By self-adaptive feature selection, the traditional mean shift tracking algorithm was improved and its robustness was strengthened for object tracking in the complicated circumstance. Since one or two fixed features (such as the color) were usually selected for object tracking in traditional mean shift tracking algorithm, object tracking would be failure in the changeable circumstance. The remarkable and non-remarkable features were determined respectively by analyzing the distinguishing degree between candidates of object feature tracked and changeable background so that the most effective features were then selected to achieve object tracking in the complicated and changing circumstance. The reliability of the improved algorithm has been verified in serial experimental results of moving object tracking in different circumstance.

The relationship between different radar targets is often nonlinear due to the complexity of target movement and environments, so the recognition rate will decrease when traditional methods for linear dimensionality reduction are used. For this reason, the nonlinear manifold structure characteristics of ISAR 2D images were analyzed in detail, and then the Spatially Smooth Locality Preserving Projections (SSLPP) algorithm of manifold learning was used for feature extraction and dimensionality reduction. Moreover, three kinds of aircraft targets were classified by k-nearest neighbor classification. Compared with other traditional subspace methods, SSLPP algorithm can earn more local information of the image space by considering the spatial relationship between pixels in ISAR 2D image sufficiently. The simulated experimental results suggest that SSLPP algorithm has better classification performance than PCA、LDA and LPP algorithms in ISAR target recognition.

According to the signal characteristics of inertial sensors in optoelectronic tracking system, an improved Kalman filtering method was designed, by which the AR model of denoised signals was built up and parameters were estimated after the signals were filtered by wavelet. Because the signals filtered by wavelet is approximative to actual ones, the variance of system noises was obtained by the variance of observation noises multiplying a coefficient less than 1. Finally, the experiments about real signals of some inertial sensors verify that the algorithm is efficient and has some catholicity.

For the problem of passive location under low detection probability, a multi-sensor fusion tracking algorithm based on sliding window batch technique is presented. Multiple low detection probability passive sensors are networked to spatially accumulate the target information and improve the detection probability. The nonlinear measurement is transformed to pseudo linear measurement by using pseudo linear estimation，and then the local optimal estimation of target state is achieved by a least squares sliding window batch algorithm. Simulations verify the validity of algorithm and analyze the effect of parameter. Results show that, the proposed method greatly improves the detection probability of the sensor network with acceptable tracking precision and real time requirement.

Electrostatic target tracking is highly complex and strongly nonlinear and the experimental data are incomplete and uncertain. In order to improve filtering accuracy and reduce computational complexity, a central difference method was applied to Extended Kalman Filter (EKF) for calculating the nonlinear function and a new improved EKF algorithm was formed based on the characteristics of the spherical electrostatic probe. The improved EKF was applied to electrostatic detection tracking. Theoretical analysis and simulation experiment show that, compared with EKF and Unscented Kalman Filter (UKF), although calculation time of the improved EKF is slightly longer than that of EKF, it is less than that of UKF. Moreover, the tracking accuracy of the improved EKF is significantly higher than that of EKF, and is nearly the same with that of UKF.

Based on background difference and level set, a novel segmentation method was presented for multiple moving targets in static camera surveillance. Firstly, background pixels in current frame were obtained by calculating symmetric difference of three consecutive frames, and the gray value of every background pixel was recorded. Then the gray value with the biggest frequency was selected for each pixel to reconstruct whole background. Secondly, the region growth algorithm based on the 8-neighbour search was proposed to detect connected regions, and small background noise regions were removed by setting a threshold. Then by analyzing connected regions, moving target regions were marked. Finally, the level set algorithm without re-initialization was adopted to segment each target region. The closed and entire target contour was gotten. Experiments show that this approach can realize automatic detection and contour segmentation of rigid or non-rigid moving targets in static camera surveillance.

To solve the problem of video-based security surveillance, a real-time method for abandoned object detection is proposed. First, frame-to-frame difference is used to obtain the pixel-wise properties of motion, and two backgrounds are constructed to detect dual foregrounds. Then, the pixel-wise properties and both foregrounds are integrated to maintain the dual backgrounds. Since the false alarms and occlusions necessarily occur and degrade the detection performance in practical applications, an evidence image is aggregated to validate the abandoned object. Experiments on several different video sequences demonstrate that the proposed method is effective in detecting abandoned objects from cluttered scenes. In addition, this method runs around 54 frames per second for a sequence with an image resolution of 352×288, which can be applied to real-time surveillance tasks.

To improve the Signal-to-noise Ratio (SNR) and detecting probability of small target in infrared image sequences, a novel method of target detection based on spatial-temporal filtering is proposed. This method considered the real-time characteristic of the detecting algorithm. Firstly, the first frame was set as a reference frame and the other frames were compensated in motion. Secondly, the frames compensated were filtered by variance Weighted Information Entropy (Variance WIE) filter in spatial domain. After that, a bi-directional inter-frame difference method was adopted in temporal domain. At last, by detecting the protruding spot of the difference images, the background and noises in infrared images were inhibited and the small target was detected. The experimental results were analyzed and compared with some other methods. The results show that this approach can highly improve the SNR of target, precisely detect the small infrared target, and express fine real-time performance.

The performance of a particle filter is strongly influenced by the choice of proposal distribution. In order to improve the performance of particle filter for target tracking, a particle filter tracking method based on motion detection is proposed to improve the proposal distribution. A new proposal distribution, which integrates the motion information of the current frame with the prior distribution, is developed. A part of the particles is sampled from the system transition density, and the others from the motion region detected by using the Gauss background modeling. Thus, the prior distribution of particles is determined by both the system transition density and the observations. The experiments show that the method is very effective under the moving background and the occlusion circumstances.

In order to study the characteristics of spectral radiance measured by imaging spectrometer with motion compensation, the relationship between measured radiance and the tilt angle of the axis of Instantaneous Field of View (IFOV) was derived. In the Visible-near Infrared (VNIR) (0.4~1.0 μm), the ratio of the spectral radiance measured by imaging spectrometer with a certain tilt angle α to that with tilt angle α=0o (corresponding to observing nadir point) was calculated. Compared with observing nadir point (tilt angle α=0o), the measured total radiance decreases with the tilt angle α increasing in the process of motion compensation. And the proportion of measured target radiance in measured total radiance also becomes small. In order to make the increment of measured radiance keep or approach the gain by design in the process of motion compensation, the tilt angle of axis of α should be less than 30o.

An improved centroid algorithm was put forward considering the actual pixel photosensitive area shape and pixel fill factor of image sensor, which is based on the characteristic of different pixel photosensitive area shape and pixel fill factor with different coordinate values of the photosensitive area center. The locating accuracy was improved by modifying the values of pixel coordinates in traditional centroid algorithm. How to simulate a star point was described by analyzing gray scale distribution characteristic. The effectiveness of the improved centroid algorithm was verified by simulation. The simulation results show that the locating accuracy of the improved centroid algorithm is superior to the bilinear interpolation centroid algorithm and the traditional centroid algorithm both in ideal and noise situations.

A new method to increase the applied bandwidth of piezoelectric Fast Steering Mirror（FSM）is proposed. Firstly, according to the Lagrange Equation, dynamic equation for simplified model of this FSM was built based on radial fasten thin plane. Then, the final theoretical expression of the steering resonant frequency was deduced for purpose of predicting the dynamical performance of the FSM. On the base of the theoretical results, we fabricated this structure of thin plane, added it to the original FSM and took the frequency-response and angle-stroke experiments. Both the experimental results and theoretical analysis results proves that the radial fasten thin plane is useful in increasing the applied bandwidth of FSM.

A novel algorithm for image fusion was put forward. Based on lifting wavelet transform, the nonsampled lifting wavelet transform with the trait of shift invariance was obtained by cancelling the odd and even split process of lifting wavelet transform. When the source images were decomposed by the nonsampled lifting wavelet transform, the spatial frequency ratio for the low-frequency was defined. Then the low-frequency coefficients were fused by means of the combination of weighting and selection, fusion genes for the low-frequency field were computed by corresponding regional spatial frequency ratio, and the high- frequency was fused by weighting fusion algorithm based on edge. Finally, the fused image was obtained through inverse nonsampled lifting wavelet transform reconstruction. The experimental results show that the proposed image fusion algorithm could describe the grey discontinuous information in a better way. Moreover, compared with traditional algorithms, obtained fused images contain more abundant details with features.

For the sensitivity of Deng algorithm to noise, an adaptive image enhancement algorithm is proposed. According to the human visual properties, the algorithm uses the basic principle of the Deng algorithm to get automatic parameters of method adaptively. It can not only keep the original advantages such as easy realization、quick calculation, but also play a certain role in restraining noise. By comparing the result of adaptive image enhancement algorithm and that of other image enhancement algorithm, the adaptive image enhancement algorithm can get automatic parameters of method adaptively. Besides the easy realization of original Deng algorithm, it can improve the performance in restraining noise.

A novel algorithm for detecting corners is presented based on Curvature Scale Space (CSS) and Multi-scale Curvature Product (MSCP). Firstly, the corners of an image are detected at different curvature scale space. Then, a multi-scale curvature polynomial is defined as the sum or multiplication of the curvature of the contour at each scale. The new method can not only enhance curvature extreme peaks effectively, but also suppress noise and prevent smoothing some corners with augment of the scale. In addition, the concavity and convexity of detected corners can be judged by the result sign of the curvature polynomial. Experiment results show that the new method is more effective in corner detection than the other algorithms mentioned in the paper.

Lots of existing image watermarking methods in Discrete Wavelet Transform (DWT) domain cannot resist lightness-adjusting attack. Therefore, a novel image watermarking method for resisting lightness-adjusting attack in DWT domain is proposed in this paper. In the method, watermark bits are embedded into DWT low frequency coefficients of original image and are extracted from the revised DWT low frequency coefficients that resist lightness adjustment of watermarked image. The method is simple and is executed rapidly and effectively. Moreover, original image is not needed when extracting watermark. Experimental results show that the method has a fine transparence of embedded watermark and is very robust to brightness-adjusting attack. Furthermore, it is strongly robust to other ordinary attacks such as re-sampling, color dithering, smoothing, noise adding, and lossy compression.

A numerical analysis model for computing the temperature field of the tank power compartment armored plate was set up. The infrared radiation feature of the tank power compartment armored plate was calculated in theory. Solid wall surface temperature field was computed by adopting coupling calculation method, coupling heat transfer between armored plate and air was computed and effect of sun irradiation on board temperature field was considered. Then, temperature field distribution and radiation spectrum intensity distribution were forecasted. The relative error of the calculating values is less than 9.034%, which can basically meet the requirements of engineering applications. The study result indicates that the top board of power compartment is the mainly source of radiation. Exhaust shutter and top armored plate near exhaust pipe are the evident region of IR radiation character of tank.

According to infrared image characteristics, an infrared image segmentation method for improved Chan-Vese model based on wavelet multisale transform is presented. First, edge information is extracted by using wavelet multiscale transform, and it is added into energy function of Chan-Vese model to improve local control ability. Furthermore, internal energy term is added, which constraints level set function to approach signed distance function so as to avoid reinitialization of level set function. Then, wavelet low frequency subbands are segmented with improved Chan-Vese model from top to bottom, and segmented results are interpolated into next subbands as initiative contours. Finally, the method can realize infrared image segmentation. Experiments show that the method is very efficient and robust to noise.

In order to test the nanometer films fast, a method of ellipsometry based on Stokes parameters with a Division-of-amplitude Polarimeter (DOAP) was introduced. The structure of instrument, theoretical calculation and parameters testing were investigated. Firstly, a DOAP was designed, in which the light path was in the same plane. This was calibrated by known polarization that combined a polarizer with one quarter-wave plate. Then, the Stokes parameters could be tested quickly by use of four photoelectric detectors with the same properties. Secondly, the ellipsometric parameters (ψ, Δ) of a nanometer film were worked out by testing the Stokes parameters of incidence and reflection light on a sample with the DOAP. Then, the thickness and refractive index of the sample were obtained. The testing error of Stokes parameters was discussed. Finally, the parallel experiment with an ellipsometer in extinction was done. The experiment shows that the related coefficient of the thickness and refractive index are greater than 85% and consistency of two instruments is good. With the conclusion, the nanometer films can be measured fast by testing the polarized states of incidence and reflection light, and the DOAP can be set up easily, adjusted conveniently and calibrated simply.

Images captured from CCD cameras installed along railway are heavily distorted in wheel profile dynamic inspection system based on light-section method and image measurement technique. This paper analyzed causes of image distortion captured by CCD cameras, and designed the factorial of 2, 3, 4 calibration model. The results of each calibration model were compared by processing the same distortion image of checkerboard. And the calibration model, which is fit to Wheel Dynamic Inspection System, was selected. Based on selected calibration model, revised results of center district and outer district of image were analyzed quantitatively. Furthermore, the selected calibration model was applied on wheel profile image of high speed train captured by Wheel Dynamic Inspection System and the revised result of wheel profile was given. Field test shows that the designed calibration algorithm achieves a high precision of 0.1 mm, which meets the demand of wheel inspection in motion.

In order to eliminate the influence of the phase-shifting error on the quality of reconstructed image in digital holography, single phase-shifting digital holography is studied. Based on the statistic property of the phase distribution of the diffraction field in recorded plane, an effective method for eliminating the phase-shifting error is presented, in which the unknown initial phase-shifting values can be calculated directly from the holograms, and the summation of the intensity bit errors of the holograms is taken as an evaluation function for an iterative algorithm to find the actual exact phase-shifting value. The feasibility and validity of this method is demonstrated by computer simulation and the reconstructed image.

Based on the practical requirement and feature parameters of high resolution detector, a continuous zoom optical system was designed. The Modulation Transfer Function (MTF) was close to diffraction limit. Some surfaces of front group, variable group and compensating group had very high tolerance sensitivity by analyzing. Therefore, the manufacture and assembly were hard to be achieved. In addition to using defocus to compensate the reduction of the performance of the system, the Z-axis group displacement of the front group was introduced to reduce the sensitivity of the first group, and the feasibility of mechanical structure of the first group was analyzed. An optimization of incident angle was put forward to reduce tolerance sensitivity. The result shows that the performance is enhanced effectively without increasing the costs.

Magnetic fluid is a new type function material which has unique optical properties, including the thermal lens effect, magneto-optical effect, tunable refractive index and magneto-chromatic effect. Magnetic fluid provides new principle and material to the design and research of the novel photonic devices and optical fiber sensors. Wavelength-division multiplexing, optical switch, optical fiber modulator, tunable grating and filter based on the magnetic fluid are presented. Finally, the principle and structure of the optical fiber current sensors based on the magnetic fluid are analyzed.

It is necessary to study and improve the modulation characteristics of Liquid Crystal Modulator (LCM) because of its wide application. A logarithm driving signal model is proposed to linearize the phase response of LCM between 0 and π. Furthermore, it is found out in the experiments that the degree of phase response linearization of liquid crystal modulator decreases with driving frequency increasing. To overcome this problem, this paper presents mapping the nonlinear phase response to linear one when LCM is driven by a periodic signal constructed of the logarithm model. The experimental results indicate that it is feasible to obtain the linear phase response between 0 and π.

In order to restrain the system temperature shift and the strain of optical fiber in the distributed optical fiber Brillouin temperature sensing system which is based on the demodulation method with the intensity of Rayleigh scattering and Brillouin scattering, a novel method, which uses the latest known temperature curve to demodulate the test temperature curve, is proposed and the experimental system is designed. The experimental result shows an excellent agreement with the theory, and it indicates that the Brillouin temperature sensing system based on the new demodulation method can improve the temperature precision, stability, and reduce the cost of the system efficiently. Moreover, the temperature test precision of the system reaches ±0.09 K.

A novel structure of fiber-optic Sagnac sensor, which could eliminate dead zone, was proposed. The optical path circulation was realized by a 2×2 coupler. The optical path was analyzed based on Jones matrix. If the length of delay loop was selected properly, the dead zone of the sensing optic-fiber could be eliminated and the disturbance of false zero frequency could be filtered. After filtering and smoothing, the signal, acquired at eight different points, was fitted by least-square method and the zero frequency was obtained. According to the experimental results, the source of disturbance was located with maximum error 68m and minimum error 3m .The validity of the structure was testified, which indicates that the structure could be used to monitor the leakage at long distance pipeline.

A novel single-ended reflecting Long Period Fiber Grating (LPFG) sensor with thin metal film overlay and the sensing system is described. An all-fiber reflection LPFG sensor with three-layer structure (core, cladding and metal) based on Surface Plasma Resonance (SPR) is established experimentally and fabricated with a pulsed CO2 laser writing system and vacuum evaporation coating system. Different nm-thick thin metal films are deposited on the reflected LPFG sensor for the excitation of Surface Plasma Waves (SPWs) and the characteristics of the reflection resonance spectra of the LPFG sensor for measuring refractive index of fluids are studied. In atmosphere condition, the peak wavelength shifts about 1.14 nm as the sensor with only 80nm Ag film coated is put from water (ns=1.33) into alcohol (ns=1.36). With the accuracy of wavelength measurement (20pm), the SPR-LPFG sensors are expected to have a sensitivity to detect ns (surroundings refractive index) with a change of ~5×10-4. It is found that different thicknesses of different metal films show different measuring sensitivity. Through the comparison of the resonance wavelength obtained in air, water, alcohol, glycerin and their mixture solution, different sensitivities of the reflected SPR-LPFG sensors have been achieved. The proposed SPR-LPFG sensing scheme offers an efficient platform for achieving high performance fiber sensors for measuring ambient refractive index.

In allusion to the principle of Surface Plasmon Resonance (SPR) and the advantage of fiber sensor, fiber SPR sensor system was designed and realized based on the surface plasmon resonance effect, and data processing model was given. Data around the resonance area was processed by averaging method, in which the noise was proved as Gaussian distribution by skewness-peak method. Moreover, mean estimate and linear estimate models were used to estimate the resonant wave length through theoretical analysis. At the same time, the least sampling number was given to denoise on the condition that the precision was ensured. Result indicates that the whole processing workload is decreased by 2%, and the processing efficiency is improved.