On the basis of our study of 55 samples collected during four cruises in the Yellow and Bohai Seas off the east coast of China (November 2014, August 2015, July 2016, and January 2017), we developed an algorithm for estimating the sea-surface density (SSD) using remote-sensing reflectance. Our results show that the multivariate linear regression model performs the best, with a determination coefficient of 0.70 and a mean absolute percentage error of SMAPE=3.49%. We used an independent dataset (27 in situ observations) to assess the performance of the model, yielding a validation result of SMAPE=3.27%. In addition, the sensitivity experiment of the model show that the observed fluctuation in the SMAPE values is <3%, indicating that our proposed model is relatively stable. Meanwhile, we applied our developed model to the geostationary ocean color imager (GOCI) satellite data recorded in July 2016 and successfully produced the SSD distribution pattern. The spatial characteristics show that the coastal waters, the central parts of the Bohai and Yellow Seas, and the waters off the northern Shandong Peninsula have relatively high SSD values, while relatively low values are distributed along the Qingdao coast.

Herein, the effect of laser pulse time delay induced by seawater scattering is studied using the Monte Carlo numerical simulation method. Further, we analyze the effects of seawater types, transmission distance, and transceiver parameters on laser pulse time delay. The numerical simulation results indicate that in clear seas, the laser pulse time delay broadening does not obviously change with the increase of transmission distance. Furthermore, when the transmission distance is less than 50 m, the channel time delay is smaller than 0.5 ns, indicating the transceiver parameters exhibit only small impact on the channel delay. However, in turbid seas, the receiving power and the channel time delay increase with the increase of receiving aperture under the multipath effect induced by channel scattering. When the receiving angular field of view (AFOV) is less than 90°, its influence on the receiving power and the time delay broadening is observed to be significant. However, when the AFOV is between 90° and 180°, the receiving power and the time delay broadening demonstrate no obvious changes with AFOV.

Herein, a mathematical model of the relationship between grating rotation angle and diffraction beam deflection angle was established based on the theory of mosaic gratings. An optical path was designed to detect the grating rotation angle using interferometry. A Zygo interferometer was used to detect the zero- and first-order diffraction wavefronts of a grating. In addition, the model error caused by the reflector installation error was studied. The experimental results show that in the range of 0-500 μrad, the maximum relative measurement error of grid line parallelism between gratings is 3.85% when the theoretical model is used to adjust the grating attitude. When pitch and yaw angles of the reflector vary in the range of 0″-20″, the maximum relative measurement errors of grid line parallelism between gratings are 4.99% and 3.77%, respectively.

A high-power fiber laser is limited by the nonlinearity of a single-fiber output. Therefore, a beam array must be formed through multiple fiber lasers that output a single focused beam. A high-power high-quality beam output must be ensured after a long-distance transmission through atmospheric turbulence. This study uses an array containing 19 fiber laser beams to design a coherent array synthesis system. The effects of beam truncation factor, angle jitter, and turbulence propagation on coherent synthetic beams are quantitatively analyzed, and the coherent propagation process and their propagation characteristics in free space are simulated to provide references for the establishment of a practical coherent laser array system.

The influence of the high-order Fourier components of an ideal triangular microwave signal on the similarity of the triangular signals is studied here. Using the numerical simulation method, the similarity between the original ideal waveform and the simulated waveform with different orders is studied. The numerical simulation results show that the similarity between the triangular photonic microwave signal and the ideal triangular microwave can reach 0.989 when the Fourier components of the triangular photonic microwave signal are retained to the third order. A scheme based on the generation of triangular microwave signals using a dual-parallel Mach-Zehnder modulator is presented. By controlling the six parameters of the system, a triangular photonic microwave signal can be obtained when the power ratio between the fundamental frequency and the third-order harmonics of the generated signal is equal to that between the first- and third-order Fourier components of the ideal triangular waveform. The experimental results show that the system can generate triangular photonic microwave signals with a repetition rate of 6 GHz.

Herein, we studied the mode characteristics of a few-mode fiber with an elliptical-core. We investigated the general expression of the Kerr nonlinear coupled-mode equations for the few-mode fiber and discussed the design process of an optical fiber parametric amplifier in detail. The dependence of the optical field mode distribution on core ellipticity was simulated using the COMSOL software, following which the phase mismatch factor in four-wave mixing and the corresponding mode field overlapping integral (or nonlinear coefficient) were calculated. In addition, we studied the degenerate four-wave mixing process of three mode groups in the elliptical-core few-mode fiber with a step refractive index distribution. The results show that when the LP21a-mode pump power is 0.5 W, the maximum gain of 20.15 dB can be obtained and the elliptical-core tolerance size is 0.6 μm in the gain deterioration range of 2 dB.

Based on the direct-current-biased optical orthogonal frequency division multiplexing (DCO-OFDM) combined with code division multiple access (CDMA), a DCO-MC-CDMA system is built. When the Gold code and m-sequence are selected as spread spectrum address codes in the proposed system, their cross-correlation is worse than that of orthogonal codes owing to the limitation of the system structure, leading to a greater multi-user interference (MUI). The maximum ratio combining, equal gain combining, and orthogonal recovery combining (ORC) are used to derive the signal-to-noise ratio expression under the influences of additive white Gaussian noise, MUI, and limiting noise. The simulation model of the Monte Carlo bit error rate (BER) in MATLAB is established. The research results show that with the increase of the number of users, the MUI makes the system BER performance worse. In contrast, the ORC eliminates the MUI induced by the difference between subchannel gains, and thus the BER performance is the best. The BER performance of the DCO-MC-CDMA system with the Gold code is better, because the cross-correlation of the Gold code is better than that of the m-sequence in

The advantages, high bandwidth and high speed, of optical Fourier transform have been proved to be highlighted since the frequency of the objective signal becomes increasingly complicated, further, it has significant applications in the fields of optical spectra and real-time detection of high-speed radio frequency signals. The optical Fourier transform based on fiber dispersion comprises two particular methods: dispersive pulse broadening and time lens. Based on the optical fiber dispersion method, the realization principles of optical Fourier transform based on dispersion broadening and time lens are compared and analyzed. In addition, it summarizes five representative approaches divided into two categories, analyzes the empirical performance including output, dispersion, spectral resolution, time-bandwidth product, and expounds the applicability and improvement of these approaches.

The training sample size for some objects on the ground is quite small when applying a deep learning model to study the classification of remote sensing images. Meanwhile, diversified remote sensing image acquisition methods generate numerous multimodal remote sensing images with different spatial resolutions. Fusing these multi-modal remote sensing images to remedy the small sample size defect and achieve a highly precise classification of remote sensing images is an urgent problem to be solved. To this end, the present study proposes a fusion method for image classification based on the correlation of two spatial resolutions. A deep learning network is utilized to extract the high-level features of the remote sensing images in two spatial resolutions. Two types of high-level features are integrated via the proposed fusion strategy and further used as the input to train the whole network model. The experimental results demonstrate that the proposed fusion algorithm can achieve high classification accuracy. Further, because different fusion rules have different classification accuracies, a suitable selection can improve the classification accuracy.

A nonlinear diffusion-filtering based feature detection algorithm was proposed with the scale space constructed by nonlinear diffusion-filtering, in which the ability of Hessian matrix weak edge detection and that of Laplace operator strong edge detection was combined, and the ratio of the Hessian matrix determinant to the Laplace operator was set as feature detection criterion. The performance of the proposed algorithm was validated with the images from simulated points and lines, Mikolajczyk standard database and the real scenes through SALSA polarization camera. The results demonstrated that the proposed algorithm can detect the strong and weak edges and corner points of an image with good robustness less affected by bad illumination or low contrast.

This paper proposes a new quality assessment method without reference images based on the convolutional neural network (CNN) and deep regression forest. First, this method performs a local contrast normalization on the original images. Second, it subsequently uses CNN to extract the discriminant features of the image quality. Finally, it utilizes the deep regression forest to predict the image quality. The method does not require any manual features, which simplifies the process of image preprocessing. In addition, fewer convolution layers are beneficial to reduce the training time of the network. The application of deep strategy to integrate the regression forests improves the prediction accuracy of a single forest. On the LIVE and TID2008 databases, the experimental results show that the proposed method can predict the image quality well and has a good generalization performance with high accuracy. The proposed method achieves a state-of-the-art performance, especially in JPEG2000, Gaussian blur and white noise distortions.

In this paper, we propose a simplified method of feature-aware for a three-dimensional point cloud. First, the k-nearest neighbor points of each point are searched by constructing an octree, and the normal vector of each point is calculated to detect and preserve the edge points. Then, the expectation maximization algorithm is utilized to cluster the point clouds and determine the points with high curvature. Finally, these point clouds are simplified by a method which utilizes the edge-aware directed Hausdorff distance, the above point clouds are merged, the duplicate points are deleted, and thus, the model is simplified. The proposed method is suitable for the models with different curvature changes, and it can display the overall contour of the model while retaining the sharp features. The experimental results show that the proposed method not only preserves the geometric features and contour appearance of the original model, but also effectively avoids the hole phenomenon in the simplification process. The geometric simplification error of the method is considerably low.

A single-image defogging algorithm based on deep learning is proposed. The convolutional neural network achieves defogging by learning the mapping relationship among the YUV(Y is luminance, UV is chrominance) channels of the foggy and clear images. The network structure comprises two identical feature modules, which mainly include multi-scale convolution, convolution and skip-connection frameworks. The experimental results show that the proposed algorithm can be used to restore images with high resolution and high contrast, regardless of the datasets with synthetic or natural fog images. Furthermore, a comparative evaluation of this algorithm with the existing algorithms confirms its superior performance both subjectively and objectively.

The traditional hyperspectral unmixing methods only consider the geological properties of hyperspectral images or the sparse properties of abundance and neglect the spectral spatial information of hyperspectral data. Thus when the pure pixels are missing, the unmixing accuracy is significantly reduced. In order to overcome these limitations, an improved spatial information constrained nonnegative matrix factorization method for unmixing is proposed. This method fully uses the spatial information and the sparse properties of hyperspectral images, and thus the properties of the traditional nonnegative matrix factorization methods are improved. Both the synthetic simulation images and the experimental results show that the proposed method has overcome the noise-sensitivity and the dependence on pure pixels of the traditional methods.

To solve the problem that hyperspectral image data has many dimensions and it is difficult to extract spectral information and spatial information, a classification algorithm is proposed based on a hypergraph and a convolutional neural network. In this algorithm, the hypergraph is first constructed based on the spectral and spatial relationships among pixels in a hyperspectral image, and then a sample with spectral space joint features is constructed through this hypergraph, which is finally sent to the convolutional neural network for feature extraction and thus the classification is finally achieved. The experiment is performed on three most commonly used hyperspectral datasets and an overall classification accuracy of 96.63% on the Indian Pines dataset is achieved. Compared with other algorithms, the proposed algorithm has a high classification accuracy and a high speed, which avoids the instability in feature extraction and fusion by traditional methods. It is verified that the spectral space joint information extracted by the proposed algorithm has a strong feature expression of hyperspectral images.

Optical coherence tomography (OCT) is a non-contact imaging modality with advantages of high resolution and high acquisition speed. OCT can provide three-dimensional image structures and detect motion information in the samples. In this study, a high-resolution spectral OCT angiography system was constructed with the lateral and axial resolutions of about 6.7 μm and about 4.7 μm, respectively. The line scan camera was operated at 140 kHz and could complete a three-dimensional scan within 2 s. Based on this, the blood flow perfusion information was extracted by differentiating the adjacent B-scan images. A micro-vasculature network image of rat cerebral cortex was obtained. The experimental results demonstrate that the proposed system is suitable for capillary imaging.

In this study, the Ni-based-alloy coatings are prepared on 45 steel substrates by laser cladding with coaxial powder feeding. The fracture morphology analysis is conducted to evaluate the fracture mode and types of cracks. In addition, the energy-disperse spectroscopy is used to further investigate the mechanisms of crack initiation and propagation. Furthermore, the effects of the laser-cladding-process parameters on the cracking sensitivity are studied. The results show that the cracks in the Ni-based-alloy coatings are due to the quasi-cleavage fractures, which are a type of brittle fracture. Moreover, it is shown that the thermal residual stress causes the crack initiation at the interface and the subsequent crack propagation to the coating surface. It is further shown that the process parameters have significant impacts on the cracking sensitivity of the coating because the uneven distribution of the hard phase (including boride and carbide) and the brittle eutectic organization result in a high cracking sensitivity. Finally, the basic strategies for controlling cracks during laser cladding of Ni-based-alloy coatings are proposed.

We propose a calibration method of the linear cameras used for measuring lithium battery coatings (LBC). First, we analyzed the LBC production system and established a vision measurement system using linear cameras. We then studied the linear camera calibration technology and developed a single-dimensional calibration method based on the LBC production characteristics and the detection requirements. We hence simplified the calibration method and improved the calibration precision. Finally, we developed a corresponding software system based on the technology in use and applied the system for the production practices. The practice results demonstrate that the proposed calibration method meets the LBC measurement requirements in an actual production environment.

Based on the known light intensity distribution of a light source and indoor spatial structure, this study established a mathematical model and adjusted the relevant parameters through experiments with respect to a hypothesis about the bidirectional reflectance distribution function (BRDF) of the spatial surface. Further, the application of this mathematical model in different types of spatial models results in an average illuminance error of less than 3%. Moreover, the plane illumination curve distribution is basically similar to those obtained using the existing methods. The working-surface related parameter values are compared with the DIALux simulation results. This study also improves the BRDF model of a wall surface in terms of the application of the uneven light energy distribution on the reflective part. The proposed method improves the illuminance calculation accuracy in a circular bottom space or a space with its bottom not equal to height where a direct calculation is not available.

This study proposes a simple and high-precision binocular ranging method. A stereo correction algorithm was first used for the stereo correction of the non-forward parallel structures of the left and right cameras. The improved Census transform algorithm was then applied to obtain accurate disparity values. Finally, the true distance information was calculated based on the special epipolar-line geometry of binocular vision. Further, the multi-center points were used to compare with the surrounding pixels in the original Census transform, which is improved as mutual supervision and fusing of multi-center points. Thus the accuracy of the stereo matching is improved. Two identical complementary metal-oxide-semiconductor (CMOS) cameras were used to build a binocular ranging platform, and the hardware, algorithm and calibration process in the flow chart of ranging were introduced in detail. The experimental results show that the proposed method performs better than the original Census transform. For a 2 m measurement, the accuracy is increased by 19.1% and the measurement error is 6.4 cm, thus meeting the requirements of high-precision binocular ranging.

In this study, we presented the scheme of a segmented pose measuring system as well as its implementation. Further, we presented a linear three-dimensional coordinate measurement model for a far-segmented measuring subsystem by applying an imaging equation of the line cameras equipped with cylindrical lenses and calibrated the extrinsic parameters of each line camera by minimizing the object space errors. We proposed an iterative pose optimization method based on a linear initial solution to solve the problems related to monocular pose determination for a near-segmented measuring subsystem. The experimental results demonstrate that the proposed system can achieve a relative position measurement accuracy of less than 0.5% and a root mean square error of less than 40 mm in the vision range of 5-35 m. Moreover, the system can realize an attitude tracking accuracy of less than 1° in the vision range within 5 m, satisfying the large vision range requirement of pose measurement.

This study proposes a new deep neural network based camera calibration method that achieves flexible, high-precision calibration in complex environments, without having to classify or extract features from input data. By optimizing the network structure, hyperparameters, and training algorithms, the deep neural network can be quickly and effectively trained. The experimental results confirm that, compared with Zhang's calibration method and the shallow neural network, the proposed method can achieve high calibration accuracy under a wide range of imaging conditions involving multiple shooting angles or high distortion. For the images produced using a highly distorted lens, the proposed method achieves an average calibration error of only 0.1471 mm over the calibration range of 633 mm×763 mm.

This study proposes a fast and accurate new edge-extraction method based on the Gaussian map clustering algorithm. First, the normals in the Gaussian sphere are clustered via agglomerative clustering and normal estimation. Then, the covariance matrix eigenvalues of the nearest neighbors of each point are analyzed to detect the edge features. The edge-extraction experiments are performed on different pointcloud objects, and the edge extraction effects and the extraction time are compared and analyzed. The experimental results indicate that the proposed method can quickly and efficiently extract the edge features from point clouds and its performance is improved compared with the edge-extraction algorithm based on original Gaussian map clustering.

An eigen-equation method under tight binding approximation is developed to find the plasmon of two-dimensional quantum dots. The results show that the plasmon charges may show monopolar, dipolar, or quadrupolar polariztion in the transverse or longitudinal direction, respectively, which may localize at the center, end, or boundary of quantum dots. According to the charge-polarization and charge-localization characteristics, it is assumed that there exist 24 or 27 plasmon modes in a two-dimensional uncharged or charged system, respectively. In addition, the size effect of various plasmon modes may differ, e.g., the plasmon frequency of the monopole-dipole end mode shifts towards the red end of the spectrum as the longitudinal size increases. In contrast, the frequency of the dipole-monopole end mode shifts towards the blue end of the spectrum as the longitudinal size increases.

In this study, a deep adversarial domain adaptation method is proposed for cross-domain classification in high-resolution remote sensing images. A deep convolutional neural network VGG16 is used to learn the deep features of scene images. The adversarial learning method is used to minimize the difference of feature distribution between source and target domains. RSI-CB256(Remote Sensing Image Classification Benchmark), NWPU-RESISC45(Northwestern Polytechnical University Remote Sensing Image Scene Classification)and AID(Aerial Image data set) are used as source domain datasets, and UC-Merced(University of California, Merced)and WHU-RS 19(Wuhan University Remote Sensing)are used as target domain datasets. The experimental results denote that the proposed method can improve the generalization ability of the model for target domain dataset without labels.

High power laser diodes (HPLDs) have been widely used in the defense sector, material processing, and pumping sources, considering their advantages such as high efficiency, long lifetime, small size, and low cost. This study describes the types and emission characteristics of the defects and the related advances in GaAs-based near infrared lasers and GaN-based blue-green lasers. By focusing on the commercial devices and using the condition-variable emission spectra for separated wavebands and the corresponding thermal imaging, the origination and spatial distribution of the emission signals related to the defects are determined. The internal catastrophic optical damage (COD) mechanism is also analyzed. Furthermore, the limitation of the current “external COD” model for interpreting the thermal evolution mechanism of the devices is pointed out.

The acquisition, pointing, and tracking (APT) technology is important to establish the reliable links in space laser communication. The conventional APT technology employs a mechanical rotation method to achieve steering control. In addition, the conventional APT technology exhibits some significant disadvantages, including large volume and rotational inertia as well as high power consumption. Further, the classical APT technology cannot satisfy the practical requirements of space laser communication, i.e., light weight, miniaturization, and low power consumption. Therefore, to study the non-mechanical beam scanning technology is of considerable significance. When compared with the conventional APT technology, an optical phased array is a promising non-mechanical beam scanning technique because it provides various advantages, including high scanning accuracy, random deflection, and strong stability. The liquid crystal optical phased array (LCOPA) is observed to develop rapidly. In this study, we briefly introduce the basic principle of an optical phased array for beam scanning. Subsequently, we review the status of current research about LCOPAs. Further, three key performance indicators, i.e., response time, deflection efficiency, and deflection accuracy, as well as the influence factors are analyzed. In addition, the methods used to improve the LCOPA performances are also summarized.

Laser frequency stabilization based on the spectral hole-burning effect in the cryogenic rare-earth-ion-doped crystal makes use of the spectral holes in the absorption of doped ions as the frequency reference. This technique has a low thermal noise limit. In comparison with the laser frequency stabilization technique based on the Fabry-Perot cavity, the proposed laser frequency stabilization technique is more insensitive to temperature, pressure, and acceleration, thereby featuring its viability to let frequency stabilization reach the theoretical limit of 10-17, which is comparable or even beyond with that of a Fabry-Perot cavity. The laser frequency stabilization technique based on spectral hole-burning effect is introduced from the aspects of theoretical principle, technical realization, and research progress at home and abroad. Furthermore, its development trend in the field of frequency stabilization is prospected.

In recent years, Raman spectroscopy is being used more frequently in biomedical analyses, especially for in-situ nondestructive detection of target samples or tissues. Raman probe, as the essential detection component in Raman spectroscopy, is also being developed for better diversity and functionality. Since Raman signal is extremely weak and noise interference is common, the new design and production of Raman probe are also extremely critical. This review provides an in-depth introduction to the existing general design and construction of Raman probe for biomedical applications, including the selection of fiber and filter/membranes, the design of probe tip, and the optimization of probe backend. Subsequently, the expansion and application of the Raman probe technology in the biomedical field are reviewed. The probes based on a combination of Raman and other spectra or imaging modes bring new vitality and prospects for broad applications in the biomedical and clinical detection of target samples or tissues.

Graphene is an excellent base material for a nanoelectromechanical system (NEMS) because of its extraordinary characteristics, including a large specific surface area, high Young's modulus, enormous stiffness, and low density. The graphene NEMS mainly studies the electromechanical properties of graphene films. Its resonant frequency has a very sensitive response to mass, force, and heat, which has a great application prospect in the sensing field. Meanwhile, the optical fiber sensor has been widely studied and applied, whose combination with graphene can possess a greater advantage. In this paper, the working principle, preparation process, sensing application of the graphene NEMS as well as the fiber-based graphene NEMS sensors are reviewed and prospected.

Based on a stack autoencoder (SAE) combined with an extreme learning machine (ELM), we built a deep neural-network prediction model, SAE-ELM. We initialized and fine-tuned the weights and thresholds of the deep neural networks using the spectral data extracted from the hyperspectral images of apples. Compared with the results of the traditional ELM model, the SAE-ELM determination coefficient of the prediction set increases from 0.7345 to 0.7703, the SAE-ELM residual prediction bias increases from 1.968 to 2.116, and the square root error of the prediction set decreases from 1.6297 to 1.2837. These research results show that the performance of the SAE-ELM model is superior to that of the traditional ELM model, and it is feasible for the proposed model to predict apple firmness.