation is. Research results will provide theoretical support for the parameters design and assembly of the display module in an integral imaging system.

With the development of space technology, more and more Fiber optic gyroscope is used in spacecraft and satellite. Due to the existence of space radiation environment, it will cause serious damage to some components in the Fiber optic gyroscope, especially the fiber ring, and have an important impact on the optical power transmission of the fiber ring. In recent years, the research shows that space irradiation has serious damage to optical elements. In this paper, for a 1550 nm Panda optical fiber, the influence of the radiation damage of this kind of optical fiber is obtained by gamma ray and electron source irradiation respectively. The transmission loss of the fiber is up to 17.5 dB/km after 1Mrad (Si) is irradiated by gamma ray, and 53.2 dB/km after 5×1015/cm² electrons is irradiated by 0.2MeV electron source. Compared with the previous 1310 nm optical fiber irradiation damage, and through the analysis of the mechanism of optical fiber irradiation damage, it is concluded that the impact of 1550 nm optical fiber irradiation damage is greater than that of 1310 nm optical fiber irradiation damage. In addition, for the degradation of 1550 nm optical fiber, the influence of fiber radiation damage on the bias drift and random walk of the whole fiber optic gyroscope system is analyzed. Finally, based on the radiation damage mechanism of optical fiber and the experimental results of radiation damage of optical fiber, some suggestions on the radiation resistance design of optical fiber ring are given.

To simulate the source region electromagnetic pulse, a composite system was constructed by spatially combining a magnetic field coil with a bounded-wave simulator. This system can support the electromagnetic environmental effect tests that have both a pulsed magnetic field and pulsed electric field. However, the simulation results show that the magnetic field coil may cause electric field waveform distortion in the parallel section of the bounded-wave simulator. To explain the mechanism of electric field waveform distortion, a viewpoint was put forward that the spatial interaction relationships between the external magnetic field coil and the upper and lower plates of the bounded-wave simulator can be represented by three equivalent capacitors. In addition, the asynchronism of electric field changes was due to the charge-discharge property of the upper and lower capacitor on the intermediate capacitor. Furthermore, a circuit model based on these equivalent capacitors and inductor was established. The calculation results based on the circuit simulation software LTspice showed that, the voltage waveforms of the equivalent capacitors can be consistent with the electric field waveforms in the corresponding spatial region. These results verify the correctness of the equivalent capacitors and the circuit model proposed in this paper.

In this paper we proposed a method of monitoring industrial robot reducer online cloud platform. Based on the theory of cyber physical model, a cloud state monitoring system is designed due to the problems of complex objects, massive failure information, high stable demand and poor real-time control ability in the modern industrial robot maintain. It includes intelligent sensor and edge computing to network cloud technology. With data model, the system uses deep learning and over-limit learning algorithm on the cloud management center server. The deep learning method trains and predicts the collected fault data, and can predict the malfunction of vector reducer. Further, the intelligent optimization scheduling algorithm is used in the cloud to obtain real-time fault in industrial robot reducer. The problem of distribution and the dynamic running performance of cloud system is monitored and improved.

Simulation analysis of radiation field characteristic of low frequency wide pulse electric field simulator using in the chamber were done by CST. The results are that the use environment in the chamber will affect the radiation field characteristic of the simultor, and the radiation field will be astricted in the magnetic field generating coils. In the chamber condition, the simulator can achieve a risetime about 30ns, a pulse width wider than 1ms, with the highest field strength 100kV/m, as well as a lower than 10% heterogeneity in 1.6 cube metre space.

In order to evaluate the attenuation characteristics of low frequency pulse magnetic field penetrating through reinforcement nets, different types of reinforcement cages are made. Radiation effect experiments under low frequency pulse magnetic field are done. Effect rules on attenuation characteristics of different diameter, different mesh size, different intersection point treatment, different connection mode and different radiation parameters are carried out. The results show that different types of reinforcement net have a kind of shielding effectiveness against low frequency pulse magnetic field. However, the shielding effectiveness is low. Under the experiment condition, the shielding effectiveness is below 7dB of signal reinforcement net. Larger the diameter, smaller the mesh size, full welding and colligation, higher the shielding effectiveness will be. The reinforcement nets with dissimilar forms can be comprehensively utilized in protecting constructions for the best result.

The increasing complexity of the electromagnetic environment poses a threat to the reliability of intelligent connected vehicles. Carrying out performance verification of intelligent connected vehicles in different electromagnetic scenes plays an important role in ensuring the safety of vehicles. A method for constructing electromagnetic scenes in the 10m semi-anechoic chamber is presented in the paper. By constructing the electromagnetic environment collection system, the main technical indicators of the electromagnetic environment collection related to the intelligent connected vehicles are analyzed, and the construction scheme of the electromagnetic environment scene library is studied. Based on the electromagnetic environment simulation system, the effectiveness of the electromagnetic scene reproduction is verified from the two aspects of test result comparison and signal comparison, which provides support for the test and verification of the electromagnetic compatibility performance of the intelligent connected vehicles in the laboratory.

Infrared(IR)photo detectors based on InAs/GaSb type II superlattice have developed quickly in recent years. Many groups show great interest in InAs/GaSb superlattice detector for its superiors as high quantum efficient, high working temperature, high uniformity and low dark current densities. Inductively coupled plasma(ICP) etching of GaSb and InAs/ GaSb superlattices were performed using Cl₂/Ar/CH₄/H₂. This paper introduceste inductively coupled plasma( ICP) etching of Inas/GaSb with SiO₂ mask by the Cl₂/Ar/CH₄/H₂ mixed-gas process. The effects of process parameters such as gas combination, ICP and RF power on the etch rate and quality of InAs/GaSb It is found that the ratio of Cl₂ flow rate significantly affects the etch rate, due to the trade-off between physical and chemical component of etching. The etch rate of InAs/GaSb increases with the increase of percent of Cl₂, there will InClx remains in the etching channel when the etching depth exceeded 2μm, which can stop the etching going on. This phenomenon can be eliminated by decrease the Cl₂ ratio,to make sure the etching depth reach 6μm under a certain low etching rate. The surface morphology and SEM of the superlattice material after etching shows that dry etching morphology is better than wet etching.After the electrode is grown, the superlattice chip have a good diode characteristic curve.

The Electrical Fast Transient (EFT) interference may affect the performance of the device or even cause the device failure. To study the anti-EFT interference performance of integrated circuit (IC), an EFT test platform of a SoC MCU chip has been set up. And the power pin of MCU integrated circuit shows better anti-interference performance than that of interface pin. Since I/O failure is mainly caused by MOSFET failure in clamp circuit, OBIRCH, SEM and other means are used to locate the failure position and accurately analyze the failure points. And the failure mechanism of the MOSFET in the I/O circuit has been analyzed according to the observation results. The MOSFET in the I/O circuit is fused under the positive feedback effect of the decrease of resistivity and the increase of current density, which leads to the I/O failure. The failure mechanism of MOSFET under high speed pulse is also verified by simulation. The results show that the only hot spot, firstly reaching the silicon melting point due to heat deposition originating from Electromagnetic Pulse (EMP) injection, is at the drain-substrate PN junction. The device fails or burns out with thermally damages due to electrical heating coupling.

Publisher’s Note: This paper, originally published on 5 November 2020 was withdrawn on 13 November 2020 per author request.

Electrically tunable metamaterial is attracting extensive attention recently for its potential in electromagnetic compatibility (EMC) and radar scattering manipulation, while still remains a great deal of challenges in its tunability. In this paper, a conductivity tunable graphene-ion gel-graphene (GIG) sandwich structure is fabricated and measured to investigate the properties of ion gel, showing its advantages over ion liquid on flexibility and tunability. Besides, a frequency and amplitude dual-tunable metamaterial absorber based on the GIG sandwich structure is proposed, modeled and optimized. From the simulated response, the absorber can achieve the tuning of absorption frequency from 5.4 GHz to 7.1 GHz and the reflection minimum from -15 dB to -45 dB. With the introduction of tunability from varactor diodes and graphene, the working frequency and absorption can be almost independently tuned. This work gives a meaningful reference for graphene based tunable metamaterial based on solid ion gel, especially in conformal and outdoor environment applications.

In the actual maintenance support application, it is difficult to carry out maintenance work according to the actual environment and conditions of the equipment to be repaired, which leads to the inaccurate timing of the equipment and the low efficiency of the equipment use. In order to study the relationship between the time interval of equipment repair and the geographical environment of the equipment more deeply, a RCNN deep learning model is proposed for training and feature extraction of equipment maintenance service information and geographical environment information, which can also predict the equipment maintenance interval. It also helps to compare among different prediction models and verify the effectiveness of the model, which further provides a method for the optimization of equipment maintenance interval.

The classification of Hyperspectral images (HSIs) has been the focus of many recent research efforts, where feature extraction plays an important role. Discriminative feature extraction methods aim to reduce the data dimension of HSIs, retain effective image information to the greatest extent, and suppress noises at the same time. Besides, according to the characteristics of pixel-by-pixel-multi-band of HSIs and data redundancy between bands, the processing of HSIs in the classifier will bring huge computational overhead. In this paper, we present a parallel implementation of the improved noise adaptive principal component algorithm (INAPC) for feature extraction of hyperspectral images on commodity graphics processing units (GPUs). Aiming at maximizing the signal-to-noise ratio (SNR) instead of the variance, we firstly deploy two SVDs and more comprehensive noise estimation in the INAPC transform and constructed a complete feature extraction process. Then we deploy a complete CPU-GPU collaborative computing solution, and use several GPU programming optimization methods to achieve the maximum acceleration effect. Through the experiments on three real hyperspectral datasets, Experimental results show that the proposed INAPC has stable superiority and provides a significant speedup compared to the CPU implementation.

A baffle effectively prevents direct rays of stray light from entering an optical system. Existing design methods can result in very long outer baffle lengths, increasing the weight and volume of the optical system, or large secondary mirror baffle sizes, increasing the central obscuration and resulting in the reduction of the illumination beam on the image. To overcome these challenges, a design method for a built-in baffle is presented. The designed built-in baffle for a typical Ritchey-Chretien optical system is compared with a conventionally designed baffle. The resulting optical system’s outer baffle length is shortened, the central obscuration is reduced, and the stray light suppression capability is good.

We propose that Deep Learning (DL) can be used to improve the performance of nonlinear structured illumination microscopy(NSIM) to enable it to reconstruct a super-resolution image with much less raw image frames. This allows for gentler super-resolution imaging at higher speeds and weakens phototoxicity in the NSIM imaging process. We validate our approach by super-resolution image reconstruction of simulated obtained data.

Hyperspectral remote sensing technology has made good progress in recent years and is often used in military and civil fields. Hyperspectral images(HIS) are three-dimensional data composed of two dimensional spatial information and one dimensional spectral information of ground objects, which can be used to classify and study HIS. The current research is generally based on closed sets, that is, the classes appearing in the testing time all appear during the training time. However, the setting of this closed set is difficult to achieve in the real situation, so the concept of open sets is introduced, that is, unknown targets that may appear in the testing time but do not appear during the training time. However, there are few effective algorithms to study the open set of HIS. To solve this problem, we propose an open set recognition method for HIS based on Extreme Value Machine(EVM). The pre-processed HIS data was used as the input of EVM algorithm and the EVM model based on Weibull distribution was established. The test data were used to detect the classification of unknown targets and known targets. Compared with other classification algorithms, EVM can classify known targets in HIS and detect unknown targets with good accuracy.

In order to precisely obtain the relative distance between probe and drogue during unmanned aerial vehicle (UAV) autonomous aerial refueling docking, a monocular vision method based on the geometric model for identifying refueling drogue and measuring the relative distance between probe and drogue is presented. In this method, firstly, using edge detection algorithm to get edge segment of image including refueling drogue, and connecting the line segments with the same deflection direction into circular arcs. Secondly, fitting circular arcs with similar radius and center into candidate circles, and validating candidate circles to obtain true circles. Thirdly, detecting lines in the true circles, and the drogue recognition is completed according to the characteristics of the drogue model. Finally, according to the known information of the drogue and the recognition drogue,the relative distance between probe and drogue can be measured. Especially,the method can be used to detect drogue accurately and has good performances in images when there are interfering objects. The experiment results show that the method has high real-time performance and high measurement accuracy, can fulfill the demand of probe and drogue aerial refueling.

Based on the user's interest in acquiring information on the status data of household appliances, a knowledge discovery algorithm combined with optical sensors for automatic control is proposed. Collecting data through optical sensors, it is analysed on the relationship between functions, status data, and family events, using fuzzy theory to convert household appliances’ status values into linguistic variables. Through the knowledge expression method of the script, the family events that the user is interested in are related to the language variables, and on this basis, the reasoning of family events is realized. Simulation experiments show that the algorithm can obtain relevant information through the optical sensor and sample to the state value of the household appliances, and realize the reprocessing of the acquired information according to the user's interest, which can solve the problems from the state of the household appliance to the family event and its inference.

Remote sensing image object detection is the primary task in the field of intelligent processing and has important practical application value. However, the current intelligent processing method of remote sensing images is difficult to meet the real-time requirements. In order to improve the effectiveness, real-time on-board processing of the collected images has become an important direction. This article compares several commonly used deep learning object detection algorithms, selects YOLO v3 for in-depth research and optimization, and transplants the debugged algorithm to the Cambrian 1H8 embedded edge intelligent computing platform for performance testing. Experiments show that the algorithm has high accuracy and the running speed basically meets the real-time requirements. It can be used to study and test the real-time processing performance of remote sensing images.

Transparent electromagnetic interference (EMI) shielding film has a wide range of applications in the optoelectronic and display systems in military, aerospace and other fields, which has attracted widespread attentions. In this paper, the electrohydrodynamic jet printing method was introduced into the additive preparation of transparent conductive films, and a batch of flexible transparent EMI shielding films with good EMI shielding effectiveness and low cost were achieved in a simple and easy way. We utilized electrohydrodynamic jet printing technique to directly print a grid-like micro structure on a flexible and transparent substrate, and a transparent EMI shielding film with a line width of 13.3 μm, period of 300 μm, and light transmittance of 73.2% was obtained. The EMI shielding effectiveness is about -18 dB in X band. In addition, the printing parameters was optimized to obtain transparent EMI shielding films that meet different performance requirements. The transparent EMI shielding film prepared by printing method requires no high-resolution mask and exposure in traditional microfabrication, and is quite low cost and convenient to adjust parameters digitally. This work proves that the additive electrohydrodynamic jet printing perfectly fits the transparent film research in stage of sample design and prototype.

When the human visual system processes the optical image, it will use the salient features of the target and transfer learning to perform deep processing to improve the accuracy of target recognition.The vision system can quickly transfer the existing experience to interpret the target image information, so that humans can recognize the external environment. This mechanism of applying a priori knowledge for image processing of new targets is transfer learning.This image processing method can be applied to the field of high-resolution remote sensing image processing.This paper proposes to use the transfer learning of remote sensing images and utilize the VGG model to expand the data to improve the accuracy of remote sensing image recognition tasks.In terms of innovative methods, it is proposed to use the training samples processed by the style transfer algorithm as the input of the convolutional neural network classification model, which can guide the model to learn more prior knowledge information than the data.In the algorithm design, the network parameters such as the training input, network structure, matrix expansion method and convolution kernel scale of the traditional VGG network were adjusted and optimized.The recognition accuracy experiments were conducted using CIFAR10 and DOTA datasets, and Google Earth was used.The results of random remote sensing images are verified.

Hexagon bolts are common standard parts in manufacturing industry and generally accompanied with anti-loosing components like lockwire and locking tab, to strengthen fastening effect. However, incorrect assembly will cause failure, even accelerate bolt loosening. Consequently, misassembly inspection is necessary. For an automatic inspection system based on computer vision, the localization information of hexagon bolts centers and corners is needed. Object detection and keypoints detection technologies develop rapidly, and some researches have applied these technologies to hexagon bolts and their keypoints detection. These existing methods use the top-down strategy, which need two separate models for each stage. This strategy is common and mature, but still has some shortcomings. First, detection models suffer from the imbalance between positive and negative samples. Second, the quality of bounding box detection will affect keypoints detection. Third, two separate models cause computation redundancy. To overcome these problems, we propose a novel bottom-up keypoints based hexagon bolts detection method. With the small dataset, relatively high-resolution images and limited computation resources, our method achieves a good balance between speed and accuracy. We build two hexagon bolts datasets, one contains locking tab with large field of view (LFOV), another contains lockwire with small field of view (SFOV). Images with large field of view have more complex background. We test our model on both two datasets, and the results show that our method is effective. For the SFOV, our model reaches 0.950 AP and 0.974 recall. For the LFOV, our model reaches 0.883 AP and 0.934 recall.

Fiber bundle have been widely used in biomedical endoscopy areas because of its flexibility and high spatial resolution. However, due to the irregular layout of the core and surrounding cladding, the captured fiber bundle images are affected by repeated honeycomb-like patterns. In this paper, we investigate the conditional generative adversarial network to reconstruct the optical fiber endoscope image. In order to improve the operation efficiency and reduce the burden of generative network, we use traditional method to obtain the fiber location map, as an additional condition to generate the network together with the input image. By injecting the position information of fibers, the generative network will pay more attention to the fiber regions and the surrounding structures, and the discriminative network will be able to assess the local consistency of the restored regions. In order to obtain more extensive contextual information, we apply multi-scale loss on the decoder side of the autoencoder. Each of these losses compares the difference between the output of the convolutional layer and the ground truth value that has been downscaled accordingly. Aside from the multi-scale losses, which are based on a pixel-by-pixel operation, we also add a perceptual loss that measures the global discrepancy between the features of the autoencoder’s output and those of the corresponding ground-truth clean image. Our experiments show the effectiveness of our approach which can effectively remove the honeycomb-like patterns and retain the original image features.

In the international environment of violence, terrorist attacks and illegal drug smuggling, security issues are particularly important. Many companies and institutions in the world have successively developed safety inspection equipment, among which x-ray safety inspection equipment has been widely used. In order to ensure the safety of passengers, it is necessary to be able to accurately identify the knives, guns, inflammables, explosives and other dangerous articles in the luggage package during the security check, so as to reduce the probability of danger. At present, airport security inspectors need to change for half an hour, which is hard to work and easy to miss. An assistant identification system of contraband based on yolov4 to assist the security inspector to judge the X-ray image was proposed to improve the efficiency and accuracy of security inspection identification, reduce the manual detection intervention of professional training personnel, and avoid the occurrence of missed detection and false detection. A novel data augmentation method is proposed to guarantee the performance of the system in the case of a small number of samples. The experimental results show that the system has strong robustness in different channel directions and complex scenes, the comprehensive detection rate of lighter is higher than 95%, and the recognition efficiency and accuracy are greatly improved compared with the traditional convolutional neural network(CNN).

As an expression of three-dimensional information, depth map provides more possibilities for many computer vision applications, which puts forward higher quality requirements for depth maps. However, the spatial resolution of depth map is always low, which limits its potential. In this paper, we present a depth map super-resolution based on joint convolutional neural network (J-CNN). The network combines RGB image to guide the reconstruction of low-resolution depth map. Our model consists of three subnets. Subnet 1 uses simple convolutional layers to extract rough features of depth maps. Subnet 2 and 3 use a progressive up and progressive down sampling network structure. This structure can greatly expand the receptive fields and help to extract more fine features of the images at different scales. We use it to extract the complex features in the depth map and RGB image. Finally, convolutional layers connect three subnets to transfer useful information from RGB image to depth map. The final up-sampling reconstructed operation is realized by sub-pixel convolution, effectively avoiding the “checkerboard effect”. Our J-CNN is evaluated on Middlebury datasets which shows improved performance compared with six advanced methods. Our model also shows strong robustness under large sampling factor (16 times).

Conventional digital holography (DH) technique largely limited by the effect of random scattering media in the imaging path, which causes great challenges for its applications in vivo imaging. As an improvement, short-coherence digital holography (SCDH) uses a low-coherence light source (near-infrared (NIR) region), where the absorption of light is at a minimum, to enhance its ability to resist scattering. However, SCDH also fails under strong scattering conditions. Here we propose to use deep learning (DL) for SCDH, and the results show that an image of a target behind a 2.30 mm chicken breast tissue can be reconstructed successfully. We experimentally demonstrate that DL-based SCDH can be used to reconstruct the object from a single measurement under some hard conditions, for example, when there is strong static or dynamic scattering media in the imaging path.

The key to improve the capability of photoelectric detection and countermeasure system is to improve the detection ability of the system. System resolution, detection distance, detection range, response time, system signal-to-noise ratio are important criteria, and system optics is the core of its engineering implementation. From the global perspective, the adaptive algorithm establishes a new theory which superior to the traditional method, and solves the problem of freely determining the core optical solution, and opens up a new field for the development of accurate detection system.

Due to the limitation of scanning conditions and the emerging clinical application of local imaging, local tomography has become a research hotspot. Traditionally, completing the projection data through interpolation or spatial transformation iteration is popular to overcome the truncation artifact, such as iterative reconstruction algorithms. Recently, deep learning networks have been incorporated in dealing with such problems. Instead of global filtering on truncated data, the proposed work focuses on the full feature extraction using U-net as well as the usage of the redundancy between projection sinogram, by performing extrapolation of truncated projection data through data learning and then filter back projection local reconstruction with high efficiency. During the learning process, 3439 projections are selected as complete projection data , and the corresponding truncated data is simulated according to the actual truncation situation. Then, 150 of the truncated data are randomly selected as the test samples, while the rest 3289 of those as the training samples in the U-net. The output sinogram data is compared with the original complete data by calculating the L2 loss function of both. And the Adam optimizer is used to continuously optimize the parameters of the network. RMSE and NMAD are used to quantitatively evaluate the reconstruction effect. Experimental results show that the proposed method based on truncated data extrapolation network can obviously suppress the ring artifacts and compared with images directly reconstructed using truncated projection data, the RMSE is reduced by an average of 43.185%, and the NMAD is reduced by 44.24%.

Low dose computed tomography (LDCT) has attracted considerable interest in medical imaging fields. Reducing tube current intensity and decrease the exposure time are the two main ways in clinic applications. Nevertheless, the resulting statistical noise will seriously degrade CT image quality for diagnosis. To make full use of the original projection data as well as further improving the small dataset processing ability of U-net, this study aimed to investigate a low dose X-ray CT image denoising method via U-net in projection domain (PDI U-net). Meanwhile, in view of avoiding the excessive smoothing of the small structures, the inception module is introduced in the encoding stage of the network. And different convolution kernel operations of 1×1, 3×3, and 5×5 are used in parallel to obtain multi-scale image features, increasing the depth and width of the network while reducing the parameters. Furthermore, the shortcut connection is utilized to transfer the low-level area local detail information to the high-level area. By merging it with the global information of the high-level area, the proposed network can maintain the image details while de-noising. The experimental results show that the method proposed can significantly improve the image quality with clear feature edges and close visual appearance to the reference high dose CT images. Compared with LDCT and Residual Encoder-Decoder Convolutional Neural Network (RED-CNN), the peak signal to noise radio (PSNR) is improved by 9.02dB and 2.74dB, and the structural similarity (SSIM) is improved by 0.43 and 0.07, respectively.

We introduced a Python-based design framework for manufacture-friendly optical ANN. It provides the cross-level interoperability between the photonic circuit chip layout and the neural networks infrastructure to enable the optical ANN with better tolerance to the device-by-device or chip-by-chip deviation. It allows a wide range of abstract on levels to describe the behavior of optical ANN: from the lowest-level functionality of manipulating the properties and arrangement of individual phase shifters on a photonic circuit chip, to the highest-level features of designing optical ANN via PyTorchlike development-library as well as its optimization with the well-established machine learning algorithms such as backpropagation. On all the levels, the physical design of the photonic circuit chip can be integrated and synchronized with the construction of the neural networks accounting the influences of the fabrication-deviation with the assistance of IPKISS, a Python-based tool for photonic circuit design. As a demonstration, we use our framework to design the LeNet-5 networks, which can be executed on the photonic circuit chip with non-uniformed grating coupling efficiency. Our LeNet-5 networks achieves the precision around 97.5% for MNIST task.