To improve the imaging quality of the geometric waveguide, a method of measuring the parallelism of the waveguide array based on the off-axis autocollimation optical path was proposed. In this study, the error evaluation model of the parallelism of the semi-transparent and semi-reflective film array in the waveguide was analyzed according to the characteristics of the array plane. Combining the imaging principle of the geometric waveguide with the autocollimation optical path, the mathematical relationship of the array parallelism was deduced. Further, the parallelism measurement system of the waveguide array was developed. The system utilized a light source controller to control the output light intensity, which prevented the measurement from being affected by the light energy utilization of different semi-transparent and semi-reflective films in the waveguide. Subsequently, the Steger algorithm was used to process the autocollimation image. Finally, the calibration, array parallelism, and verification experiments were conducted. The experimental results show that the measurement uncertainty of the parallelism measurement system is 1.14″, and the maximum repeatability error is 0.32″. The proposed method can quickly and accurately measure the parallelism of the geometric waveguide array, which is of guiding significance to the attitude correction of the waveguide array and imaging quality improvement.

Quantitative research on energy and material exchange between the land or ocean and the atmosphere, especially the monitoring of carbon dioxide (CO2) exchange flux, play an important role in the study of global carbon cycle and climate change.We developed an open-path CO2 detection system based on direct absorption spectroscopy technology and derivative spectroscopy technology for on-line measurement across an order of km air path. A movable platform was developed capable of detecting atmospheric variations of CO2 in real time with the integrated optical system of transceiver.Allan variance analysis shows the detection limit of the system is 0.08×10-6 at the integration time of 100 s.The feasibility of the method with the second derivative spectrum is verified by using standard gases with different concentrations for the concentration calibration, and the correlation is 0.998. The continuous operation was carried out at Shenzhen Eco-environment Monitoring Station for 1 month, and the detection results have obvious daily variation periodicity.Comparing the data with the Licor7550-CO2 monitor installed at different points nearby, the data change trend is consistent, and the stabilityof the prototype is better.

Ultraviolet (UV) detectors based on the Surface Acoustic Wave (SAW) technology have become an active field of research owing to their advantages such as fast response, high sensitivity, simple fabrication, and wireless passive monitoring. In this study, the progress in the research of SAW UV detection, based on the classification of UV-sensitive films, was reviewed. The sensitive mechanism of SAW UV detection was analyzed and the characteristics, preparations, and UV performances of gallium nitride, doped gallium nitride, zinc oxide, doped zinc oxide, nanostructured zinc oxide, aluminum nitride, and other sensitive membrane materials were presented. This work also demonstrated the latest advancements in SAW UV sensors and their prospects in future development trends and challenges.

To analyze the fast penetration of a 3D droplet into a granular medium and its flow, an optical measuring system comprising a laser, fluorescence, and a high-speed camera has been developed. The system consists of three parts: the penetration segment, wherein droplets contact a medium, adding index-matched fluids to reduce light scattering; the optical signal acquisition segment, which captures the liquid morphology in the medium at different conditions and performs a quasi-3D measurement using laser sheets; and the data processing segment, which analyzes experimental data. The moving distance, speed, and acceleration of the droplet in diverse directions at different times are measured to describe such dynamic process. Moreover, the data can support the establishment and verification of the dynamic theory, comprising force, momentum, and energy. The spatial resolution of this system is 0.02 mm, and its time resolution is 0.5 ms. They can address requirements such as real time, strong stabilization, and high spatial and desired time resolutions.

To improve the sensitivity and cost-efficiency of a fiber bending sensor and to increase its linear range, a method based on a deep neural network was proposed to classify different bending angles and directions of plastic fiber. Plastic fiber with side throw sensitization processing was used to collect speckle images of different bending angles at the output end of the fiber. Data set one was made with five types of bending angle and data set two contained seven types of bending angle. After the pretreatment of image data, a multilayer convolution neural network was used to analyze the speckle image. The convolution and pooling provided speckle image features. A softmax classification was used for classification accuracy. Finally, the effect of two different convolutions on the classification of the neural network model was compared. The results show that the classification accuracy reaches 96% when the angle interval of fiber bending in the data set one is 5°. The theoretical and practical analysis results show that the scheme has a high recognition rate. Moreover, the realization of this method is expected to provide a new type of simple and efficient fiber bending sensor.

To ensure absolute calibration of the solar band of the Medium Resolution Spectral Imager (MERSI), a cross-calibration algorithm utilizing Sea-Viewing Wide Field-of-View Sensor (SeaWiFS) instrument over six Pseudo-invariant Targets (PTs) is developed and used to calibrate FY-3C＼MERSI data. The technique involves three key steps: Capturing the actual Bi-directional Reflectance Distribution Function (BRDF) characteristics using the well-calibrated SeaWiFS radiance data; Parameterizing the Spectral Band Adjustment Factors (SBAF) between SeaWiFS and MERSI sensor; and Cross-calibrating each screened MERSI observation based on the above two modules. The employed BRDF model can accurately predict ρTOA as observed by SeaWiFS with small biases (within approximately 3% for most cases). Furthermore, the fitted SBAF are nearly unbiased estimations of the simulated values. The time series of the cross-calibration result is very stable and exhibits no obvious trend, which is consistent with the characteristic of long-term stability and regular change of radiation signal of PTs. Hence, there is a systematic deviation from the calibration result of L1. This method fully considers and solves the uncertainty due to orbit drift, i.e., the difference between observed geometric conditions and spectral response in cross-calibration, and can effectively monitor and correct the radiation performance of satellites in orbit.

Extrinsic fiber Fabry-Perot Interferometer (EFPI) sensors are difficult to install and can be damaged easily in the internal insulation of power transformers. To address this issue, a high-sensitivity EFPI sensor was designed in this study, following which an external method was proposed to install the EFPI with the oil cavity on the transformer tank. On the basis of the principle of EFPI sensors and diaphragm vibration in gas-solid media, the relationship between the EFPI detection sensitivity and its diaphragm size was first analyzed, following which the high-sensitivity sensor size was designed and manufactured. Subsequently, the cavity oil structure of the external installation was designed. The Piezoelectric ceramic Transducer (PZT) was used to build an EFPI testing system, using which the amplitude-frequency characteristics and detection sensitivity were analyzed. Finally, a board-to-board electrode and transformer model were employed to build a PD detection system, and the external EFPI and PZT sensors were enabled to detect the PD acoustic signals synchronously. The experimental results show that the first-order natural resonance frequency of the EFPI sensor is 113 kHz, and the static pressure sensitivity is 7.5 nm/kPa. Moreover, the external EFPI sensor is significantly more sensitive than the PZT sensor. The external EFPI structure is effective for ensuring protection of the sensing diaphragm, and its installation is considerably less difficult. Furthermore, potential risks are prevented when the EFPI sensor is placed in the high-voltage field strength area.

Phase information is an important parameter in the wave function of a Bose-Einstein condensate in an optical lattice. However, in experiments, the phase information of the wave function cannot be obtained directly from the atom distribution in momentum space by absorption imaging or in-situ imaging. Thus, a deep learning network model was developed to study the influence of the phase distribution of a Bose-Einstein condensate on the atom distribution in momentum space. Thirty-two thousand data sets obtained by theoretical calculations were used as training and verification sets. Based on the analysis of the phase characteristics and momentum space of the wave function, a method for predicting the momentum of supercooled atoms in an optical lattice was developed using a convolutional recurrent neural network model. After the model verification, a difference between the model training and Schrodinger equation results is 1.76, which is 83% less than the average error of a back propagation neural network. Our approach provides a new application of machine learning in the field of physics.

Considering the development of transparent ocean strategy, low-cost staring imaging equipment have unique advantages in underwater optical imaging. However, it is difficult to separate the backscattering and imaging target, as well as to capture a clear image at a long distance. More importantly, before acquiring the effective target image, the strong scattering noise saturates the image and prevents subsequent processing. Thus, we propose a novel imaging method that combines short coherent illumination and polarization imaging. The short coherent illumination simplifies the separation of the backscattering and imaging target while the polarization technology prevents image saturation in advance and ensures the effective acquisition of the target image. In addition, we built a large-scale underwater optical imaging platform and conducted imaging tests at a long-distance of 22 m. The experimental results showed that the signal-to-noise ratio increased from 0.50 dB to 13.57 dB, and the anti-image early saturation ability of the device increased 1.42 times. These results are superior to the traditional polarization imaging. The proposed composite imaging method can provide technical support for large-range underwater optical monitoring.

Commercial Light-Emitting Diodes (LEDs) exhibit a -3 dB modulation bandwidth of only several MHz, which severely affects the information capacity and transmission in visible light communication. In this study, to fabricate an LED with high bandwidth, the influencing factors and mechanism of the modulation bandwidth of an LED device were investigated. Three LED chips with PN junction areas of 200 μm×800 μm, 300 μm×900 μm, and 300 μm×1 200 μm were designed and fabricated. By analyzing the photoelectric and modulation characteristics of the three LEDs, the relationship between the PN junction area and the modulation bandwidth was determined. Then, the capacitance-voltage curves of the three samples were compared, and the influence of the capacitance on the modulation bandwidth was analyzed. The results indicate that the LED with a PN junction area of 200 μm×800 μm exhibits the minimum capacitance, resulting in the maximum -3 dB bandwidth of 49.9 MHz among the three LEDs. This demonstrates that the parasitic capacitance caused by package, LED drive circuits, etc., has a significant effect on the modulation bandwidth of LEDs. Thus, LED devices with high bandwidth can be fabricated by decreasing the parasitic capacitance of the LED.

In order to obtain a reliable method for the fabrication of nanometer metal tips, the fabrication process for pyramidal Ni tips based on microelectroforming technology was designed and verified experimentally. First, silicon templates with inverted pyramidal pits were fabricated on (100) single crystal silicon wafers using anisotropic etching. Then, a 200 nm thick Ni thin film as seed layer was deposited onto the templates by magnetron sputtering combined with lift-off technology, followed by a Ni microelectroforming step. Finally, the pyramidal Ni nanotips were released by etching the silicon template in KOH solution. The results show that the relative error of square etch-windows on the silicon templates can be reduced by approximately 9% using inductively-coupled plasma dry etching instead of hydrofluoric acid etching in the patterning of the SiO2 etch-windows. The bottom side length of the pyramidal Ni nanotip is approximately 140 μm, and the smallest radius of curvature of the nanotips is 54 nm. The replication accuracy from the templates to the nanotips reached as high as 99%. The results show that the template microelectroforming technique can be used to achieve stable and mass fabrication of Ni nanotips, which reduces the cost and lays a solid foundation for the application of metal nanotips.

A selective laser melting method for manufacturing the composition gradient material part was investigated to solve the problem of manufacturing metallic gradient material parts with complex structures. A composition gradient design combined with several groups of scanning path data files and a material txt format file was proposed to obtain additive manufacturing data of the composition gradient material part. A real-time powder mixing and distributing device was used to mix and distribute the composition gradient powder by swinging a hopper along two axes. The principle of flexible cleaning and recycling of powder was used to clean and recycle different powders in the same layer during the manufacturing of the gradient material part through selective laser melting. The experimental verification was carried out using a self-developed selective laser melting system for forming of gradient material part. A 4340+CuSn10 gradient material part was manufactured during the experiment, which showed a significant gradient change in color. Energy-dispersive X-ray spectroscopy analysis on the front side and the upper surface of the part showed that the average mass percentages of Fe in the middle three gradient areas were 4.94%, 36.49%, and 59.16% in the vertical direction, and 12.88%, 41%, and 53.59% in the horizontal direction. The composition of the experimental sample gradually changed in different layers and different regions of the same layer, which verified the feasibility of the method and provided a new choice for the manufacturing of composition gradient material parts.

The existing neutral line expressed through displacement superposition under the small deformation assumption leads to a large positioning deviation of the flexspline teeth. By taking the double disk wave generator as an example, the neutral line in the wrap angle was determined using the CAM isometric line. The outside neutral line expressed using a spline function was constructed based on the continuous conditions, such as slope and curvature, at both ends, and the difference between the tension of the neutral line was equal to the elongation produced by the circumferential force. A Finite Element Method (FEM) model of the tooth ring in contact with a circular cam was developed to verify the geometric interpolation method and the neutral line expressed with radial and circumferential displacements numerically under small and large deformations, respectively. The differences between the verification examples were calculated using the large deformation results, which were considered more suitable for the actual conditions. The examples showed that both theoretical solutions and the FEM model based on small deformation had more significant differences with the more reasonable FEM results for the large deformation case. In the 30° wrap angle model, the pole angle of the proposed method decreased by approximately 0.006°, and the pole diameter decreased by approximately 0.013 mm. The method derived based on the geometric condition in this study was closer to actual conditions and easy to implement when the model envelope angle ranged between 25° and 65°. The proposed model eliminates the influence of large geometric deformation, contact nonlinearity, and neutral line elongation in providing more accurate gear tooth positioning for the tooth profile design of harmonic drives.

Optical glass is the key material for manufacturing optical devices such as lenses and filter mirrors. Owing to its high hardness, high brittleness and low fracture toughness, edge damage has been the major defect during the machining of optical glass. Based on the drilling experiment of BK7 optical glass, the damage characteristics at the entrance and exit of the machined hole were analyzed; subsequently, the formation mechanism of edge damage was discussed by using the compression fracture theory of brittle materials; and finally, the influence of drilling parameters was studied. Consequently, the suppression effect of Rotary Ultrasonic Drilling (RUD) and Rotary Ultrasonic Pecking Drilling (RUPD) on edge damage was studied. The results indicate that ‘local edge chipping’ is the main form of the entrance damage, and the superposition of ‘layer separation’ and ‘discontinuous chipping’ is the main character of the exit damage. The subsurface lateral crack expansion, caused by squeezing the tool end abrasive, is the main cause of generating entrance damage, whereas the middle crack expansion and axial collapse are the main causes of exit damage. The RUD has a clear effect on the reduction of hole edge damage: its maximum entrance chipping size Lin~~max and the maximum exit separation size can be decreased by 15%—50% and 45%—65% of common drilling, respectively. The RUPD can promote the expulsion of abrasive scraps; thus, achieving less exit damage than the RUD.

In this study, a variable stiffness friction damper was proposed that effectively suppresses micro-vibration output during the orbit attitude adjustment of the spaceborne flywheel and improves dynamic environment during launch. Depending on the actual load conditions of the spaceborne flywheel in orbit and active launching phases, the harmonic balance method was used to obtain the frequency domain force/absolute displacement transmissibility curve. The transmissibility curve was verified via experimental methods. The experimental results show that the variable stiffness friction damper can provide greater damping at the resonance frequency and control the amplification factor to a smaller range. The measured results agree well with the theoretical solution. The research results show that under the effect of small load on-orbit phase, the variable stiffness friction damper can effectively suppress the micro-vibration of the flywheel output. On ensuring high frequency vibration isolation performance, the peak value of the force transmissibility is approximately 6.5. Under heavy load during the launching phase, the peak value of the absolute displacement transmissibility is approximately 2.3. Notably, the variable stiffness friction damper can simultaneously consider the influence of the change of the flywheel load conditions in different working phases, and significantly inhibit its dynamic response.

Kinematics calibration is an important way to improve the accuracy of Parallel Manipulators (PMs). The robustness of kinematics calibration to measurement sensor noise can be improved by choosing measurement poses based on maximizing the observability index related to the identification Jacobian. The identification Jacobian matrix is a function of position and orientation of the PM, but the magnitudes of identification Jacobian matrix elements corresponding to position and orientation are different. This leads to a difference in the robustness of position and orientation to measurement sensor noise. Therefore, the identification Jacobian matrix was first normalized and the measurement poses were then selected through the observability index related to the normalized identification Jacobian to ensure the same robustness of position and orientation to measurement sensor noise. Through numerical examples, the precision of the structural parameters calibrated by using this method under three kinds of measurement noise is greatly improved compared with that of the traditional method, increasing from 1.007 5 mm to 0.336 7 mm, 0.100 9 mm to 0.033 7 mm, and 0010 1 nn to 0.003 4 mm, respectively. Compared with the traditional method, the calibration precision of position using this method basically remained unchanged, but the attitude accuracy was sharply higher, increasing from 0.015 2° to 0.003 3°, 0.003 3° to 0.000 3°, and 1.5×10-4° to 3.3×10-5°, respectively. The measured pose selected in this method was used to calibrate the Stewart parallel mechanism, and the mean error values of the position and attitude before and after calibration were reduced from 2.321 mm to 0.242 mm and from 0.246° to 0.025°, respectively, effectively improving the pose accuracy.

Given the large number of species of flies and their individual complex characteristics, recognizing a particular type of fly has proved to be time consuming and, for the most part, inaccurate. In this paper, a method for the facial recognition of a fly using deep learning technologies was proposed, specifically a Convolutional Neural Network (CNN), and its related face recognition algorithms. Initially, a multi-task convolutional neural network was utilized and optimized for the image alignment process. Thus, depth-wise separable convolutions were applied to reduce the number of calculation parameters as well as the image preprocessing time. Next, we combined the rough extraction of contour features and fine extraction of specific parts to derive more abundant feature information. The convolution and pooling layers were harnessed to elicit contour eigenvalues of the image, while Inception-ResNet and Reduction networks were administered simultaneously to obtain eigenvalues of specific parts. Finally, the above methods were coalesced to enhance the accuracy and comprehensibility of the resultant feature information for network training. Experimental results show that the mean average precision of the proposed method is 94.03%. When compared with other network training methods, this method not only improves the computational efficiency but also ensures high accuracy.

To minimize position information extraction inaccuracy while using Position Sensitive Devices (PSD), and to overcome noise jamming resulting from components and signal processing circuits, a Feedback stacking model based on Extreme Learning Machine (FsELM) was proposed. FsELM employed Extreme Learning Machine (ELM) as the basic training block, updated the input data based on the differences between the truth values and monolayer predicted results, developed the feedback stacking models by cyclic training, and realized the effective depth extraction information of the PSD signals. Further, a one-dimensional PSD-based laser triangular displacement detection experiment was designed to evaluate the performance of the algorithm. The processing abilities of traditional filtering methods, such as classical learning algorithm, ELM, its variants and the proposed FsELM were compared. The FsELM exhibited a significantly higher prediction accuracy compared to other processing methods. The mean square error and prediction accuracy are 1.4×10-5 and 0.78%, respectively. In addition, the operating speed of the FsELM is higher than that of all the other methods, except for the models with single training structures, such as ELM. The results demonstrate the efficient management of random noise interference and accurate prediction ability of the FsELM in uncertain environments.

An algorithm for joint cell formation and power distribution in a visible light communication network was proposed to maximize energy efficiency. Based on user-centric design, a novel user clustering algorithm was proposed, and a between-within proportion clustering index was introduced to effectively reflect the separability and compactness of user clustering, determine the optimal clustering number, and form virtual communities by associating Access Points (APs) with clustering users according to the proposed measurement. Under the constraint of quality of service, the joint power distribution was solved iteratively using the Dinkelbach algorithm and dual projection sub-gradient algorithm. The optimal power distribution scheme and APs participating in the communication within the cell were obtained, effectively improving the energy efficiency. Simulation results show that the proposed algorithm can significantly improve energy efficiency compared with traditional algorithms.

To solve the problem in which a traditional ResNet101 model cannot effectively describe the detailed features of remote sensing images, leading to the unclear segmentation boundary between roads, trees, and buildings in complex scenes, a Multiscale-feature Fusion Dilated Convolution ResNet (MFDC-ResNet) was proposed. First, to obtain large-scale building feature information of remote sensing images, a dilated convolution was introduced in the deep residual network to capture richer multi-scale details. Second, to enhance the expression ability of the center point of dilated convolution on the building of feature images, a 3×3 convolution kernel was proposed to extract features in the local area of remote sensing images. Finally, a spatial pyramid pool model of multi-scale feature fusion was proposed to fuse the multi-scale features, obtain the building contextual information of different scales of remote sensing images, and complete the accurate segmentation of buildings. The results of the experiments show that the mean Intersection over Union (mIoU) of building segmentation in WHU is 0.820 and the recall rate is 0.882. The developed method can effectively overcome the influence of roads, trees, and other factors. Moreover, the building boundary can be extracted clearly and smoothly from the remote sensing images and the segmentation accuracy is improved.

Three-dimensional LiDAR is widely used in unmanned driving systems, mainly to detect the road environment and for collision avoidance detection. A real-time method to segment the point cloud based on depth projection was proposed to increase the segmentation accuracy of a point cloud scanned by LiDAR. Voxel filtering was first used to remove noise points, after which progressive morphological filtering was used to remove ground points, and finally the point cloud was subjected to point depth projection. The adaptive angle threshold method for the depth projection image was used to segment the point cloud, and after segmentation of the point cloud target, a hybrid hierarchical bounding box was constructed for collision detection. The experimental results show that this method constitutes a significant improvement in time efficiency compared with traditional clustering algorithms, and can effectively reduce the problem of over-segmentation. The proposed method increased the segmentation accuracy rate in the experiment to 78.82%. The combined hierarchical bounding box algorithm is applied to the segmented points.

The prediction accuracy of a Conventional Recursive Least Square (CRLS) predictor is strongly correlated with the inter-spectral correlation and is sensitive to the sequence in which the pixels are predicted. In view thereof, a lossless compression method for hyperspectral images was proposed. The method, which was based on the CRLS predictor, was modified to enable the selection of adaptive bands and to optimize the prediction sequence mode. First, to improve the correlation between the reference bands and the band to be predicted, the bands of the hyperspectral image were reordered according to the criterion of the maximum inter-spectral correlation coefficient in the preprocessing stage. Subsequently, the adaptive band selection strategy was used to select multiple bands with the highest correlation with the band to be predicted for use as prediction reference bands. Afterwards, the CRLS predictor with the best prediction sequence mode, selected by the minimum prediction residual entropy, was used for inter-spectral prediction. Finally, the arithmetic encoder was used to encode the prediction residual. Experiments on the AVIRIS 2006 dataset show that this method achieves bit rates of 3.314, 5.594, and 2.395 bpp on a 16-bit calibrated image, 16-bit uncalibrated image, and 12-bit uncalibrated image, respectively. These results indicate that this method can effectively improve the prediction accuracy of the CRLS predictor without significantly increasing the computational complexity. The best result of the proposed method closely approximates or is superior to that of other similar methods.

The initial point cloud model acquired by 3D laser scanning equipment contains more noise points that is not good for the later point cloud processing. Therefore, the noise needs to be deleted. A hierarchical point cloud coarse-to-fine denoising algorithm was proposed for effective retention of the sharp geometric features of the point cloud. The tensor voting matrix of the points and their neighbors was constructed. In addition, the diffusion tensor was constructed by calculating the eigenvalues and eigenvectors of the matrix. The diffusion tensor-based anisotropic diffusion equation was applied for cyclic filtering, to realize the initial coarse denoising of the point cloud. Further, the curvature feature of the point cloud was calculated post-filtering. To achieve fine denoising, the noise points in the point cloud were further deleted according to the curvature value. Finally, the point cloud entropy was calculated for quantitative evaluation of the denoising algorithm. The experimental results demonstrate that the proposed point cloud denoising algorithm exhibited a smaller denoising error, higher entropy value, and high execution efficiency. The proposed hierarchical point cloud denoising algorithm can quickly and accurately delete noise points, while retaining sharper geometric features of the point cloud. Therefore, it is an effective point cloud denoising algorithm.