For the detection of small remote sensing targets with complex backgrounds， an improved multi-scale feature fusion-based single shot multi-box detector （SSD） method was proposed. First， a feature map fusion mechanism was designed to fuse the shallow high-resolution feature maps and deep feature maps with rich semantic information， after which feature pyramids were built between the feature maps to enhance small target features. Subsequently， the channel attention module was introduced to overcome the background interference by constructing a weight parameter space to provide more attention to the channels that focus on the target region. Finally， the scale between the priori box and the original map was adjusted to better fit the small remote sensing target scale. Qualitative and quantitative tests based on image datasets from a remote sensing aircraft were then performed， with the results showing that the proposed method improves the detection accuracy by 4.3% when compared with the SSD method and can adapt to complex multi-scale remote sensing target detection tasks without reducing the detection rate for small targets.

To identify the friction model parameters of a modular joint， an off-line identification method that compensates the joint friction is proposed. First， the structure and control system of the modular joint are presented， and the dynamic model of the joint is established. Second， the LuGre friction model is developed. The grey wolf algorithm and piecewise least-square algorithm with a pseudo random sequence are then used to identify the respective model parameters. The results of two methods are compared and analyzed， and a feed-forward compensation algorithm based on the LuGre friction model is designed and verified experimentally. The experimental results indicate that compared with the piecewise least-square method， the identification accuracy of the grey wolf algorithm improved by 19.2%； the joint velocity tracking error decreased from 0.295 （°）/s to 0.183 （°）/s when the given velocity signal was a sine wave with an amplitude of 1 （°）/s and a frequency of 10 Hz； and the velocity loop bandwidth increased from 12 Hz to 18 Hz after friction compensation. Several experiments are repeated， and the identified data exhibit a high repeatability， which verifies the suitability of the proposed method. The proposed feed-forward friction compensation algorithm can be used to improve the dynamic performance of the joint control system.

To explore the dehazing priors and constraints among the physical parameters during imaging under haze conditions and improve dehazing accuracy， we propose a decomposition–composition and recurrent refinement network based on the physical imaging model for image dehazing. Unlike existing dehazing methods， it contains a transmission prediction branch and a clear image prediction branch. Both branches are built based on the multi-scale pyramid encoder–decoder network with a recurrent unit that can utilize multiscale contextual features and has more complete information exchange. Considering the transmission map is related to the scene depth and haze concentration， the transmission map can be regarded as a haze concentration prior and guide the clear image prediction branch to estimate and refine the dehazing result recurrently. Similarly， the clear image that contains the scene depth information is regarded as a depth prior and guides the transmission map prediction branch to predict and refine the transmission map. Then， the predicted transmission map and clear image are further synthesized as the haze image that serves as the input of the network in each recurrent step， enabling the predicted transmission map and clear image to meet the constraints of the physical imaging model. The experimental results demonstrate that our method not only achieves a good dehazing effect on both synthetic and real images， but also outperforms existing methods in terms of quality and quantity. The average processing time for a single hazy image is 0.037 s， indicating that it has potential application value in the engineering practice of image dehazing.

Object detection is the basis of autonomous driving and robot navigation. To solve the problems of insufficient information in 2D images and the large data volume， uneven density， and low detection accuracy of 3D point clouds， a new 3D object-detection network is proposed through an image and point-cloud fusion with deep learning. To reduce the calculation load， the original point cloud is first filtered with the flat interceptor corresponding to the object's frame detected in the 2D image. To address the uneven density， an improved voting model network， based on a generalized Hough transform， is proposed for multiscale feature extraction. Finally， Normal Three-Dimensional Distance Intersection over Union （N3D~~DIOU）， a novel loss function， is extended from the Two-Dimensional Distance Intersection over Union （2D DIOU） loss function， which improves the consistency between the generated and target frames， and also improves the object-detection accuracy of the point cloud. Experiments on the KITTI dataset show that our algorithm improves the accuracy of three-dimensional detection by 0.71%， and the aerial-view detection accuracy by 7.28%， over outstanding classical methods.

Widespread video sensors record rainfall information continuously. Video-based rainfall data estimation， with high spatio-temporal resolution， has become one of the most promising methods of rainfall data collection to date. However， due to the complexity and variability of sensor devices， video scenarios， etc.， the quality of rainfall data estimated can often contrast between individual visual sensors. Further processing is required to ensure the quality of rainfall inversion results. Inspired by Tobler's First Law of Geography， this study presents a precision control model （PCM） for video-based-rainfall inversion results correction. The model uses the spatio-temporal information between camera nodes， within the Visual Sensor Network， as the constraint. Rainfall events were analyzed from the dimensions of spatio-temporal consistency， situational consistency， and correlation， to achieve a high-precision representation of rainfall data. A multi-granularity filtering method was adopted for rainfall inversion using mutual verification of rainfall information among video nodes. The experimental results show that the PCM model can effectively improve rainfall inversion accuracy and stability in various rainfall scenarios. The mean value of the relative error of rainfall intensity （RI） is reduced by approximately 14.85% in light or medium rainfall scenarios， and approximately 19.90% in heavy or violent rainfall scenarios； For the standard deviation of the related error of RI， approximately 40.87% reduction for medium and light rain scenarios， and approximately 40.96% reduction for heavy rain scenarios. The results of this study confirm that the proposed PCM can provide support to produce high-quality rainfall data.

The key step of optical interferometry is phase unwrapping， which is expected to be computationally fast， highly precise， and widely applicable. According to the feature of wrapped phase that between different order fringes there are significant edges, a fast unwrapping algorithm based on region segmenting with mathematical morphology(RSMM) is proposed. First， mathematical morphology is applied to extract the boundaries and segment regions from the phase map. Then, phase differences between adjacent regions are calculated in order to determine the phase order and elevated quantity of each region, and so are phases of the pixels on boundaries. Finally， wrapped phases in regions and boundaries are elevated individually according to the quantified elevation to obtain the unwrapped phase map. Simulations and experiments indicate that RSMM requires less than 1 second to unwrap and generate a phase map for 1 000×1 000 pixels， and this required time is less than a quarter of the computation time of conventional least-square algorithms. In addition， the phase unwrapping performance is not influenced by phase boundary， data dropout， and noise. The RSMM algorithm has the advantages of high speed， broad adaptability， and high accuracy and is promising for measurement applications with a commanding requirement for computation speed， such as dynamic interferometry， optical holography， and fringe projecting profilometry.

To ensure high accuracy and a wide measurement range for wind vector measurement based on ultrasonic sensors in mixed noise containing α and Gaussian noise， a novel FLOM-based dual-phase measurement method is proposed in this paper. First， the FLOM operator is used to suppress mixed noise containing α and Gaussian noise； this eliminates the shortcomings of the traditional second-order moment and high-order cumulant， which cannot be used for mixed noise containing α and Gaussian noise. Then， the time delay estimation method is transformed into the phase estimation method， and a FLOM-based dual-phase measurement method based on the orthogonality of the reference signals is proposed. This method effectively eliminates the influence of the amplitude variation on the measurement accuracy. The simulation results show that the measurement accuracy and measurement range of the proposed method are higher than those of the traditional time delay estimation method under wind speeds of 0-70 m/s. Even when the SNR is -10 dB， the RMSE of wind speed measurement is less than 1.5 m/s， and that of wind direction angle measurement is less than 2°. Practical application results show that the RMSEs of wind speed and wind direction angle measurement are 0.104 m/s and 0.54°， respectively， under strong winds. The proposed method can estimate the wind vector in mixed noise containing α and Gaussian noise more accurately than the time delay estimation method can.

The articulated arm coordinate measuring machine （AACMM） is one of the portable non-orthogonal coordinate measuring instruments with a series structure and an angle error that has an amplifying effect on the measurement accuracy， making it difficult to improve the measurement accuracy. To solve the above problems， the AACMM kinematic modeling method with a tilt motion error compensation for the rotation axis was proposed based on the DH model. The method for separating the tilt motion error for the rotating axis was studied， and the corresponding test system was built. The calibration model for the structural parameter errors was established based on the spatial distance， and the corresponding experiments were performed. The experimental results show that the measured standard deviation for the AACMM whose kinematic model included the function of compensating the tilt motion error for the rotating axis is reduced from 0.055 mm to 0.037 mm， compared to that for the DH model. The model can effectively promote the accuracy of AACMM.

Micro-electromechanical systems （MEMS）-based extrinsic optic fiber Fabry-Perot ultrasonic sensors have the characteristic of anti-electromagnetic interference. They have features such as long signal transmission distance， small size， and light weight， as well as good detection and positioning ability for ultrasonic signals released by partial discharge， Owing to these properties， they have potential for numerous application prospects. Currently， most of the sensitive structures that have been reported to use this type of sensor typically have completely circular diaphragms and are relatively simple to process. However， the temperature difference or pressure imbalance between the inside and outside leads to deviations. Thus， reducing or eliminating this effect is a basic prerequisite for promoting the industrial application of sensors. In this paper， we propose an optic fiber Fabry-Perot ultrasonic sensor based on a porous sensing diaphragm. The sensor was manufactured using the MEMS process with a thickness of only 5 μm. The results showed that the sensor had a good ultrasonic response in liquids. Moreover， the static pressure sensitivity could reach 1.25 V/Pa， and the performance of the distance decay and directional response was consistent with that in air. Additionally， the pores could prevent the air pressure imbalance inside and outside the sensing diaphragm. Therefore， the proposed sensor has potential for applications in the detection of partial discharge in liquids.

This study was aimed at dealing with the reduction in the precision of the angle measurement system in precision reducers due to the installation tilt and the deformation of the circular grating. The traditional error compensation method is limited by the number of working faces of the multi-sided prism， making the compensation effect insufficient. Thus， an error compensation method based on multi-data correlation was proposed. The autocollimator was combined with a regular 24-sided prism and used to calibrate the angle measurement error of the circular grating. By changing the relative position of the polyhedral prism and the circular grating in the circumferential direction， several groups of calibration data for the measurement of the angle error were obtained. The harmonic wave error compensation method was then used to preprocess multiple groups of measurement data； subsequently， the different groups of data were correlated and trained by a back propagation neural network to obtain the angular error model for the circular grating. This method does not need to employ the compensation formula， and it can also overcome the problem of fitting oscillation data when there are several sample points in the harmonic wave error compensation method. The experimental results show that the peak value for the residual error of the proposed compensation method is within 0.94" and the standard deviation is 0.25"； these values are 33.3% and 37.9% less than those of the traditional harmonic wave error compensation method， respectively. The proposed method effectively improves the accuracy of the angle measuring system to a sub-arc second level.

To explore the damage law and evolution mechanism of silicon wafers during precision grinding， a single diamond grain cutting experiment of a monocrystalline silicon wafer was carried out. The characteristics of the fracture damage morphology at the entrance of a scratch， inside the scratch， and at the outlet of the scratch were analyzed when the edge of the silicon wafer was covered by an adhesive and when it was not. Thus， the close internal relationship between the AE intensity， grinding force， cutting depth， friction coefficient， and crushing damage was established. With an increase in the loading pressure or penetration depth， the damage on monocrystalline silicon becomes increasingly serious， and the intensity of the acoustic emission signal with release increases. The critical threshold conditions for the internal fragmentation of monocrystalline silicon are as follows： the load， penetration depth， and acoustic emission intensity should be approximately 80 mN， 2 μm， and 10%， respectively. The toughening effect of the adhesive coating on the edge of the monocrystalline silicon wafer was remarkable. The critical threshold conditions for the edge chipping of monocrystalline silicon are as follows： a load of approximately 800 mN， a penetration depth of approximately 6 μm， and an acoustic emission intensity of approximately 55%.

The existing artificial anal sphincter clamping mechanism has the disadvantages of low biocompatibility and long response time. Based on the defecation mechanism and physiological structure， this paper discusses the design of a Novel Closed-link Artificial Anal Sphincter （NCAAS） clamping mechanism. The NCAAS clamping mechanism consists of three sets of interlaced clamping arms and a transmission mechanism based on groove cam swing rods. To study its biocompatibility and response time， the clamping force equation of the clamping arm was derived from the principle of virtual work， and the processing of the finite element in instantaneous dynamic simulation. During simulated stool control， the maximum clamping force of the NCAAS is 1.6 N， and the anorectal angle is in the range of 62.2°~95.2°， which meets the demands of human daily stool control mass. The weight of the NCAAS prototype is 55.19 g， with a height of 42.7 mm and length of 68.2 mm. The response time of the improved system is 7.25 s. An in vitro experiment with a pig colon verified the biocompatibility of the NCAAS， which draws the result of a 700 g controlled stool mass at an anorectal angle of less than 90°. The NCAAS is small and light， with better biocompatibility and a response time that is significantly shorter than that of traditional artificial anal sphincters.

To improve the prediction accuracy and robustness of the spindle thermal error compensation model of computer numerical control （CNC） machine tools， this study investigates the application of the principal component algorithm to the thermal error modeling of CNC machine tools. First， a selection algorithm of the temperature sensitive point and thermal error modeling algorithm based on principal component algorithm are proposed. Second， a three-axis vertical machining center is used to measure the spindle thermal error over an entire year. Thereafter， the principal component regression （PCR） model of the spindle thermal error is established based on the experimental data obtained. Then， the prediction accuracy and robustness of the PCR model are compared with those of the multivariate linear regression， back propagation （BP） neural network， and ridge regression models. The experimental results show that the PCR model has the highest prediction accuracy （6.8 μm） and robustness （2.4 μm）. Finally， the developed PCR model is used to predict the thermal errors of machine spindles that operate according to the speed spectrum. In this case， the model exhibits a prediction accuracy and robustness of 6.12 μm and 3.43 μm， respectively. Finally， the PCR model is embedded into the thermal error compensation controller for performing thermal error compensation experiments to verify the effectiveness of the proposed modeling algorithm.

It is difficult for three-dimensional （3D） depth information to be detected by monocular cameras， and there are limitations of binocular stereo vision when the surface features are not apparent. In addition， the proportional-integral derivative （PID） controller parameter tuning is not ideal. To address this， linear structured light technology is used to detect vibrations of the flexible manipulator， and the bacterial foraging algorithm is used to optimize the PID for vibration control. First， a linear structured light visual vibration measurement system is developed， and a mathematical model is established. Then， the visual calibration is completed， and the feasibility and accuracy of the proposed method are verified. By using the bacterial foraging algorithm， the parameters of the PID controller are optimized， the simulation results are compared with the results of the empirical tuning method， and the active vibration control experiment of the flexible manipulator is performed. Experimental results show that for the first-order modal vibration of the flexible manipulator， the average effect of the PID control method based on the empirical method control is 25.65% compared with the average effect of 39.32% when using the bacterial foraging algorithm. Consequently， the superiority of the proposed bacterial foraging algorithm when optimizing PID controller parameters is demonstrated.

Non-contact diffuse optical tomography （DOT） is a novel technique exhibiting higher spatial detection density compared to traditional DOT. The dynamic range of the detector in non-contact DOT considerably influences the accuracy of the reconstruction of the absorption coefficient. In this study， the effect of low dynamic range of the detector on the reconstruction results was examined by conducting experiments and simulations. A method based on deep learning was proposed to extend the dynamic range of the detector. The training samples were generated in the NIRFAST software by setting different phantom absorption and scattering coefficients and incident light field parameters. A fully connected network was built for model training. The model was used to recover the detection data in the simulations and optical experiments. The data recovery and reconstruction results of the experiments indicate that the model recovers the error data of the detector with a low dynamic range while extending the dynamic range of the detector from 256 to 109. The proposed method can effectively reduce the reconstruction error caused by the low dynamic range of the detector and provide an effective technical solution for non-contact DOT using detectors having a low dynamic range.

In order to solve the problem of unclear images in the rough obstacle avoidance link of lunar landing in weak light environments， this study proposes two image preprocessing methods， namely "region" pixel binning and camera background value removal， to improve the imaging sensitivity， image signal-to-noise ratio （SNR）， and contrast ratio of lunar landing. First， based on the analysis of the principle of traditional pixel binning， a region pixel binning method based on the n~~taps imaging data format is proposed. Then， according to the characteristics of large-scale obstacle recognition in lunar landing rough obstacle avoidance， the image contrast is improved by removing the background value of the camera. Finally， the methods of pixel binning， camera background removal， and the combination of the two methods are repeated using two cameras that have wide and narrow fields of view. The experimental results show that the SNR and contrast of images can be improved effectively by combining pixel merging and camera background value removal under weak light environments. In 2~~Binning mode， the SNR and contrast of the camera with a wide field of view can be improved by 5.901 4 dB and 0.254 7， respectively， and the SNR and contrast of the camera with a narrow field of view can be improved by 5.764 4 dB and 0.265 4， respectively. In 4~~Binning mode， the SNR and contrast of the camera with a wide field of view can be improved by 11.689 9 dB and 0.210 2， respectively， and the corresponding values for a camera with a narrow field of view can be improved by 11.401 5 dB and 0.284 0， respectively.

An optical film with microstructures for use in MiniLED backlight modules is designed to reduce the backlight module thickness. First， two diffusion principles based on the diffusion particle and refraction lens are analyzed. It is found that when the optical distance （OD） is very small， none of the principles are effective in ensuring that the light is fully diffused in order to meet the requirement on uniformity. Then， based on the principle of total reflection， the surface source is discreted to a cluster of point sources， the energy returned by a single point source to the lamp board is analyzed， and the energy returned by all the discrete point sources is accumulated. Notably， the profile of a microstructure can be decided when the energy returned by the surface source is maximum. The simulation results show that when the double-layer microstructure optical film is placed on the surface of a MiniLED chip， the brightness uniformity is 79.9%， which can be further increased to 89.2% by adding a diffusion film at an OD of 0.9 mm. Maskless direct writing lithography is used to prepare a sample optical film. The experimental results show that the brightness uniformity of the double-layer microstructure optical film placed on the surface of the MiniLED chip is 79.6%， which can be increased to 88.7% by adding a diffusion film at OD=0.9 mm. The optical film thus designed can be used to realize the ultrathin backlight module without the requirement of a precise position and therefore presents strong practicability.

To monitor the quality of sapphire substrate processing， a specific system for sapphire substrate surface morphology measurement and evaluation was developed. The classic vertical scanning white light interference technology was adopted. Based on the fluctuation of the interference images， the cross-section variance calculation method was used as the definition evaluation method to realize the automatic focusing of the system. Referring to the substrate measurement points and their location distribution recommended by the national standard GB/T 29505-2013， the corresponding measurement points for different sizes of sapphire substrates were increased according to the distance between adjacent centers of the same layer unit circle. Two-dimensional and three-dimensional roughness evaluation parameters were selected as the system evaluation parameters. The multi-point location roughness statistical parameters were analyzed. The error of the system was compensated by standard parts with Ra values of 0.1， 0.2， 0.4， and 0.8 μm， and the error compensation equation was obtained. The commercial 3D optical profilometer Zygo7300 was used to perform comparative experiments， and five different positions of the standard part with 0.8 μm Ra were measured. The error range of Ra measured by the two interferometers is -0.012-0.011 μm. After the error compensation， the 2-inch and 4-inch sapphire substrates were measured and evaluated. The experimental results show that the accuracy of the measurement system has a nanometer level， and it can measure and analyze the surface morphology of sapphire substrates with different sizes as well as evaluate the surface quality distribution.

To solve the problem of ultra-low stray light in highly sensitive refractive optical systems and super narrow spectrum detection systems， the theory of ghost images formed by the focal plane is used. First， according to the characteristic parameters of the photosensitive surface， the theory of ghost images formed by the lens and focal plane was proposed. Subsequently， the Light-tools software was used for modeling and analysis. The principle， position， and energy relationship between the ghost points and the image were calculated and simulated. Finally， a series of ghost image experiments were designed and carried out to validate the accuracy of the data. The simulations and experiments indicated that the ghost image and the original image were isometric， confocal， and centrosymmetric. The results showed good agreement between the conclusions of the simulations and experiments. Thus， the validity and rationality behind the formation of ghost images were proved. The results can be used as a guide for the design of other refractive optical systems.

Endoscopic biopsy is the main approach to the clinical diagnosis of early gastrointestinal cancer to date. However， this approach warrants a long period to obtain the final diagnosis. Endocytoscopy is a type of endoscope with ultra-high magnification， which， combined with intraoperative staining， can directly observe the pathological structure of the lesion such as the nucleus in vivo. To make endoscopists more accurately analyze the pathological features of the nucleus during the operation， a nuclear staining and segmentation method was previously developed for the esophageal mucosa tissue of pigs based on the endocytoscopy system with high magnification. Firstly， 1% toluidine blue was used to stain the nucleus of esophageal mucosa tissue， and the stained nuclei were observed successfully under the microscopic imaging mode of endocytoscopy. Based on this， the deep learning method was adopted to train the nuclear segmentation model， which effectively realized the segmentation and extraction of stained nuclei. The pixel accuracy reaches 99.23%， specificity of 99.54%， sensitivity of 84.37%， and the Dice of 0.813 8， laying a foundation for the study of artificial intelligence-assisted diagnosis of endocytoscopy.

Fiber Bragg grating strain gauges are usually used to access the structural strain and monitor the health of the structure. However， the performance of the structural health monitoring system is dependent on the status of such a large number of sensors， which are quite easy to break down after a long-term service in various conditions. Based on the phenomenon that similar characteristics will be observed for the outpout of the fiber grating strain gauge in the same structure， we propose a novel Malfunction-diagnosis method for the sensors in the fiber Bragg grating strain gauges. The proposed method extracted the eigenvalue s （i.e.， sample data length， standard deviation， energy value and principal component period） using signal processing algorithm. The eigenvalue convergence center points were determined by loop iterations. The eigenvalue distances from the convergence centers were standardized and merged into a comprehensive index for the health monitoring of the sensors. The simulation proves that the malfunction points can be effectively identified when the number of malfunction points is below 20% of the total points. The proposed method was used to monitor the 416 fiber Bragg grating strain gauges of five-hundred-meter aperture spherical telescope （FAST） health monitoring system， the results showed that the eigenvalue lists of 317 sensors could be extracted and 4 malfunction points and 14 abnormal points could be identified. Instead of a large amount of training with prior knowledge， the proposed method could provide reliable malfunction diagnosis for the sensors in fiber Bragg grating strain gauges just based on the characteristics of the data， which might be practical useful for the health monitoring of the FAST structure.

An open resection of liver tumors is the most effective method for their treatment. The intraoperative three-dimensional （3D） reconstruction of liver surface based on binocular structural light plays an important role in the accurate localization of liver tumors. However， due to factors such as illumination in the surgical environment and the location of images captured by the binocular camera， a local brightness saturation occurs in the captured images， leading to hindrance in the three-dimensional reconstruction of the liver surface. To solve this problem， an adaptive optimal fringe grating based on the segmentation projection algorithm of different reflectivity regions is proposed in this paper. First， a physical model and a sketch of illumination on the liver surface under binocular system were established， and the influence of local brightness saturation on the liver surface was analyzed. The region was then divided according to the different reflectivity of the liver surface. The intensity of the adaptive optimal fringe grating was calculated according to the region segmentation projection algorithm of different reflectivity， so as to achieve the partition projection of the liver surface. The saturation points of the left and right camera images were transferred to the streaks cast by the projector. The effectiveness of the proposed method to overcome the local brightness saturation of the liver surface was verified by the local brightness saturation experiment. Finally， the experimental results of porcine liver showed that the deletion rate of porcine liver 3D reconstruction surface decreased from 1.1%， 2.9%， and 1.4% to 0%， respectively， under three different operating conditions due to local brightness saturation， and the accuracy of porcine liver 3D reconstruction reached 0.75 mm. The method described in this paper can essentially meet the requirements of doctors， including the intraoperative 3D reconstruction accuracy of liver surface to be less than 1 mm with no defects in the reconstructed surface.