A model with radial basis functions based on surface slope (RBF-BS) has advantages in terms of its characterizations and optimal system design. However, using this model can lead to some problems such as low optimization efficiency and difficulties in conducting tolerance analysis owing to the use of too many basis functions in RBF-BS. Thus, a method of achieving the optimal design and performing tolerance analysis with this model was proposed. This study combined global and local optimizations to develop a new optimization design method based on the local characteristics of the RBF-BS model. The linear relationship between the coefficient of basis function and freeform surface sag was determined by the method of mathematical statistics, and the initial range of tolerance of coefficients was directly determined. The reasonable tolerance value of freeform surface sag was acquired by large-sample statistics. When the proposed method of achieving the optimized design and performing tolerance analysis was applied to a head-mounted display system with a freeform surface using the RBF-BS model, the results indicate that the average modulation transfer function in the entire field view exceeds 0.3, and the largest distortion is 3.45%, which are consistent with the system design target. Based on the results of tolerance analysis, the experimental system integration successfully achieves the display of picture. The proposed method of achieving the optimized design and performing tolerance analysis provides valuable information about RBF-BS and other local freeform surface models.

To optimize a gas-stimulated Raman scattering system, the phenomenon of gas-stimulated Raman scattering was demonstrated through a combination of theory and experiments. The Raman gain coefficients of CH4 and D2 at different pressures were calculated theoretically. Stimulated Raman scattering experiments were then conducted using a 355 nm Nd∶YAG laser with an output energy of 70 mJ. The output energies of Stokes light at gas pressures of 1×105 and 2×106 Pa and focal lengths of 500 and 750 mm were measured. It is found that the CH4 Raman gain coefficient increases with an increase in gas pressure. In addition, the D2 Raman gain coefficient reaches the maximum at approximately 1×106 Pa and then do not change with increase in gas pressure. Experimental results indicate that reasonable values for Raman cell gas pressure and focal length of the coupled focusing lens must be selected for gas-stimulated Raman device applications. For example, a longer focal length and a higher pressure must be used for a methane gas system, whereas a longer focal length and lower pressure must be used for a helium system. The theoretical calculations corresponded with the experimental results. These research results shall play a major role in optimizing the light source of NO2 differential absorption laser radar.

To acquire the zero-gravity surface figure error of an aspherical mirror in a ground fabricating environment, a high-precision rotation method based on gravity compensation technology was established. First, the basic principle of the N equal interval rotation method is introduced. Second, combined with an aspheric ultra low expansion (ULE) mirror with a diameter of Ф1 290 mm processing, the rotation angle and off-center control methods are given, respectively, and the angle error and off-center error are better than 0.1° and 0.1 mm, respectively. Third, in the low-precision stage, the rotation result is processed by three-position rotations and the surface accuracy of the mirror quickly converges to 0.029λ-RMS. At the same time, the symmetry error on the mirror surface is cumulatively amplified due to the application of the rotation method. After removal, the surface accuracy further converges to 0.023λ-root mean square (RMS). Finally, the six-position rotations are used to guide the optical manufacturing. The surface figure error of the mirror in the six directions is 0.012λ-RMS and 0.010λ-RMS by removing the gravity deformation error, which allows it to be considered as a zero-gravity mirror in the space environment after the satellite is in orbit. The method described in the studyis not only applicable to the fabrication of one meter-level mirrors, but also to larger space aspheric mirrors with the goals of a zero-gravity surface figure error.

The optical fiber Surface Plasma Resonance (SPR) sensor is usually based on the fiber core as the resonance substrate. It is necessary to remove the optical fiber cladding by adopting complex processing technologies such as etching, side polishing, grinding, etc. However, there are problems such as evanescent waves that are not easy to leak or sensor probe making them difficult to observe. In this study, a step-index multimode fiber clad SPR sensor with fiber cladding as the SPR resonant substrate is proposed. The transmitted light of a single-mode fiber is injected into step-index multimode fiber cladding by core-shift welding and the 50 nm gold film is plated on the step-index multimode fiber cladding. In the probe sensing section, all the light field energy is distributed in the step-index multimode fiber cladding, the SPR effect is sufficient, and a deeper resonance valley is obtained than by using the conventional fiber cladding SPR sensing structure. When the refractive index measurement range is 1.333—1.385 RIU, the average sensitivity of the sensor can reach 2 307 nm/RIU. This study also explores the influence of different parameters on the diameter and length of the multimode fiber core in the sensing section. The step-index multimode fiber cladding SPR sensor proposed in this study is simple to fabricate and can effectively solve the evanescent wave problem making information difficult to obtain between fiber cladding and air interface.

A design method for a common aperture and multi-band optical navigation sensor was proposed with the aim of solving the problem in which single-band navigation sensors can only get limited information from a target, as they have a low recognition rate. The design method of the initial structure of the common aperture optical system with primary and secondary mirrors as the common part and working in different fields of view was presented based on the theory of optical power distribution. Visible, infrared, and laser channels share a Ritchey-Chretien system to meet the design requirements of a large aperture. To realize the functions of visible imaging, infrared imaging, and LiDAR detection, the receiving part of the LiDAR was separated and the use of spectroscopic elements was reduced by the selective transmission of an optical film on the secondary mirror. The image of the Ritchey-Chretien system was divided into two paths by a prism; one was to the image photodetector and the other was to the image in the uncooled infrared detector. Through a large relative aperture, the radiation response of an uncooled detector was improved. The design results show that the Modulation Transfer Function (MTF) of the visible system is above 0.4 at 90 lp/mm, the MTF of the infrared system is above 0.4 at 14.7 lp/mm, the transfer function of the imaging system is close to the diffraction limit, and the energy of the laser receiving system is up to 90% at 30 μm from the center of the mass of the detector. The imaging quality of the system is suitable at -20—40 ℃.

To analyze the efficiency of geometrical optics and its influencing factors on a Compound Parabolic Collector (CPC), a theoretical basis was provided for the optimization design and practical application of a linear Fresnel reflector system. According to the CPC characteristics of a linear Fresnel reflector system, a model was built using Matlab software; the geometrical optics efficiency was simulated by a ray-tracing method. The geometrical optics efficiency model was validated by TracePro software. The model was used to analyze the influence of outer diameter tolerance, position deviation of the absorber tube, and the profile error on the geometrical optics efficiency of the CPC. Results show that the mean geometrical optics efficiency of CPC with a gap of 32.5 mm, half-acceptance angle of 55°, truncation ratio of 0.3, and an absorber tube outer diameter of 70 mm, can be reduced rapidly when the outer diameter tolerance of the absorber tube is less than -0.4 mm, there is a horizontal offset, a positive vertical offset, or an increase in the profile of the CPC. In practical applications, the above factors should be avoided as much as possible or the range of the incidence angle of the CPC should be reduced in design to reduce the influence on its mean geometrical optics efficiency.

The Finite Difference Time Domain (FDTD) method is used to adjust the interval period multilayer structure of the narrow band-stop filter in mid-infrared of graphene nanometer film and refractive index regulable lithium niobate (LiNbO3). First, it is necessary to control the number of layers for the multilayer structure to adjust the filter characteristics. The numerical result shows that with the increase of the layer number, the transmission rate decreases, the wavelength corresponding to the minimum transmittance is shortened, the width of the band-stop significantly narrows down, and the narrow-band characteristics curve gets very sharp. Second, a change in the chemical potential of the dielectric constant for each layer of graphene by adjusting the voltage to achieve the adjustment of the narrow-band filter characteristics is necessary. The numerical result shows that for the same number of layers, the wavelength corresponding to the minimum transmission rate is shortened with the increase of the chemical potential. When there are more layers, the minimum transmission rate changes in the short wavelength; when there are few layers, the minimum transmission rate changes in the long wavelength. Third, a change in the refractive index of the lithium niobite (LiNbO3) crystal in the multilayer structure through voltage changes to achieve the adjustment of narrow-band filter characteristics, is also necessary. The numerical results show that for the same number of layers, the wavelength corresponding to the minimum transmission rate gets longer with the increase of the refractive index of the lithium niobite (LiNbO3) crystal. When there are more layers, the minimum transmission rate changes are in the short wavelength; when there are few layers, the minimum transmission rate changes are in the long wavelength.

To solve the problem in which the contrast between a space target and its background is too low to be distinguished in a dark scene, the focal-plane array polarization imaging system is used to image the outdoor dark scene and the indoor space simulation environment. At the same time, to compensate for the shortcomings of the decreased resolution of the focal-plane array polarization imaging system, the bicubic interpolation algorithm is used for upsampling. On one hand, the intensity image under four different polarization angles can be obtained by a single exposure of the focal-plane array polarizer camera, then, the Degree of Polarization (DOP) and the Angle of Polarization (AOP) images are compared to the intensity image. On the other hand, the bicubic interpolation algorithm is used to upsample the four intensity images to improve the image resolution and then, calculate the DOP image compared with the DOP image obtained without the upsampling procedure. The experimental results show that compared with traditional intensity imaging, the contrast of the target is improved, the edge information and texture information are better displayed by imaging polarimetry, and the final imaging resolution is improved by the bicubic interpolation algorithm. The measure of enhancement (EME), a contrast evaluation index, is approximately 17% more in the DOP image than in the intensity image, proving that the focal plane polarization imaging system using a bicubic interpolation algorithm has potential application value for the identification of space targets in the dark scenes.

Freeform optics are greatly in demand for use in next-generation space cameras for achieving higher optical design freedom as compared with the commonly used spherical and aspherical elements. Freeform optics have the advantages of excellent optical performance and simplified system structure. The development of optical remote sensing technology imposes higher requirements on these optics. Therefore, a series of optimization procedures of grinding process parameters for zerodour mirrors were conducted, and the grinding of a 300 mm aperture freeform mirror was completed using the optimized process parameters. The results show that the surface figure of the mirror increases to 8.89 μm PV, and the depth of the subsurface damage layer reduces to less than 18 μm. Realizing high precision and low damage grinding of freeform optics is of great significance for reducing the workload of subsequent lapping and polishing.

A piezo-actuated stick-slip device, which consists of a micro-motion stage and a slider, can realize long-range motions with high resolution. Induced mechanical vibrations in the micro-motion stage impose an upper bound on the speed of the piezo-actuated stick-slip device. To address this issue, this study proposes a composite control scheme with active damping. First, the saw-tooth wave signal is filtered and made smoother. Then, a delayed position feedback controller is introduced to improve the damping of the micro-motion stage and mitigate its vibrations. Both tracking and feedforward controllers are designed to reduce the tracking errors and increase the control bandwidth. Finally, the proposed controller is implemented on the prototype of a piezo-actuated stick-slip rotational device. The experimental results show that, compared to a proportional integral controller, the proposed controller improves the control bandwidth from 32.7 to 1 466.5 Hz. In addition, compared to a conventional feedforward controller, the proposed controller under a 100-Hz saw-tooth wave signal with a duty cycle of 0.2 improves the angular velocity of the slider from 3.52 to 9.03 mrad/s. The device angular speed is improved significantly.

In order to improve the manufacturing accuracy of gear involute artifacts, this study investigates the influence rule of installation eccentricity for gear involute artifacts on their profile slope deviation. Firstly, based on the involute generation principle of the double roller-guide, a mathematical model of the influence of installation eccentricity on the profile slope deviation of a gear involute artifact is developed, and the tooth profile form deviation are separated from the tooth profile slope deviations. Based on the mathematical model, the installation eccentricity corresponding to compensation of the profile slope deviation of gear involute artifact is deduced, and it is verified via test experiments. The experimental results demonstrate that the profile slope deviation of a gear involute artifact can be reduced from -3.53 μm to -0.06 μm, and the requirements for a class-1 gear involute artifact are satisfied. These conclusions on the influence rule of installation eccentricity on the profile slope deviation of gear involute artifacts can be used to compensate their tooth profile slope deviation and provide technical support for the development of high-grade gear involute artifacts.

In order to solve the problem of in-orbit optical load ground test vibration source simulation, a multi-dimensional micro-vibration simulation platform based on a parallel mechanism was designed, which can effectively reproduce the characteristics of spatial micro-vibration distribution frequencies and small vibration levels. Firstly, the virtual frequency principle and the Newton-Eulerian equation were used to derive the analytical formula for the natural frequency of the system. This was combined with the design index to optimize the configuration, and the structural design was configured based on this, so that the natural frequency satisfied the analog bandwidth of 5-250 Hz. Finally, a control method based on transfer function was proposed, which verified its correctness and solved the working ability of the platform. The fundamental frequency corresponding to the sixth stage of the platform was observed to be 3.4 Hz, and the fundamental frequency corresponding to the seventh order was observed to be 356 Hz, which satisfied the bandwidth requirement. The maximum error between the output and the target value obtained via the transfer function control is 1.54%, which indicates that the method is suitable for platform control. The maximum translational acceleration of the upper platform is observed to be 399.3 mg, and the maximum angular disturbance is detected to be 1 979.3 μrad, which meets the requirements of the index. The platform exhibits large analog bandwidth, high load capacity, and small vibration levels. It can be used as space micro-vibration ground test vibration source simulation equipment.

With increasing applications of Magnetorheological Finishing (MRF) in the field of ultra-precision manufacturing, enhancement of the efficiency of MRF is imperative. To address this requirement, a technique based on pH adjustment was developed in this paper, which in turn, optimized the properties of the polishing slurry. The dispersion behavior of polished particles and the rheological characteristics of the polishing slurry were investigated via scanning electron microscope, particle-size analysis, and Zeta potential test. The results demonstrate that an adjustable pH facilitates the dispersion of nanoparticles. A pH value of 12 has been identified to be the most suitable for the dispersion of the polishing particles. The absolute value of the Zeta potential is observed to be 33.28 mV and the particles diameter of D50 is detected to be 260 nm under this pH. After polishing of fused silica, the alkaline MRF fluid is demonstrated to not only lead to superior material removal rate, with the corresponding peak removal rate and volumetric removal rate increasing by 87% and 66%, respectively, but also achieve a precise level of surface roughness.

Mechanical connection mechanisms used in space optical systems experience problems with lubrication, thermal expansion coefficient matching, problems related to its complex structure, and small docking eccentricities. In this study, a flux-pinning interface scheme, which can be applied to a large segmented reflect mirror in space, was proposed. The mathematical model that contained the current equivalent distribution model was built and calculated by the H-formulation method,then, verification experiments were designed to demonstrate the effectiveness of the model. The relationship between the levitation force, stiffness, and relative position of the flux pinning interface under three working condition was obtained using the finite element method. The results indicate that when the distance between the superconductor and permanent magnet is 5 mm, the stiffness of the vertical direction ky can reach 7 000 N/m; when the distance between the superconductor and permanent magnet is 10 mm and it is in a central position, the stiffness of the lateral direction kx can reach 3 800 N/m. The stiffness of the vertical direction ky when Δx=4 mm drops by 20% as compared with the one with Δx=0 mm, therefore, the mechanism can provide greater Δx values than traditional mechanisms and sufficient stiffness, which can also be used for buffering when docking.

In order to enable robots to accurately perceive and grasp objects, a novel type of tactile sensing unit using an iron-gallium wire as the sensitive component was designed, and a test platform was developed to evaluate the output characteristics of the sensing unit. The relationship between the output voltage and the applied static and dynamic pressure was tested. A sensor array structure was designed using the tactile sensing unit as the core; then, the sensing array was mounted on a mechanical hand, and the gripping experiment was performed.The experimental results show that the tactile sensing unit, consisting of a Fe-Ga wire with a length of 16 mm and diameter of 0.8 mm in a bias magnetic field of 1.908 kA/m and a static pressure of 2 N, can achieve an output voltage of 96 mV and a sensitivity of 48 mV/N. Under the dynamic action of 1-4 Hz frequency and 0-2 N pressure, the output curve of the sensing unit is detected to be smooth, and the sensitivity is observed to be high. The sensor array is capable of sensing multiple sets of pressure information to accurately display the force distribution of the mechanical fingers.These properties make the unit widely applicable in the field of accurate grasping and intelligent control.

To overcome the trade-off between high performance and system chattering in conventional Sliding Mode Control(SMC), and to improve the reliability and pointing accuracy of an aeronautical photoelectric stablization platform based on Permanent Magnet Synchronous Motors(PMSM), an approach law for sliding mode controller was proposed. The law can effectively weaken the chattering of the system and achieve a better tracking effect. On this basis, to increase the bandwidth of the disturbance observer and the accuracy of the observation, an Extended State Observer(ESO) was introduced into the servo system of the photoelectric stabilization platform to observe the total disturbance of the system.Further, the lumped disturbance was compensated within the sliding mode controller to better suppress the chattering of the system and improve the system′s ability to resist external disturbances. Experimental results demonstrate that the sliding mode controller, combined with ESO, performs better than the traditional PI+DOB control method. In the uniform speed tracking experiment, the RMS value of the system′s position pointing error is observed to be merely 0.005 7°, which completely satisfies the requirements of the aeronautical photoelectric stabilization platform, and the accuracy is observed to be approximately three times as high as that of the classical PI+DOB control method. In the sinusoidal wave tracking experiment, the proposed method is observed to greatly reduce the phase lag of speed tracking, and the position pointing error is detected to be merely one-sixth of that of the PI+DOB method. In the trianguler wave tracking experiment, the RMS of the position pointing error is approximately one third of that of PI+DOB method.

In order to meet the demands of transmitting mixed micro-fluidics and micro-fluid media in the fields of biology, chemistry, medicine, aviation, etc., a compositedrag-reducing fluid and avalve-less pump that integrates fluid mixing and pumping was proposed.It enabled simultaneous vortex inducing and mixing by using a bluff body. A new interpretation and expression of flow resistance was established based on fluid flow theory. Aresistance analysis of the vortex-inducing composite bluff body was carried out by a "comparative method of pressure around the wake flow" and a "fluid unit momentum analysis", which revealed the essential mechanisms of the composite bluff body-causing valve and pump. The structure of a traditional valve-less pump was modified, and swirl mixing was realized while the pump-back flow was reduced by introducing a shunt ring and a composite bluff body group. A pump prototype was developed, and the pump flow and fluid mixing tests were carried out at a driving frequency of 11 Hz and avoltage of 180 V. The instantaneous flow rate reached 40.1 ml/min and the fluid was well mixed by the motions of the fluid vortex. These tests verify that the new type of pump can mix and transmit fluidsbetter than other types. It improves the structure and function of the valve-less pump, and at the same time, lays a foundation for the useof valve-less pumps in the field of micro-fluid transmission.

Magnetic field information plays an important role in determining the heading angle and attitude information, but it is highly susceptible to interference from surrounding ferromagnetic materials. The traditional ellipsoid fitting algorithm for magnetic field calibration requires high quality magnetometer data, which is not customer-friendly. This paper implemented a more convenient and accurate real-time magnetic field calibration using a smartphone-based platform. Firstly, according to the physical relationship between angular velocity and magnetic field, the state transition equation of magnetometer data was derived by using the gyro data recursively. The magnetometer measurements were used in the measurement equation, and the extended Kalman filter algorithm was used to complete the real-time EKF magnetic field compensated by the gyroscope. Then, the data from the magnetometer and the gyroscope were retrieved in real time on the mobile phone, and a real-time EKF magnetic field calibration operation was performed on the two, producing the calibration result as the output.It was then compared with the uncalibrated data and the calibration data obtained via the ellipsoid fitting algorithm.During the comparison experiment,the gyroscope compensation algorithm can not only achieve real-time calibration with any change to the external magnetic field environment, but can also complete the calibration of the magnetometer data within 2 s. In terms of calibration accuracy, when the mobile phone is stationary, the EKF algorithm can reduce the magnetic field interference in real time, with the value of the quality parameter Q of the calibration being 0.72.When the mobile phone is wound around "8", the quality parameter Q of the calibration is observed to be 0.53, which is better than 0.03 - its value for the ellipsoid fitting algorithm. When four scenarios of daily activities on the mobile phone are simulated, the quality parameters Q obtained are 0.73, 0.54, 0.52, and 0.48, respectively, and in these cases, the ellipsoid fitting algorithm is observed to be incapableof calibration. Thus, real-time, high-precision, easy-to-use magnetic field calibration is achieved. Experimental testsdemonstrate that the magnetic field calibration method based on extended Kalman filter can dynamically calibrate magnetic field interference in real time. Its calibration speed is fast, its precision is high, and its anti-interference ability is adequate. Therefore, it has a wide range of applications in consumer electronics, vehicle inertial navigation systems, and the military.

Several problems are encountered when measuring the flatness error of large and complex workpieces in a production line, such as abroad area of the detection surface and a vast amount of data. To improve the efficiency and accuracy of flatness error detection, an optimization algorithm was adopted to increase the speed of flatness error evaluation. The Differential Evolution (DE)algorithm was implemented for solving these problems, and the optimization method of Particle Swarm Optimization(PSO) algorithm was integrated into the DE algorithm framework to increase the convergence speed by improving the mutation operation. This study proposed a mathematical model using the minimum zone method for the flatness error evaluation of large workpieces and expounded the principle and implementation steps of the improved DE algorithm. Finally, using the outer panel of a forklift truck as an example, the convergence speed and accuracy of the algorithm were verified by evaluating the flatness error of the outer panel. The results demonstrate that the convergence result of the improved DE algorithm is stable for evaluating the flatness error of large workpieces, and the error is close to zero. The accuracy of the proposed algorithm is 36.83% higher than that of the genetic algorithm, and the convergence speed is 58.33% and 28.57% higher than those of the genetic algorithm and standard DE algorithm, respectively. The proposed algorithm can be satisfactorily applied to the flatness error detection of large workpieces to improve the detection efficiency.

To solve the problems of the inability of LCM in detecting damaged targets in local bright areas and its low separation efficiency, a damage detection method based on local contrast image enhancement of a neighborhood vector local dot was proposed in this study. First, a nine-dimensional neighborhood vector was generated from the 3×3 neighborhood of each point in the image, and the maximum value in the neighborhood was extended to the nine-dimensional extreme vector in calculating the dots of the neighborhood vector and extreme vector. Second, the Neighborhood Vector Dot Contrast (NVDC) of each pixel was calculated. Third, the NVDC of the image patch of each pixel was calculated,i.e., the maximum NVDC value of each pixel in a relatively large area(5×5) was taken as the final Neighborhood Vector Dot Local Contrast (NVDLC) of the corresponding pixel. Fourth, the NVDLC image was binarized, and the targets were separated from the background. The experimental results demonstrate that the LSNR of the damaged image is improved from 3.775 to 12.445, and the signal of the damage target is significantly enhanced using the image enhancement method proposed in this study. After the maximum contrast image of NVDLC is enhanced, the damage targets with size of less than 2 pixels can be directly separated from the background using the adaptive threshold formula, which meets the accuracy and efficiency requirements of the weak contrast damage target detection.

Fusing a low-resolution Hyperspectral Image (HSI)with its corresponding high-resolution Multispectral Image (MSI) to obtain a high-resolution HSI is amajortechnique for capturing comprehensive scene information in both spatial and spectral domains. To exploit fully the spectral and spatial information of an image, an algorithm based on total-variation-regularized local spectral unmixing for HSI super-resolution was proposed in this study. Spectral features and corresponding spatial information were extracted from both HSIs and MSIs through coupled encode-decode networks, respectively. The decoder of the coupled network could effectively preserve spectral features, and regular terms integrating local low-rank and vector total variation constraints could make full use of spatial structure information in MSIs to extract a stable abundance matrix.Finally, the angular differences between representations were minimized to reduce the spectral distortion.Experimental results reveal that the reconstruction errors in CAVE and Harvard datasets reach 3.78 and 1.66, respectively, and the spectral angle maps are 6.57 and 3.03, respectively,thus outperforming the state-of-the-art methods. The proposed algorithm can make full use of the spatial properties and thus produces a better HIS super-resolution effect.

To reduce noise and distortion of a 3D range image obtained from a laser rangefinder, an anisotropic adaptive smoothing method was introduced. The method consisted of stochastic signal estimation and scale-space representation. A feature estimation model was then derived from neighboring pointsand was used to predict the manifold topological relations between those neighboring points. To achieve anisotropic diffusion smoothing, the Mahalanobis distance between the original image and the estimated model was calculated asa similarity measure,which could then be usedtoconstruct a convolution kernel. This method enabled the distortion of the original image to be effectively corrected and noise to be suppressed.It also made the main imagefeatures more apparent. Experimental results indicate that the peak signal-to-noise ratiogain of the adaptive algorithm reached 16.41 dB, and the mean square error was reduced to 66.16% when the noise variance was 4.0×10-4 m2. Our smoothing method can thus improve the quality of noisy 3D range imagesand can provide technical support for 3D sensing and measurement modeling based on laser rangefinders.

To solve the problem of image clarity and contrast degradation in fog scene image restoration, a single image defogging algorithm based on residual learning and guided filtering was proposed. A residual network was first constructed by using foggy images and corresponding clear images. Multi-scale convolution was then used to extract more detailed haze features. Taking advantage of the anisotropy of the guided filter, the algorithm then obtained a clearer fog-free image after the residual network was filtered to maintain image edge characteristics. Experiments produced the following results as compared with the dark channel prior, CAP, super-resolution convolutionalneuralnetwork, DehazeNet, and multi-scale convolutional neural network algorithms.On synthetic foggy images, the peak signal-to-noise ratio reacheda maximum of 27.840 3 dB, the structural similarity index measurereacheda maximum of 0.979 6, and the runtime on natural foggy images was as low as 0.4 s.In addition, the subjective and objective evaluations proved to be better than those of the other comparison algorithms.Thus, the proposed defogging algorithm not only produces a better defogging effectbut is also faster, there by offering a greater practical valuefor defogging applications than the other algorithms.

The sparse regularized image restoration method based on animage group adopts the adaptive structure group dictionary to replace the traditional learning dictionary based on the entireimage block.However, because some parameters in the algorithm have not been optimized, the complexity of the algorithm remains relatively high.Therefore, this study proposed a sparse regularization image restoration method based on an adaptive image group in terms of roughness.First, global and local image roughnesses were calculated.Then, the number of self-adaptive regularization iterations was calculated according to the global roughness, and the number of samples required for learning the dictionary was adjusted based on the local roughness.Finally, the adaptive parameters were applied to the process of sparse regularization image restoration based on an image group.The method proposed in this study was applied to a case involving image restoration of text removal for images with different degrees of smoothness. The experimental results show that the efficiency of image restoration can be greatly improved when a similar restoration effect is guaranteed, particularly in relatively smooth images, where the speed-up ratio can reach nearly 30 times.

To tackle the problem where by DeepLabV3+ loses considerable detail information during feature extraction, which leads to poor segmentation results in the edges of the objects, this study proposed a semantics segmentation algorithm based on DeepLabV3+ and optimized by superpixels. First, a DeepLabV3+ model was chosen to extract semantic features and obtain coarse semantic segmentation results. Then, the simple linear iterative clustering algorithm was used to segment the input image into superpixels. Finally, high-level abstract semantic features and detailed information of the superpixels were fused to obtain edge optimized semantic segmentation results. Experiments conducted on the PASCAL VOC 2O12 dataset show that compared to DeepLabV3+, the proposed algorithm had superior performance in terms of detail parts such as edges of objects, and the value of mIoU reached 83.8%.The proposed algorithm thus outperformed other state-of-the-art algorithms in terms of semantic segmentation.

Skull registration is one of the most important steps in craniofacial reconstruction. Its accuracy and efficiency have amajor impact on craniofacial reconstruction results. To improve the accuracy and efficiency of skull point cloud registration, this study proposed a hierarchical optimization method for skull point cloud registration. We divided skull registration into two processes, coarse and fine. First, the skull cloud model was denoised, simplified, and normalized. Then, the feature points were extracted from the skull point cloud model and their feature sequences were calculated. The initial corresponding point pairs were constrained based on the feature sequence, and the algorithm was used to eliminate the mismatched points to achieve coarse registration of the skull. Finally, an improved Iterative Closest Point (ICP) algorithm with geometric feature constraints was used to achieve fine skull registration to achieve the goal of accurate skull registration. In this study, experiments on rough, fine, and first coarse and then fine registration were conducted. Results show that in the coarse registration process, the registration accuracy of the optimized coarse registration algorithm is improved by approximately 35%, and the algorithm time consumption is increased by approximately 6% as compared with the unoptimized coarse registration algorithm. In the fine registration process, the registration accuracy and convergence speed of the improved ICP algorithm are improved by approximately 20% and 43%, respectively, and the time consumption of the algorithm is reduced by approximately 47% as compared with the ICP algorithm. For the complete registration process, the registration accuracy and convergence speed of the algorithm are better than those of the other two methods. Therefore, this method is an effective skull point cloud registration algorithm that can achieve accurate registration of a skull point cloud.