To process uncertain data obtained in sensors,a model of Situation and Threat Assessment (STA) based on fuzzy Bayesian network was proposed. The uncertain sensor data were divided into vagueness domain and probability domain. The vague data type of the threat target was handled by fuzzy comprehensive evaluation,and the dynamic threat degree was obtained in fuzzy domain. And then the fuzzy dynamic threat degree was translated into the probabilistic type by the possibility/probability theory. All the uncertain data were figured in probabilistic type and processed by Bayesian network to produce the threat level of the target. An example indicates that the fuzzy Bayesian network can obtain the real threat levels and is feasible for weapon system to automate decision-making on target-selecting and target-striking.

A new multi-target detection and tracking technology based on circular opto-electronic detector is proposed. The detector with circular structure based on 6 linear CCDs,is firstly developed to achieve 360°field of view. In order to achieve the detection and tracking of multi-target at high speed,the target recognition criterions of Object Signal Width Criterion (OSWC) and Horizontal Scale Ratio Criterion (HSRC) are proposed. As a result,the algorithm based on the two criterions,which is suitable for the proposed circular detector,is developed. Finally,an experimental system,which has the functions of signal sampling,process,and results display,is developed. The detection function of the proposed methods is also demonstrated by using the developed experimental system. The experimental results indicate that the technology offers significant advantages in Detection and Tracking,Computer Vision,etc.

This paper addresses the problem of detecting small targets with very small spatial extent that are masked by spatially strong background clutter in image data. A novel adaptive prediction algorithm of background clutter is proposed based on decreasing the Cumulation-squared Error (CSE) of gradient in neighborhood. The algorithm can improve the Signal Clutter Noise Ratio (SCNR) of small targets evidently. Many experiments prove that the algorithm has better performance than traditional methods. And a fast detecting algorithm based on statistical analysis is introduced. It can get higher probability of detection based on lower probability of false alarm than other methods. The theoretic analysis and many simulations verify the validity of the method.

Two-dimensional pointing mirror as well as detector array is often used in the searching and tracking system. Based on the vector theory of optical reflection,the imaging characteristics and scanning traces of the two-axis pointing mirror are studied. The relationship between image rotating angle and the pointer’s elevation and azimuth is analyzed in detail. When the pointer rotates around the axes which aren’t parallel to the pointer’s surface,the images on the detector rotate as well and the trace of the pointer is non-linear. Furthermore,the calculation methods for image rotation of detector array and the influence of rotation on the tracing system are also given.

In allusion to the flaws of general dynamic programming algorithm with maximizing energy,a new improved method with the most steady states is proposed based on the previous methods. Considering the stabilization of the gray of the targets and the continuity of the movement trajectories,probabilities of the energy stabilization and direction stabilization are computed respectively. Probability of state stabilization is computed with information fusion and energy of candidate point whose stabilization probability is most is accumulated. Simulation results show that the proposed algorithm can well solve the energy scattering problem. The area of energy scattering is decreased by 10 times when targets’ SNR is 2.5. Targets’ SNR can be increased stably to 1.5 times when the number of accumulation frames is 5 and the Targets’ SNR is 2.5. The number of false alarm points is decreased by more than 10 times when the targets are separated by threshold.

Based on the situation of the space-based small moving target detection on space surveillance system,we introduce the way working on staring model,and then discuss the way of small moving target detection based on image registration technology. Through finding out the local min distance between stars as star map characteristic quantity,a novel star map characteristic quantity is proposed. This characteristic quantity is used to align the neighbor star maps in the sequential images. Through subtraction,the background consisting of evolving clutter can be removed and the small moving target in space is detected. The result of the test shows that the local min distances in star maps distribute well proportionally and the star maps match well.

Based on multirate output feedback control law and region pole assignment technology,a discrete sliding mode control algorithm is presented. The control law is adapted to seeker servo system,which is disturbed by friction,coupling and restriction moment. The method to determine the parameters of the controller is also given. The disturbance doesn’t need to satisfy the match condition and the boundary of the disturbance is known. Theoretical analyses show that the control strategy can guarantee the stability of the closed loop system. This methodology doesn’t need the system states for feedback as it makes use of only the output samples to design the controller. Thus,it is more practical and easier to be implemented. Simulation results show that by adapting the presented control law,the servo system of seeker can reach good performance in the presence of disturbance. The advantages,such as easy implementation and good dynamic performance,make the control law more promising on control system design of seeker.

To satisfy the high performance of image-stabilization platform,a new compound adaptive inverse control method based on Elman neural network and PD was proposed for speed loop design. Through analysis of the Elman neural network model and the control object,index-trace loop and disturb-attenuation loop were designed independently. Training of the Elman neural network and identification of the object were implemented on line. Simulation results show that the method is feasible to effectively overcome the influences of the nonlinear factors,such as disturbances and the slow time-variation of parameters.

Adaptive optics system requires the high calculation capability and high real-time capability of wavefront processor. In view of the requirements,a new method for real-time adaptive optical wavefront processor based on systolic array is proposed. The method maps wavrfront gradient calculation,wavrfront reconstruction and wavefront control calculation to systolic array respectively in order to build the deep pipelines of data reasonably. Meanwhile,the latency of the wavefront processor based on Field Programmable Gate Arrays (FPGA) technology is analyzed. A 61-element 48 sub-aperture adaptive optical system has been implemented using a Xilinx Virtex-Ⅱ XC2V3000 FPGA. Experimental results indicate that the latency is only 8.6μs. As a result,the method can greatly increase the wavefront processor real-time performance,integration level,expansibility and general-purpose character.

A novel extraction algorithm for infrared moving target is proposed based on spatio-temporal information. The proposed algorithm efficiently utilizes the target intensity feature,surrounding background and the moving information. In time domain segmentation,Gauss distribution model of frame difference background is established to determine the infrared target region via change detection mask. The spatial relation matrix between pixel and region is constructed to constrain the classical fuzzy C-Means clustering (FCM). And then,the region which contains the entire target is segmented efficiently based on the improved fuzzy clustering algorithm. At last,infrared target extraction is achieved by the fusion of spatio-temporal results. Experimental results verify the effectiveness and robustness of this extraction algorithm which can extract the infrared target correctly.

A multiple hypothesis method for tracking multiple motion objects is proposed based on the object detection,which is robust to changes illumination and shadow conditions in image sequence. After video preprocessing including background suppression,temporal filter and clutter removal,the initial results of object detection is obtained in each frame image. Based on the detection results which may have missing and/or false detection,the multiple object tracking algorithm accepts the probabilities of initial object detection and keeps multiple hypotheses of object in a graph structure. Through extending and pruning the graph based on the previous image information,the multiple object tracking method gives feedbacks which are predictions of object locations to the object detection module. Therefore,the method integrates object detection and tracking tightly. The most possible hypothesis provides the multiple motion object tracking results. The experiment and evaluation results illustrate the validity and advantage of the proposed algorithm.

The dead-zone attached with all-digital close-loop Fiber Optic Gyro (FOG) which employs square wave bias modulation and step wave feedback modulation was researched. The dependence between the height of step wave and open-loop output of the FOG was obtained with a correlation coefficient equaling 0.97,and an open-loop detection model taking this correlation into account was established. Based on the equivalence of digital step wave height set up by fixed step or feedback step,the above open-loop model was modified to close-loop model. The close-loop model was analyzed by theory and simulation. The simulation dead-zone range was compatible with the experimental one with an error of 6.7%. The results show that the linear correlation between the height of step wave and the output is one of the factors of dead zone.

To compensate the temperature drift in a Fiber Optic Gyro (FOG),a linear multi-variable model was proposed. The model was composed of the autoregression part and the polynomial distributed lag (PDL) part. The former represented the effect on FOG’s current output by its own lags,while the latter described FOG’s drift caused by temperature variation. Due to the linear character of the model,its coefficients were obtained by using least-square estimation method. In the testing,two FOGs' field temperature drift data sets were collected to validate the effectiveness of the proposed model. The results show that the linear multi-variable model is effective,and the FOGs’ accuracy may be improved by more than 50% after compensation.

Aiming to the errors which were induced when wavefront was simulated with Zernike polynomials on the discrete points,a set of orthogonal polynomials expressed on a discrete point to simulate turbulent wavefront was proposed. According to the turbulence statistics theory,the turbulent wavefront expressed by Malacara polynomials was constructed with Gram-Schmidt orthogonalization. Compared with the method directly using Zernike polynomials,the results show that the method introduced in the paper is more accurate under the same conditions.

Based on the principle of Moiré Deflection Technology (MDT) and the definition of lenses’ power,relationship between moiré fringes’ tilting angle and the tested lens’s power is deduced respectively according to the combination of Talbot effect and principle of grating shadow and Fresnel-Kirchhoff theory. The calculated results are uniform that is lenses’ power is in the direct ratio to the tangent of moiré fringes’ tilting angle. Some factors that will affect the measuring error of lenses’ power are discussed. Analysis of the results shows that measurement of lenses’ power using MDT has high precision and it is feasible for lenses with low or middle-high power if proper period of gratings and reasonable angle between the two gratings are adopted. In the end,images of moiré fringes gotten by a single lens,an aspheric lens and a progressive addition lens are provided and analyzed.

The principles of profilometry with color face-structured projecting grating was introduced in this paper. The sine coded projecting grating was adopted,so that the rapid direct demodulation of grating fringe method could be used to carry out the measurement of single view angle of the object,and a standard step was measured by the method. The reason that caused errors in the measuring process was analyzed,and the corresponding improving method was put forward. Finally,the foot of complex free-form surface was measured by the method,and the results show that the method can obtain the 3D data of the object quickly and accurately.

A customized optimization model of phase plate for Extended Depth of Field (EDOF) is introduced based on the various requirements of different EDOF optical imaging system. The requirements contain customized DOF,resolution and minimum Modulation Transfer Function (MTF) value. In the optimization model,insure that the related MTF values is bigger than the minimum MTF value,and the integration of the Fisher information of the Optical Transfer Function (OTF) in the designed range is employed as the optimization target function. Then we optimize the cubic phase plate of the wavefront coding system and the exponential phase plate for several practical requirements. The subsequent image restoration filter is also introduced in the paper. The results show that the combination of the customized phase plate from the optimization model and the restoration filter can meet the requirements of system. And the whole MTF of the system is near the diffraction-limited system’s MTF.

The proposed algorithm for digital imaging system ensures correct exposure to the main object of scenes with large dynamic range. The image window was divided into M×N tiles. Using the contrast between the main object and the background of large dynamic scenes,the algorithm identified the main object through a certain search process,detected the lighting condition based on the search result,and then assigned different weights to the main object and the background so that the overall brightness level was made to carry more information of the main object. The experiment results show that the lighting conditions of all the scenes for evaluation are correctly detected. For the high dynamic range lighting conditions,the number of preview frames and the exposure error are less than 8.8 frames and 6.56% respectively.

Polarized light could be described completely by Stokes Parameters. A Division-of-Amplitude Polarimeter (DOAP) system was designed,in which one quarter-wave plate was set between the lenses of Division-of-Amplitude,and then the light path was in same plane. Intensity of four light beams related to Stokes Parameters was tested by four detectors with the same performance. The system was calibrated by known polarization that combined polaroid with one quarter-wave plate. Through matrix calculation,the Stokes parameters were tested quickly and the polarization of light was work out. Experiment shows that the system is set up easily and adjusted conveniently. The deviations of four dimensions of Stokes parameters between testing value and theoretical value are less than 6%.

A pushbroom microscopic hyperspectral imaging device was developed to study the normal and leukemic blood cells of human. The microscopic hyperspectral images of blood from normal and leukemic human were collected and processed. Some typical transmitted spectrum curves of blood cells were extracted and analyzed. The results show that the transmittances of leukaemia blood cells are 50 percent higher than those of the normal at 541.3 nm. From the images and spectrums of the blood,it can be seen that the microscopic hyperspectral imaging device can be used to study the changes of spectrum characters and physical chemistry composition of the human blood.

Aiming to the defects that some traditional contour descriptors can’t classify similar contours and that others o not have a good internal rotation-invariant property,the sector-projection-wavelet-descriptor was proposed. Contour to be recognized was first orientation normalized by its principal axis. And then,the normalized contour was projected on the prepared N-sector-areas. Finally,the curve gotten from previous steps was analyzed by wavelet theory. This descriptor can not only decrease the error caused by orientation normalization,but also can classify some contours which can not be classified by ring-projection or Fourier descriptor. The experiments compared with other contour descriptors show: sector-projection-wavelet-descriptor has a good recognition result whatever the contours to be classified are similar or largely different.

A new sub-pixel algorithm based on the principle of the parabola fitting is proposed. According to the condition of the correlation peak,the outline of the longitudinal section along the climax of the correlation peak is fitted by parabola. The longitudinal section is executed along different direction to acquire different parabola,and then the climax of parabola can be gained from every parabola. The optimum position is the point,where the sum of distance is the least from it to all other points. Thus,it is realized that the line of parabola is described by point and the surface of correlation peak is described by line,which can show more truly the condition of the correlation peak. The experimental results show that the algorithm maintains a high precision of the subpixel extraction.

First DPCM forecast transformation was implemented to a medical image,and then integer wavelet transform was done to the image. The low frequency coefficients and high frequency coefficients were processed by Huffman code and vector quantification respectively. According to the properties of the wavelet coefficients,most of the energy was focused on the low frequency part. Lossless entropy code was implemented to the low frequency coefficients,and vector quantification was taken to the high frequency ones in order to reduce the redundant information,which was insensitive to person eyes. Finally,the processed low frequency and high frequency coefficients were reconstructed to obtain the compressed image. Compared with the compressions based on Discrete Wavelet Transformation (DWT),JPEG and JPEG2000,experimental results indicate that a high compression ratio and the good compression effect can be obtained by the proposed method.

In order to restore the blurred image to obtain the cognizable target,we analyzed two kinds of blur characters,which produced by the high speed relative motion between camera and target. For the question that the degraded function of whole degradation process was difficult to ascertain,we proposed a method of mutual layered restoration based on image interpolation. First,the method improves the analysis ability of low resolution images through image interpolation. And then we chose the optimal parameter combination of the two degradation functions mutually under the optimization rule of mean squared error. The layered restoration could be implemented with the two degradation functions of blurred image. We restored the image and obtained the main characters of the target image in simulation experiment. The result shows that the method is effective.

Corners in an image were detected by multi-scale Harris operator,and they were taken as initial interest points. Since adaptive non-maximal suppression eliminated lots of potential matching points,condition theory was applied to control the number of initial interest points. The bad conditioned points were eliminated,so the computational complexity of the following process was decreased and the efficiency of the algorithm was improved. Meanwhile,matching points were mostly kept. Features under the control of condition theory were called well conditioned features. Since the location of initial Harris corner had offset and false corners were produced,sub-pixel localization technique was used to determine the location of corners,and the false and unstable corners were eliminated in this process. PCA-SIFT was applied to describe the feature points and their neighbors to get vectors. Finally,Euclidean distance between vectors was used to determine whether two points match or not. Experimental results show that the efficiency is improved significantly,the proposed algorithm is good at feature matching,and has good robustness to geometrical transformations,image noise and illumination change.

Aimed at features of infrared and visible images and combined with Mitianoudis’s fusion method,an improved Independent Component Analysis (ICA) domain multimodal image fusion algorithm was presented. According to Mitianoudis,linear transformation was performed to the source images using ICA bases obtained from offline training. Then the image in ICA domain was segmented into different regions: active regions and non-active regions. For the active region,the "max-abs" fusion rule was used,while for the non-active region,different fusion rules were used according to the region segmentation result of the target sensor image. Finally,the fused image was reconstructed by the inverse transform. Experiment results demonstrate the effectiveness of the method.

Based on the spectrum and image analysis,a method of predicting the converter end-point was proposed. A flame light and video capture system was designed to analyze the data with the edited codes. The result shows that the light intensity and the image characteristic information have the remarkable change law in the steelmaking blowing process,which coincides essentially with the experience steelmaking. Through the statistics reverse analysis,some characteristic parameters were selected,and then the end-point time mathematical model was established by the use of the multivariate regression analysis method. The experimental results indicate that the model passes the significance test,and the predication accuracy is 90.4% when the measurement error is less than 4 second. The model is stable and provides a convenient method to predict the converter steel-making end-point on line.

In iris preprocessing,the located iris was divided into 8 layers at first,and each layer was normalized by different standards to unwrap the iris image into stepped shape,which avoided the data repeat or loss caused by traditional normalizing method. By using the frequency-selective character of 2D Gabor wavelet,the iris features were extracted and recognized respectively in different spatial frequency bands. In the end,by comparing the recognition results in different frequency bands,the spatial frequency characteristics of iris texture were obtained. The experiment shows that when the spatial frequency changes from 0.005 to 0.04,the recognizable iris features can be effectively extracted,and the recognition can be successfully realized. However,the extracted features will be full of noises and the iris will not be effectively recognized when the spatial frequency is lower than 0.005 or higher than 0.04.