In order to meet the correction frequency and imaging quality requirements of adaptive optics systems for large ground-based high-resolution imaging telescopes, a 349-unit wavefront processor was designed, which realized a wavefront correction frequency of 1 500 Hz. The design of a real-time wavefront processor was proposed whose main components were a control computer, a FPGA wavefront slope processor, a GPU matrix multiplication handler and a modularization D/A converter. A closed-loop calibration result of the dynamic aberration produced by the 349-unit adaptive optics system was reported. In this experiment, the system realized an effective correction of the simulated atmospheric turbulence whose atmospheric coherent length r0 is 6 cm and the Greenwood frequency is 160 Hz. After the closed loop evaluation of the adaptive optics system, the average square root value of 1 000 frames of the wavefront aberration drops from 1.07 λ to 0.11 λ(the central wavelength is 600 nm). The proposed 349-unit adaptive optics system can achieve a higher imaging quality under the wavefront correction frequency of 1 500 Hz. Moreover, the wavefront processing delay is better than 235 μs. The analysis of the power spectrum reveals that this adaptive optics system can correct wavefront distortion which was less than 100 Hz.

With the fast development of integrated circuits, the automotive industry, advanced manufacturing, and astronomy, fringe reflection profilometry (FRP, also called deflectometry) is widely used to measure the three-dimensional (3D) shapes of specular surfaces. FRP reconstructs the 3D shape of an object by calculating the local gradient or depth information of the specular surface using the phase information in the reflected fringe patterns. In order to gain a complete understanding of the recent developments in FRP, this paper reviews the basic principles of fringe reflection, generation of fringe patterns, phase calculation methods, calibration of system parameters, and slope integration to obtain depth data. The paper also introduces direct FRP without the integration procedure. The merits and demerits of FRP are presented through applications and examples. Finally, the future research directions of FRP are highlighted. This review paper provides a useful reference for understanding and studying the 3D surface measurements of specular reflection objects.

The control technology of a distortion control system for zoom scanning imaging was investigated to eliminate ground distortion of a long wave infrared camera on-orbit scanning imaging. A high-precision multi motor position synchronization algorithm was proposed taking the multi motor position synchronization of zoom scanning control into consideration. The synchronous positions of the zoom group motor and the compensating group motor at each point of the scanning time were set according to the design principle of zoom scanning imaging control. In particular, the zoom control system was designed to realize synchronous timing with the scanning control system, calculate the running velocity of the motor every 0.01 s according to the current position and the next moment synchronous position, and control the velocity of the motor in real time. The experimental results show that the position synchronization errors of the zoom group motor and the compensating group motor are within ±0.003 mm and ±0.002 mm, respectively, and the maximum ground sample resolution deviation in the track direction is within ±0.047 m. The images obtained from the long wave infrared camera in the process of continuous zoom scanning control are quite clear, and the distortion has been significantly eliminated, thus realizing zoom scanning imaging.

For general attitude sensors, there is a deviation between the actual measurement results and design precision of non-on-orbit calibration equipment. In addition, they lack the simulation function for the target celestial bodys characteristics. In this work, we combined a polarization splitting prism and semi-reversing semi-transparent prism to complete liquid crystal on silicon (LCOS) optical splicing manner. With sub-pixel display technology, in which many pixels were used to express one feature point, a high-precision dynamic celestial simulator was designed for solving the problems mentioned above. First, the reasons for the low contrast ratio of two light valves in the traditional LCOS splicing manner were described in detail. Then, the improvement plan for the splicing structure was put forward. Subsequently, an apochromatic collimation optical system with small distortion and high imaging quality was designed. Furthermore, ideas and methods of sub-pixel display technology were investigated Finally, sub-pixel technology to conduct simulate a dynamic star map and target celestial body was employed. The results show that the contrast ratio of the LCOS splicing screen of a dynamic celestial body simulator, which can simulate a target celestial body, is high. Moreover, the angular travel error between two stars in the dynamic star map is less than ±6″. In addition, the stellar magnitude simulation context reaches eight continuous grades. The stellar magnitude simulation precision is better than ±0.3 mV. The results meet the precision requirements for feature point recognition of an attitude sensor. The high-precision dynamic celestial simulator may meet the basic requirements for ground precision calibration and functional testing of a sensor.

Scanning slit is an important element for controlling exposure dose in step-and-scan lithography. Too large a penumbra of the blades edge in scanning slit could affect the exposure performance. Firstly, according to the illumination principle of a step-and-scan lithography, the calculation formula of the penumbra of the blades edge at the reticle plane was derived by analyzing the relationship between the intensity distribution of the reticle and the thickness and location of the blades in the scanning slit. The theoretical value of the penumbra of the coplanar and noncoplanar blades edges at the reticle plane was verified using the simulating light model of a 0.75 NA lithography. Secondly, a high-accuracy scanning slit apparatus with four coplanar blades was developed, which not only meets the synchronization performance requirements of the step-and-scan lithography, but also reduces the penumbra of the blades edges in the X and Y-direction. Finally, the dynamic performance of the scanning blades and the actual penumbra of the blades edges at the reticle plane were also experimentally tested. The test results indicate that when the scanning speed reaches 470 mm/s, the dynamic position error of the blade is within ±30 μm and the penumbra of all edges does not exceed 0.5 mm and meets the requirements of the 90 nm step-and-scan lithography.

A nonmydriatic optical system for stereo imaging fundus camera was designed in this work. It could simultaneously capture two images of the same fundus retinal region from different angles, and could enable three-dimensional reconstruction of the fundus. It comprised an imaging system and a lighting system. In the imaging system, the Gullstrand-Le Grand eye model was used to simulate the normal human eye, and the principle of stereomicroscope was employed to design the splitting optical system. In the lighting system, the stray light produced by the retinal objective lens was eliminated using a black dot board, and an annular diaphragm was used to prevent corneal light reflection. Simulation results reveal that the modulation transfer function of each visual field at the cut-off frequency of 95 lp/mm is greater than 0.2, the distortion of the imaging system is less than -3%, and the field curvature is less than 0.1 mm. The stereo imaging system is highly efficient in chromatic aberration correction and focusing, and can provide clear images for the human fundus with the diopters ranging from -10 m-1 to +5 m-1.

Although optical extensometers based on Digital Image Correlation (DIC) method have widespread applications in strain measurement, there is a lack of systematic research on the strain measurement accuracy. For the uniaxial tensile test, testing equipment including the traditional 2D-DIC optical extensometer, three-dimensional digital image correlation (3D-DIC) optical extensometer, 2D-DIC optical extensometer equipped with dual-reflector imaging capability, and strain gauge were employed to measure the strain of stainless steel simultaneously. The results of axial and transverse strain measurement using different optical extensometers were compared, and the corresponding sources of measurement error were analyzed. The experimental results revealed that the traditional 2D-DIC optical extensometer yields low measurement accuracy because of the presence of out-of-plane motion. The 2D-DIC optical extensometer equipped with dual-reflector imaging capability can eliminate the effect of out-of-plane rigid body displacement and thus, ensure higher measurement accuracy, especially for transverse strain measurement, compared to other optical extensometers. Notably, the strain measurement results obtained using the 2D-DIC optical extensometer equipped with dual-reflector imaging capability were in good agreement with those obtained using a strain gauge. The measurement accuracy of transverse strain using 3D-DIC optical extensometer is apparently lower than that of axial strain.

The control method of the drive axis was investigated with the goal of rapid startup of the MEMS gyroscope with a high Q-factor. The change rate of the vibration phase of the high Q-factor resonator with frequency was analyzed, and the reason for the long starting time of the phase-locked loop scheme in addition to its sensitivity to the initial frequency deviation was elucidated. The variation rule of the amplitude at the initial stage of the self-excited oscillation scheme with time was deduced by using the averaging method. A control scheme that combines the self-exited oscillation loop and the phase-locked loop (SEOL-PLL) of the drive axis was proposed. The self-excited oscillation mode was used to quickly start the gyroscope, and then the phase-locked loop was used to maintain the accuracy and stability of the vibration frequency. In the experiment, when the initial frequency deviation of the PLL is less than ±10 Hz, the starting time of the gyroscope is 2 s. Whereas with the SEOL-PLL scheme, a frequency error of 0.01% is achieved in 0.3 s and an amplitude error of 0.1% is achieved in 0.4 s, as long as the initial frequency deviation is within ±1 000 Hz. The “SEOL-PLL” solution considerably reduces the start-up time and is insensitive to the initial frequency deviation of the gyroscope. It is therefore suitable for the mass production of MEMS gyroscope and has a good adaptability to ambient temperature change.

A precise cable drive system is a flexible friction transmission method that utilizes the power transmitted from drive capstans to a driven pulley through a proper preloaded transmission medium. It has many advantages, such as flexible layout and high precision, besides being lightweight. It is widely used in many dexterous precise electromechanical devices and servomechanisms. The progress in the applicability of the precise cable drive is summarized in the aspects of precision pointing mechanism, harmonious robot, haptic interface mechanism, running robot, dexterous hand and endoscopic surgery. Next, the urgent issues of the basic theory and application research are dealt with, including studies on transmission mechanism, applied studies on servo systems, and engineering design methods for typical applications. Eventually, some suggestions regarding further research on theory analysis and engineering design technology of the precise cable drive system have been put forward.

A method of pulse electrochemical machining in discrete fields and domains was introduced to solve the problems in microchannel arrays machining on difficult-to-cut metal surface. This method has the characteristics of processing microchannels with high surface quality and a non-penetrating structure. The machining zone of each channel was separated by the specially designed fence-like template made of nonconductive material, and hence the electrolyte flow field and electric field are confined and improved, because of which a higher process stability and good machining accuracy and consistency can be achieved. Both simulation analysis and experiments verified the effectiveness of the method. The distributions of flow field and electrical field of this method are better than those of the through-mask Electrochemical Machining (ECM) method and electrochemical pattern transfer method. Nine channels with a 538.76 μm width, 25.78 μm depth, and an overcut of 19.38 μm are obtained within 1 min. Single-factor experiments were conducted and the results show that good process performance can be obtained with a short machining time, low voltage amplitude, low duty cycle, and high pulse frequency. Thus, the presented method of pulse EMC in discrete fields and domains is suitable for mass production of non-penetrating microstructures with the advantages of high efficiency, high quality, and low cost.

Images with resolution beyond the diffraction limit can be achieved by combining conventional microscopy with a microsphere. In order to position the microsphere on the field of interest of the sample surface and to expand the observation area, a method to manipulate the microsphere by combining it with a multi-axis translation stage was proposed in this paper. Images were obtained by scanning the microsphere, which was positioned accurately by driving the translation stage with four degrees of freedom. The influence of the probe on the super-resolution image was analyzed by performing an optical simulation. Kinematic analysis of the translation stage was studied for determining the manipulation strategy of the microsphere. Force analysis of the microsphere in a liquid medium was carried out to evaluate the possibility of detachment of the microsphere from the probe. By using a microsphere, the gap between the Blu-ray disc stripes could be clearly observed. The experimental results indicate that the amplification factor of the microsphere is 3.52 and a resolution of 130 nm (approximately λ/4) can be achieved. In addition, by scanning the microsphere along the S-shaped trajectory at a speed of 5×10-6 m/s over the sample surface, the super-resolution image over a large continuous area was achieved. Therefore, by using this method, the imaging area could be expanded and the observation efficiency was improved.

Secondary mirror adjustment mechanisms are core elements of active optics used for the adjustment of space telescopes. For real-time rapid adjustments in orbit, great demands are put on their adjusting accuracy and dynamic characteristics. Because of nonlinearity, uncertainty, disturbances, and coupling of the system, sufficient control cannot be easily achieved by traditional methods based on intermixing PID systems and inverse kinematics. Focusing on this problem, an uncoupled electromechanical model was proposed by transforming the joint couple dynamics into external disturbances. The model uncertainty was represented by the multiplicative uncertainty resulting from the system identification, and the mixed sensitivity function was designed according to the model error bound and performance index. By using the extended method to deal with the imaginary axis pole, the complex dynamic system was transformed into a stacked S/T/KS problem. A robust controller was designed by MATLAB and a digital algorithm was realized using DSP. The improved robustness, disturbance rejection, and dynamic characteristics were verified by both simulations and experimental results, proving that space-reliable application can be achieved.

To realize precise actuation of micro/nano-manipulation systems, a type of nano-motion platform based on the stick-slip principle was designed. Flexure hinges, a mass block, and an elastic component were integrated as an independent stator base. The stator consisted of a stator base, a piezoelectric stack, and a ceramic ball installed in the base bottom. The vertical position of the end of the elastic component can be adjusted by rotating an adjustment screw to change the pre-pressure between the stator and moving platform. Thus, the optimal driving force can be obtained. Because of the motion mechanisms of stick-slip driving and the influence of various parameters on the platform motion, mechanical modeling was carried out. Friction force was of key significance for stick-slip driving. In order to accurately express the friction mechanism of stick-slip driving, the LuGre friction model was introduced into the mechanical modeling. The simulation analysis was performed using MATLAB/Simulink software. The overall size of the stick-slip driving platform is 40 mm×40 mm×18 mm, and its mass is 32 g. Experiments show that the platform can achieve a minimum step size of 10 nm, its highest speed is 2.5 mm/s, and its stroke is 22 mm.

In order to overcome the shortcomings of traditional mechanical and electronic pressure switches, such as the production process is complex and it is difficult to integrate with the follow-up circuit, bulky and so on, the design and preparation of passive MEMS pressure switch were developed by using glass frit encapsulation technology with metal lead coverage in this paper. The overall structure of the passive MEMS pressure switch was designed to include the pressure sensitive film, the silicon island and the upper electrode on the silicon cover and the micro-barrier bumps, the glass paste and the lower electrode on the glass substrate. The key dimensions of pressure sensitive film, silicon islands and upper and lower electrodes were optimized by simulation. After three wet etching processes, double barrier bump, silicon island and pressure sensitive film were fabricated. The silicon cover, the glass substrate and the metal lead were then bonded together as a whole by glass frit thermal-pressure bonding process. The results show that the height of the double-barrier bumps and the thickness of the pressure-sensitive film are well controlled at 8 μm and 50 μm, respectively. And the threshold pressure of the MEMS pressure switch is 125 kPa.

In order to improve the applicability, maintainability and the cooling efficiency of piezoelectric computer chip water cooling system, a computer chip water cooling system driven by combined piezoelectric pump was proposed. Initially, the output performance of the combined pump units in piezoelectric water-cooling system with different combinations (series/parallel), pump quantity and the relative position were tested and analyzed under 220 Vpp square wave. Subsequently, a research on water cooling efficiency of piezoelectric water-cooling system was conducted, based on the experimental results of combined pump units. The experimental results indicated that as the double pump units working in series combination, the performance of the working units located at the first and tail position (AD) was optimal with maximum output pressure (25 kPa) at a frequency of 30 Hz. When four pump units were working in series combination, the maximum pressure (23.5 kPa) and maximum flow (13.5 mL/min) were obtained at 35 Hz and 55 Hz, respectively. While double pump units were working in parallel combination, the performance of the double-working pumps which were located at the first (AC) position was relatively poor; When four pump units were working in parallel combination, the maximum output flow (22 mL/min) and maximum output pressure (12.6 kPa) were obtained at 50Hz and 60Hz, respectively, through switching of series/parallel and the pump quantity. It was found that different combinations and pump quantities could acquire different cooling efficiency. The driving parameters of the combined piezoelectric cooling system are obtained, which provide a novel approach for the effective cooling of computer chips.

The detector is the core component of the photoelectric conversion in the satellite-borne, and once launched, it is almost impossible for the maintenance, repair and replacement. Using The screened commercial infrared detector without life data has the risk. The service life of the thermoelectric cooler(TEC) is the weak link of the life of detector assembly, and its reliability has an important influence on the normal use of the detector. Therefore, it is important to study the life test of the infrared detector assembly. Firstly, the working mode of the satellite-borne infrared detector and the failure mechanism of TEC were analyzed. The accelerated life test method with constant stress combined with fixed number truncation was developed to reduce the time cost.Secondly, the hardware and software of the infrared detector life test system was described in detail, demonstrated the equipment and temperature cycle of the infrared detector; Finally, the life test system was operated, the test time accumulated for about 170 days, and the temperature cycle of the infrared detector accumulated about 30 000 times. The test result indicated that the maximum changes of the relative photosensitivity of G12180 and G12183 were 1.45% and 4.44%, respectively. The maximus increments of the cooling current of G12180 and G12183 component TEC were 4.30% and 7.50%, respectively. It can be seen that the life and performance changes of the detector assembly meet the aerospace load requirements.

In order to improve the efficiency of transdermal drug dilivery and reduce the pain of traditional injection on the human body, the microneedle array is needed. In this paper, the fabrication technique for microneedle array was introduced. A microneedle array was obtained on a polymethylmethacrylate (PMMA) substrate by using two moving X-ray lithography and alignment X-ray exposure process using the synchrotron radiation source AURORA of Japanese Ritsumeikan University. A hollow microneedle array was fabricated. The different specifications had been successfully fabricated by using different mask pattern and X-ray lithography for different hollow positions. X-ray lithography alignment device had been developed for the problem of alignment during fixed X-ray lithography. The experimental results show that the device could achieve the preparation of hollow microneedle array. Then, the piercing text was carried out, and the results showed that the microneedle had enough strength. Besides, in order to achieve the goal of low cost batch replication of microneedle array, the reverse mold and replication experiments of microneedle array were also carried out, and the Ni solid microneedle array was successfully manufactured. Finally, in view of the distortion of the side shape of the microneedle array structure during the process of X-ray lithography, the simulation prediction was established for the moving X-ray lithography, the simulation prediction results were compared with the experimental results. The results showed that the simulation error of developing depth was 5%.

In order to realize the ECG monitoring under electromagnetic interference environment, an anti-EMI ECG monitoring system based on the optical modulator was proposed in this paper. Firstly, the calculation formula of output photocurrent was discussed, and the influence of half wave voltage and insertion loss on the sensitivity was analyzed. Secondly, the Lithium Niobate (LiNbO3) electro-optic crystal was used to establish experimental setup, and measure the ECG signals of five health subjects. Then the proposed system and electrical ECG acquisition system were used to detect the ECG signals of subjects under both of normal and electromagnetic environments. The measurement results showed that the proposed system could obtain clear ECG signals as same as the electrical ECG acquisition system under normal circumstances, but under the electromagnetic environment, the proposed system could obtain better ECG signals than the electrical ECG acquisition system. Finally, the signal-to-noise ratio (SNR) of the two systems were calculated. The calculation results showed that the SNR change of the proposed system was 0.54 dB(V2)/0.49 dB(V4) under the electromagnetic interference, and the SNR change of electrical ECG signal acquisition system was 2407 dB(V2)/16.75 dB(V4). Therefore, the proposed wearable ECG monitoring system can accurately monitor the ECG under the electromagnetic environment.

For decreasing the effect of background noise, the traditional imaging method of laser radar requires take long time in accumulation sampling and generating statistical histogram of photon countingto obtain the depth estimation of target. A 3D time-domain denoising algorithm based on photon-counting laser radar was proposed in this paper. Combined with the Poisson statistical model, the method proposed did not need to generate photon counting statistic histogram but used the different distribution feature of signal and noise in the time-domain, which increased the detection probability of signal photons and separated the signal from the noise to recover an accurate depth image of scene in the environment of low signal-to-noise rate. Experimental results demonstrate that the method increases the imaging accuracy by 3-fold at least comparing to the traditional maximum likelihood depth estimation. The method is conducive to the use of laser radar 3D imaging in high background noise environment and could broaden the application range of Lidar.

There are position errors between the images from each split optical path of the amplitude-division imaging polarimeters, image registration between the images from each split optical path is a precondition for polarized detection. In order to solve the problem that the features of the target are not obvious and difficult to extract, and that the change of gray scale of the image is large, a function of similarity measurement was proposed and it was suitable for the image registration of the amplitude-division imaging polarimeter. Based on the function, the images from each split optical path had been registered. First, according to the principle that the position error between the images can cause the abnormal information areas in the polarized image, the algorithm of extracting the similarity metric function was discussed. Next, according to the characteristics of the imaging system, the parameters of the geometric transformation between images were determined. Then, the genetic algorithm as the parameter optimization search algorithm was used to get optimal parameters of the geometric transformation between images, and the image registration algorithm was finished. Finally, the algorithm was verified by using constructed images and collected images respectively, and the image mutual information value (MI) is determined as the metric of the accuracy of image registration. The experimental results show that the MI between reference image and registered sensed image is 2.692 5 for the constructed image, and it is 1.849 3 for collected images.It indicates that the accuracy of registration by using this method is higher than that by using method based on feature. It can satisfy the registration requirements of the system.

Aiming at the problem of image quality degradation caused by insufficient illumination, a retractive algorithm of bright channel was proposed to compensate the illumination intensity of the image. The algorithm assumed that the local constant light could initially satisfy the uniformity of illumination and was similar to the scene, and the bright channel operation was used to estimate the weight of the light component. The problem of blocking was usually solved by the local processing, but this would make the compensation image texture blurred or even lost, and the fusion strategy based on image structure similarity was designed. Finally, the Retinex theoretical model was used to compensate for the light. The experimental results show that the proposed algorithm is simple and efficient, and can compensate for the low illumination area of image shadows or nighttime images. Compared with the traditional algorithm, the peak signal to noise ratio (PNSR) is improved by about 5 dB and the structure similarity (SSIM) increased by more than 7%. The algorithm in the pure software system PC (CPU frequency 2.4 G) processing 640×360 color video can reach 6-12 ms/frame, processing 320×256 infrared video to reach 4-10 ms/frame, to meet the needs of the project.

In order to realize the arbitrary shape object extraction from LiDAR point cloud data, a method based on irregular marked point process was proposed. Firstly, a random point process was defined on ground plan, in which random point positioned the object projection on the plan. Then the marks associating individual points were defined with a set of nodes to depict the shape of object on the ground plan. Assumed that the elevation values of ground points followed an independent and identical Gauss distribution, and that of objects were also characterized by Gauss distributions individually. According to the Bayesian inference, the object extraction model was obtained; The RJMCMC algorithm was designed to simulate the posterior distribution and estimate the parameters. Finally, the optimal target extraction model was obtained according to the maximum a posteriori. LiDAR point cloud data was extracted by using the proposed method. According to the experimental results, it can be seen that the detection accuracy of the algorithm is above 80%, the highest accuracy is 99.43%. In this paper, the traditional rule mark process is extended to irregular marking process, and it can be used to fit the geometry of arbitrary shape target effectively. Experimental results show that this method can effectively fit the arbitrary shape objects.

To improve the detection rate of pulmonary nodules in early lung cancer screening, a low-dose CT pulmonary nodule detection algorithm based on 3D convolution neural network was presented. First, the multi-directional morphological filtering algorithm was used to preprocess low-dose sequence CT image. The improved 3D region growth algorithm combined with the convex hull algorithm was used for lung parenchymal segmentation. Then the 3D candidate nodules were routed and illuminated in order to solve the convolution neural network on the sample imbalance sensitive issues. Finally, in situations of different network parameters, four groups of experiments were performed on the 50 sequences of low-dose lung cancer screening data in ELCAP database. The results showed that accuracy, sensitivity, specificity and ROC curve of the AUC values were 84.6%, 88.89%, 80.32% and 0.924 4 respectively by the constant optimization of network parameters. The proposed algorithm can correctly detect low-dose lung nodules, with the the accuracy, sensitivity, and specificity increased by 5.37%, 5.6% and 10.42%, respectively, which is more comprehensive and can provide effective help for lung cancer screening compared with conventional lung nodule detection algorithm.

Measurement of solid rocket nozzle thrust line is one of a key technique in the field of spacecraft precise installation and is a representative application of rotation axis line extraction of rotational symmetric object. Given that the existing methods have the problems such as nonobjective, poor reliability and low adaptability, an automatic solution for rotation axis extraction from 3D scanned high density of point cloud data was proposed, in which all the surface points and their normal vectors were utilized as constraints. Firstly, the normal vector of each point in point cloud was calculated and unreliable points were eliminated according to the standard deviation value of local planar fitting for normal vector calculation. Then, the initial value of rotation axis was achieved through planar and spherical fitting of reference points which had same latitude with the randomly selected reliable seed point. Finally, refinement result of rotation axis was calculated by solving the objective function listed based on the relationship between normal vectors of each reliable point and the rotation axis. The test experiments were performed, and the accuracy and precision of the method were verified by simulated and measured data. The experimental results indicate that the deflection degree is under 0.003° and the transverse distance is under 0.02 mm, which satisfies the requirements of rotation axis extraction of rotational symmetric object.

Reconstruction of 3D object is an important part in machine vision system, and the semantic understanding of 3D object is a core function for the machine vision system. In this paper, 3D restoration was combined with the semantic understanding of 3D object, a 3D semantic scene recovery network was proposed. The semantic classification and scene restoration of 3D scene were achieved only by using a single RGB-D map as input. Firstly, an end-to-end 3D convolution neural network was established. The input of the network was a depth map. The 3D context module was used for learning the region within the camera view, then the 3D voxels with semantic labels were generated. Secondly, a synthetic data set with dense volume labels was established to train the depth learning network. Finally, the experimental results showed that the recovery performance w improved by 2.0% compared with the state-of-art. It can be seen that the 3D learning network plays well in 3D scene restoration, it owns high accuracy in semantic annotation of object in the scene.

In order to solve the problem that the information from the source images is easy to interfere with each other which influences the quality of infrared and visible image fusion, a new image fusion method based on Guided filter, Gaussian filter and nonsubsampled directional filter bank was proposed. The low-frequency approximation components, strong edge components and high-frequency detail components were obtained by combining Guided and Gaussian filter. Then the high-frequency detail components were filtered to obtain the detail directional components with the use of nonsubsampled directional bank. The low-frequency approximation components were fused by a fusion rule based on regional energy and the strong edge components were fused by a strategy based on convolutional sparse representation. The detail directional components were fused by a rule based on improved pulse coupled neural network. Then the final fused results were obtained by using inverse transform through fusing the fused components. Experimental results show that the proposed algorithm outperforms traditional methods in terms of visual inspection and objective measures. Compared with the image fusion algorithm based on discrete wavelet transform and sparse representation, which possesses the better fusion effect in the traditional methods, the fusion quality indexes of the proposed method, such as Standard deviation(STD), Information entropy(IE), Mutual information(MI), Average gradient (AG) and Spatial frequency(SF) increased by 20.28%, 2.24%, 47.41%, 5.34%, 8.02% averagely.

The micro-vision images captured by the CCD camera typically show low gray value, non-uniform illumination, large dynamic range, poor contrast and missed or unrecognized fine structure due to the coaxial light source and optical diffraction in micro-vision system. Based on the raised cosine and variable gain sub-band decomposition algorithm, a micro-vision image adaptive enhancement method was developed to improve the image quality in this work. First, on the basis of image characteristic, the original micro-vision image was enhanced by using adaptive Log gain function to improve the contrast between minutiae and background which located in bright or dark regions. Then, the fast illumination estimation was carried out in terms of the adaptive cosine convolution. After that, the independent spectral sub-bands of the image in each channel were obtained through the adaptive variable gain sub-band decomposition algorithm. The ultimately enhanced image was generated after the implement of intensity correction, image fusion and color restoration. The algorithm had been successfully applied to a micro displacement measurement system and the relative error was less than 20%. Moreover, extended the algorithm to ordinary image processing can improve the contrast and minutiae. The relative increasing rates of the average quality of three experiments were 81.46%, 71.18% and 93.75%, respectively; and the average consuming time were 3.86 s, 0.24 s, and 1.27 s, respectively.

In order to reduce redundancy of detecting extreme point and various adverse effects of image change factors during using SIFT vehicle logo recognition algorithms. An improved SIFT algorithm based on edge constraint and global structure was proposed, it took advantages of the image moment invariant theory and the image edge detection algorithm to only detect edge regions of target image, eliminating extreme points that have nothing to do with vehicle logo recognition regions, and it divided each feature point neighborhood into circular regions and calculated the maximum curvature of pixel in each group of concentric circles that obtained by the division to construct the global SIFT combination feature vectors, which made the SIFT descriptors had a global describing nature. It also combined the SVM model such that a feature vector classifier of vehicle logo image was created to classify features and recognize vehicle logos. The simulation experiment data indicates that the improved SIFT vehicle logo recognition algorithm may reduce redundant extreme points by about 25 to 45 percent, which enhances the effectiveness of detecting extreme points, and make the average recognition rate reach more than 97 percent, which improves the real-time trait of recognition. It can be seen that higher recognition rate and faster recognition speed can be obained in comparison with several common image feature extraction operators.

In order to realize the three-dimensional measurement of micro electro-mechanical systems (MEMS), it is usually necessary to use a microscope or an optical profilometer. However, the traditional method has problems such as low precision and high cost of detection equipment, and it cannot achieve good results for an MEMS structure with many fracture surfaces. A new method for 3D measurement and image processing was presented in this paper. Unlike the three-dimensional reconstruction measurement method of multi-angle microscopes, this study firstly used a monocular microscope to obtain a series of single point measurements of the target, thus acquiring the depth information through single axial images. Then, a defogging algorithm was adopted for image preprocessing, denoising, and achieving effective information extraction. Subsequently, the depth information of the object to be tested was obtained using a focus measure algorithm. Finally, 3D fitting was performed using the data processing software. Based on the above principle, the focal plane array (FPA) was taken as the target to be measured. The experimental results show that the proposed 3D reconstruction method and image processing algorithm can obtain a more accurate FPA morphology, and can clearly show the release hole reflecting surface and the leg part on the reflector. The leg length of the FPA is measured to be 110.6 μm, and the pixel size of each reflector is about 120.8 μm×70.8 μm, which is consistent with the design value. It solves the problem of measuring the fracture surface, and reduces the difficulty and cost of microstructure measurement. The multi-directional image-fusion measurement technology using a monocular microscope is of great significance for MEMS 3D profile measurement. The defogging algorithm has significant application value in image fusion and 3D measurement image processing.