Aiming to address the unsatisfactory expression ability of semantics， which results in inaccurate text descriptions in video captioning， a chained semantic generation network （ChainS-Net） for video captioning is proposed. A multistage two-branch crossing chained feature extraction structure is constructed that uses global and local domain modules as basic units and captures the video semantics from global and local visual features， respectively. At each stage of the network， semantic information is transformed and parsed between the global and local domains. This method allows visual and semantic information to be cross referenced and improves the semantic expression ability. Furthermore， it allows a more effective semantic representation to be obtained through multistage iterative processing， thereby improving video captioning. Experimental results on MSR-VTT and MSVD datasets show that the proposed ChainS-Net outperforms other similar algorithms. Compared with the semantics-assisted video captioning network， SAVC， ChainS-Net shows average improvements of 2.5% in four metrics of video captioning.

The current point cloud registration methods cannot effectively address resolution mismatches， partial overlaps of point clouds， and numerous noise points when used for cultural relic models such as Terra-cotta Warriors. Hence， a ResUNet registration model based on the dynamic graph attention mechanism is proposed. The model integrates the residual module into the U-Net， performs three-dimensional （3D） sparse voxel convolution to calculate the features of point clouds， and applies a new normalization technology known as batch-neighborhood normalization to improve the robustness of features against point density changes. To improve the registration performance， the model aggregates local and context features via self- and cross-attention mechanisms. Finally， a random sampling consensus algorithm is used to estimate the change matrix between the source and target point clouds to complete the robust registration of the Terra-cotta Warriors model. To verify the effectiveness and robustness of the proposed method， four datasets （3DMatch， 3DLoMatch， 3DMatch with resolution mismatches， and two sets of terra-cotta warrior data） were used to test the registration model. Experimental results show that the registration recall was 90.1% and 61.0% in the 3DMatch and 3DLoMatch datasets， respectively. In the mismatched-resolution 3DMatch dataset， compared with feature learning-based registration algorithms， our algorithm improved the registration recall by 5%–20%. In the terra-cotta warrior dataset， the relative rotation and translation errors were less than 0.071 and 0.016， respectively， which are several times to one order of magnitude lower than those of other algorithms. The model proposed herein can extract key feature information from a 3D point cloud and is more robust to variations in point density and overlapping compared with other models.

To improve the quality of the fusion of infrared and visible images， this study proposes a novel method based on structure and texture-aware Retinex （STAR）. It first decomposes the source images into reflection and illumination components according to the STAR model. This decomposition can separate the texture and structure of the source images accurately and extract the detailed features of the visible images with low luminance. Subsequently， it merges the reflection component using a weight map， which is constructed using the second-order gradient of the source images as the input. Moreover， it merges the illumination component using a gamma function， which can make the fused image have more brightness information. Finally， it reconstructs the fused reflection and illumination components to obtain the final fusion image. According to the test on 38 pairs of widely used images in the TNO infrared and visible image database， the proposed method can generate excellent fused results with high visual quality. Furthermore， compared with five state-of-the-art methods for the fusion of infrared and visible images， the proposed method achieved significantly better objective evaluation results in mutual information， nonlinear correlation information entropy， and feature measurement based on image phase consistency. This study involves the use of STAR model for fusing infrared and visible images and establishes a direct fusion framework based on Retinex， which improves the fusion results of the existing methods in terms of detailed features and global contrast.

An improved Fisher’s criterion-based deep convolutional generative adversarial network algorithm （FDCGAN） is proposed in this study to solve the problem of quality deterioration in generated images when the training sample size is insufficient or number of iterations decreases. In this method， a linear layer is added to the discriminative model to extract category information. Then， Fisher’s criterion is used in backpropagation to combine label and category information. To minimize errors， the weights are adjusted iteratively while maintaining small intra-class and large inter-class distances such that the weights can rapidly approach the optimal value. A comparison of the experimental results of the FDCGAN model with that of the most recent six network models shows that the proposed model achieves better performance in all the FID metrics. In addition， applying the proposed model to the current advanced models in generalization tests yields more satisfactory results.

The impact of deformation error on the end positioning accuracy of high-precision attitude adjustment equipment cannot be ignored. To improve the accuracy of a 2RRPU/2RPU/U two-axis parallel attitude platform， an error compensation model is proposed based on a stiffness model to predict the error trend and a neural network algorithm to improve the prediction accuracy. The theoretical stiffness model is first established based on the full Jacobi and elastic deformation matrices of the attitude-adjusting platform. The validity of the prediction of the loaded deformation trend is verified by comparing it with the prediction by the Ansys data stiffness model. Then， a Simulink-Adams-Ansys-OPC-based simulation environment is built， and the platform full attitude simulation data is collected under random load. Next， the attitude and drive error trends are predicted based on the stiffness model and velocity Jacobi matrix， and the mapping from end error to drive compensation is realized based on the velocity Jacobi. The accuracy of the error prediction is further improved by using a neural network algorithm. The simulation results show that the attitude accuracy of the platform is improved by 9% after adopting the error compensation model， which verifies the effectiveness of the “stiffness prediction-neural network” model in the improvement of the platform attitude accuracy.

To address the problems of large data fluctuation and the difficulty of the accurate control of large climatic environment test chamber temperature with strong disturbances， the temperature structural composition， temperature control characteristics， and heat load disturbance source of test chamber temperature are analyzed. A dynamic fuzzy-proportional-integral-derivative-coordinated control algorithm for the cool-hot end temperature is proposed for the precise temperature control of large test chambers. The control principle and its specific algorithm are studied. The temperature data processing method used is Kalman filtering， and the control effect evaluation method is based on the Allan analysis of variance. Experimental results show that the deviation caused by thermal load disturbances in measuring temperature data can be effectively removed by Kalman filtering， reflecting the actual temperature change realistically. The control algorithm studied is compared with the traditional proportional-integral-derivative control algorithm， which has less temperature fluctuation and shorter steady-state system response time， and can be applied to the precise temperature control of large climate environmental test chambers.

Solar panels and satellite antennas on spacecraft are coupled systems with multiple flexible bodies. Multiple flexible coupling structures have the characteristics of close modes， low stiffness， and small damping； they are easily excited by disturbances to produce long-lasting and large amplitude coupled beat vibrations. To study the active vibration suppression and anti-disturbance ability of coupled multi-flexible structures， aiming at a three-coupled flexible beam （TCFB） system， the machine vision is used for vibration detection and feedback， and a nonlinear H∞ control method is used for vibration control. A three-coupling flexible piezoelectric beam experimental platform is established based on visual inspection and spring connection. The finite element model is established， and the parameters of the model are corrected according to the excitation identification of the free vibration and sinusoidal response signals. The anti-disturbance index is given based on the H∞ algorithm. By using the identified finite element model and the designed nonlinear control law， the H∞ controller is obtained based on the anti-disturbance， control speed， and energy consumption indices. The experimental study of disturbance vibration control based on visual feedback is carried out. The experimental results show that the vibration suppression speed of the H∞ control is approximately the same as that of the PD control， but the vibration amplitude of the H∞ control is lower when the vibration is stable. The experimental results indicate that the H∞ control method has better anti-disturbance ability than the PD control method.

The long transmission distance of satellite ground quantum communication and the relative motion between the satellite and ground increase the tracking performance requirements of a quantum tracker. To improve the tracking accuracy of satellite ground quantum communication and reduce the error rate of quantum communication， the tracking and pointing control system of a quantum tracker is investigated in this study， and the total disturbance affecting the tracking accuracy is compensated. First， the satellite ground quantum communication link and the quantum communication process and its influencing factors are introduced. A mathematical model of the quantum tracker is then constructed. Based on the mathematical model of the tracker， an active disturbance rejection control （ADRC） model predictive control algorithm is designed and proposed. To solve the problem of mode switching， a fractional order tracking differentiator is proposed to optimize the trajectory and thus reduce overshoot. Experimental results of quantum communication with the “Mozi” satellite reveal that the fractional order tracking differentiator improves the target acquisition speed by 22% following mode switching. Compared with the traditional PI control， the proposed ADRC model predictive control has stronger anti-interference capabilities， weakens the miss distance peak of pitch reversal， and has a tracking accuracy that reaches 2.9". The quantum polarization data reception is improved， and the total bit error rate is reduced to 1.18%. The proposed control algorithm can further improve the tracking accuracy of the quantum tracker， reduce the bit error rate， and meet the accuracy requirements of satellite ground quantum communication.

Herein， a method based on Scanlab is proposed to satisfy the technical stability requirements of special collimating light sources. The relation between the angle of incidence and transmittance is used as the principle of signal regulation. A quartz plate is fixed with Scanlab， and the transmittance of quartz is adjusted when Scanlab rotates. Real-time data feedback is recorded using a monitoring detector. To achieve power stability of the light source， the rotation angle， transmittance parameters of quartz， and angular range are analyzed based on Fresnel's law. A proportional-integral-derivative algorithm is used to modulate the error in the signal of the monitoring detector and the voltage corresponding to the reference power to accurately adjust the output of Scanlab. A stabilizer is used to control the power of a He-Ne 632.8 nm laser， and a trap detector is used to verify the level of power stability. The experimental results pertaining to the stability of the laser after modulation are as follows： The standard deviation CV is 0.016% （1 800 s）， and the peak-to-peak fluctuation SV is ±0.042% （1 800 s）. Compared with the free-running result， SV and CV are improved by factors of 8.79 and 13.76， respectively， and the power stability of the laser is enhanced.

This paper proposes a two-dimensional heterostructure photonic crystal on a GaN-based active platform for achieving a resonant microcavity in the blue band with a high quality （Q） factor and small mode volume （Vm）. The working mechanism of the heterostructure photonic crystal resonator is explained through band structure analysis. Based on the band edge and band gap principles applicable to photonic crystals， photonic in-plane feedback， a high Q factor resonance， and out-of-plane vertical emission are realized. The resonance characteristics of the core and cladding regions using different cavity parameters are studied by finite difference time domain （FDTD） simulation. The relationships among the microcavity structure and the resonance Q factor， cavity loss， resonance frequency， and Vm are discussed. These findings pave the way for designing high-Q/Vm resonators in the blue band and establishing a method with a theoretical basis for studying heterostructure photonic crystal microcavities with excellent resonance characteristics.

Because conventional three-dimensional digital image correlation （3D-DIC） equipment cannot be used to fully measure the deformation field of the partial surface of an aero-engine， a mirror-assisted muti-view measurement method is proposed and a set of multi-view 3D deformation measurement systems is constructed. The proposed method includes a high-precision camera calibration model， a correction method for reflection imaging， and a classical digital image correlation algorithm. In addition， using a photogrammetric technique， the method can align the measurement data of multiple 3D-DIC units to a common coordinate system to obtain multiple 3D shape and deformation fields simultaneously. The accuracy of the proposed method is verified through a specially designed uniaxial tensile test， where the accuracy of shape reconstruction is 0.03 mm， and the error of strain measurement is less than 50 με. Finally， the proposed system is applied to the loading test of an aero-engine casing. The shapes and corresponding deformation fields of multiple ribs of the casing are measured successfully， the results of which are consistent with those of finite element simulations. The results all demonstrate that the method is suitable for multi-view measurements of aero-engine casing deformation and has important engineering value for the structural designs and strength tests of complex cavities.

To achieve high-speed online detection of banknote surface coatings， a high-speed and real-time design method for a near-infrared （NIR） weak signal detection system based on high-speed banknote quality inspection machines is proposed. First， according to the characteristics of coating materials， driving and weak signal acquisition circuits are developed by using NIR light-emitting diodes and photodiodes to achieve high-speed real-time weak signal detection. Second， the real-time data transmission and process are realized by using a field programmable gate array （FPGA） and a graphical virtual instrument platform. Finally， a peak standard deviation method for calculating the peak intensity of the signal is proposed， and the real-time performance of the algorithm is analyzed. The detection of the coating on the banknote surface and the real-time simulation experiment of the algorithm is carried out. The experimental results indicate that when the paper speed reaches 22 m/s， whether the substrate is simple or complex， the signal from the uncoated banknotes is significantly higher than that of the coated banknotes， and the differences are about 7000， 4000， and 1400， respectively. The more complex the substrate， the smaller the signal difference. The time from the beginning to the end of the algorithm is about 2-3 ms， which is far less than the time interval of 22 ms. The speed of the algorithm meets the real-time requirements. The system developed in this paper is high-speed， real-time， and stable； it can effectively detect whether the coating on the surface of the paper is available. It is expected to provide a new method for online quality detection of coatings.

In terms of control， this study systematically reviews the advanced technology and research progress of image shift and image rotation compensation control technology for aviation optoelectronic imaging with a large field of view and high-resolution requirements. Under the development trend of image shift and rotation compensation control， the technology can be divided into two major categories： single- and multi-actuator cooperative compensation controls. For single-actuator compensation control， compensation methods based on gimbal control， optical path folding， and imaging medium control are summarized. For multi-actuator cooperative compensation control， the collaborative control scheme is resolved and analyzed from the three perspectives of command coordination， mechanical linkage， and multi-actuator information interaction， and the difficulties and future development trends of image shift and image rotation motion compensation control are prospected from five research angles. The study will contribute to a rapid and comprehensive understanding of the research status and development trends of image shift and rotation compensation control in the field of aviation optoelectronic imaging and act as a helpful reference for further improving the synthesis performance of aviation optoelectronic imaging equipment.