A new approach, which combines the radar’s dynamic information and several scans’ measurements to deal with the data initial association problem in high clutter environment of tracking-while-scanning systems, is proposed. First, the dynamic initial track numbers are estimated by using the measurements’ number in the sliding window, the target detection probability and the receiver’s noise sampling information. Then the estimated initial track numbers are adopted to adjust the initial correlation region and the normalized residual square threshold adaptively. Therefore, lots of false initial trajectories are suppressed. Simulation results demonstrate that our approach can solve the real time data initial association problem better than logic-based method.

The Forward-looking Infrared (FLIR) camera is installed in the front of ship. In the FLIR surveillance video sequence used for moving ship detection, it can be achieved to avoid ship-ship collision so that the safety sailing is ensured. Ship detection and tracking algorithm in FLIR images from complicated background along the inland waterway is proposed. Firstly, a method is proposed to extract the sky-water line, and the Region of Interest (ROI) is extracted based on the sky-water line. Then, multi-scale fractal feature related with K (MFFK) is presented to detect ship. Finally, an improved mean shift algorithm for moving ship is proposed to track ship in inland waterway. Experimental results have shown that the approach is robust and achieves real-time and reliable ship detection and tracking.

In this paper, three main indices of multitarget tracking in random set framework were compared，including Circular Position Error Probability (CPEP) distance, Wasserstein distance and Optimal Subpattern Assignment (OSPA) distance. Their characters, conditions, pros and cons were first investigated. Besides, the existing formula for the CPEP distance, which may cause confusion, was modified. The result shows that the CPEP distance can be used to evaluate target lost. Compared with the OSPA distance, the Wasserstein distance is easy but may be intuitively inconsistent in some cases, while the OSPA distance is more reasonable. Nevertheless, it needs to select the level parameter and this is subjective. Finally, two experiments are proposed to validate these indices.

Fusion method of subdivision band images of mid-wave infrared is presented based on both morphology processing and wavelet packet transform. Local fusion of two subdivision band MWIR images is implemented with morphologic erosion and dilation processing and weighted stacking method in order to eliminate saturating regions in the first subdivision band image that are done by sun irradiation. And global fusion between local fused image and the second subdivision band image is achieved by two dimension discrete wavelet packet transform in order to combine detail information of two images and to get legible results. The experimental results show that local standard deviation and local entropy of image obtained by twice fusion, compared with two original images, increase by 21.06% and 9.20% respectively, which proves the validity of fusion method above.

System error is change in some fusion system. In order to resolve the problem of system error registration, this paper combine improved Particle Swarm Optimization (PSO) with Simulated Annealing (SA).This method integrates the capacity of fast optimization seeking in improved PSO with the characteristic of probabilistic leap in SA, which not only solves the defect that PSO easily plunges into local optimum, but also ensures population variety and avoids population degradation. The simulations show that compared with PSO and GA, improved SA-PSO has better precision in system error registration.

To study a backup navigation scheme for allowing temporally GPS faults or degradations, a natural-landmark scene matching visual navigation method based on dynamic keyframe is proposed for Unmanned Aerial Vehicle(UAV).Firstly, this method could autonomously describe and check featured natural landmarks by analyzing image sequence from on-board camera. After abstraction of keyframes including featured natural landmark, UAV will be located by the means of NLSM (Natural-Landmark Scene Matching) based on dynamic keyframe. Additionally, this navigation scheme adopts interframe scene matching algorithm in order to improve the navigation performance of accuracy, reliability and runtime. Experiments show that the backup navigation scheme proposed meets the requirements of navigation in complex and unknown environment for UAV.

A multi-target tracking algorithm based on multi-model is proposed to solve uncertainties in real environment.The algorithm uses dual threshold method to ensure real-time performance and tracking accuracy. The method firstly preliminary determines the targets associated with the measurements, and then tracks these targets. The algorithm also determines the flight conditions according to the target state estimation. Different method are used in accordance with different flight conditions to improve its robust and real time performance. The measured data verifies the efficiency of the algorithm when tracking targets.

A model of parallel pair is proposed to overcome the limitation of the traditional one and it is the kernel of the designed algorithm for parallel feature extraction in Synthetic Aperture Radar (SAR) images. After the original SAR image is filtered to suppress the speckle noise, Ratio of Exponentially Weighted Averages (ROEWA) operator which has constant false alarm rate is used to detect edge. Then the approach of primitive parallel line segment extraction is carried out. Finally, the breakpoints are connected by means of the heuristic link idea. The experiment results show that the proposed algorithm can extract the parallel characteristic of SAR images efficiently, and it can be used for auto target recognition, such as roads, airports and rivers.

A method of small target detection in infrared image sequences is proposed based on the Fuzzy Support Vector Machine (FSVM) temporal background predication. Firstly, the pixel gray value sequences at the same position of the previous frames are fitted with function by using the FSVM optimized. The pixel gray value at the same position of the next frame is predicted by the fitted function. Then, the estimated image subtracted from the source image gives the residual image. The residual image is segmented using two-dimensional Tsallis-Havrda-Charvat’s entropy thresholding method. The real small target is confirmed by the continuity and consistency of its movement. The experimental results were given and analyzed. They were compared with those of the existing method of spatial or temporal background predication .The results show that the proposed method can precisely detect the small infrared target and it is superior to the existing method.

The light could be split into lots of beams with Digital Micromirror (DMD), and focused by convex lens. Then,a point light source array will be formed, in which the size of point and the space between points can be controlled conveniently. The DMD is more flexible and cheaper than the traditional micro optic elements. With a digital parallel light source constructed with DMD used in parallel confocal microscopy detecting, an optical path is designed for current measurement environment. The experiment result indicates that the axial resolution could be improved with smaller period of point light source array, and the conclusion is based on the pixels of array meeting the needs of DMD and CCD.

Coherent light sources in the Deep Ultraviolet (DUV) spectral regions become more important with the development of photolithography, laser micro-machining as well as laser spectroscopy. KBe2BO3F2 (KBBF) is now the only crystal to produce deep-UV coherent light just by using Second Harmonic Generation (SHG). The technical parameter of the SHG, Fourth Harmonic Generation (FHG) and sixth harmonic generation using the Nd:YVO4 laser and the FHG using the Ti:Sapphire laser are obtained according to the frequency doubling in KBBF. The theoretical conversion efficiency of the Gaussian beams was obtained by using the numerical simulation on the crystal KBBF in the matching type I and type II for the first time. These academic results are very useful in DUV-DPL experimental studies.

To ensure turning-off the laser in the system of continuous wave cavity ring-down technology, a new method for modulating laser current, named as laser detuning is proposed. With this method, the resonant state between the laser and ring-down cavity is ended. Hence, the incident laser to the ring-down cavity can be turned off rapidly. This detuning method does not require external optical switches (such as the acousto-optic modulator) to turn off the laser source so the system is simpler than previous ones and the cost is lower than traditional systems. The continuous wave cavity ring-down system based on the laser detuning is designed. The operating principle and performance of various devices are analyzed. The characteristics of the detuning process are analyzed and the difference between the laser detuning and laser turn-off are compared. As a result, the turning-off time in the technology of modulating laser current is shorter than the traditional optical switches, and the exponential curve is more distinct.

At the beginning of 21st century, a new method of driving for large telescope was developed, which was based on Arc PMSM (Arc Permanent Magnet Synchronous Motor). This method has many advantages over other traditional drive methods, on the other hand there is a problem of thrust ripple, which will affect the tracking precision of the telescope. In order to solve the problem of thrust ripple, which will affect the tracking precision of the telescope, a control method based on the principle of Space Vector PWM (SVPWM) was introduced. A model of Arc Permanent Magnet Synchronous Motor (PMSM) was built up under the condition of Matlab/Simulink. And SVPWM method was adopted to construct Arc PMSM control model which contained velocity and current control loops. Simulation results show that the motor with the load can get a fast dynamic response and ideal three-phase sine wave current. The speed andthrust are smooth, and the thrust ripple is smaller than 5%, so this control method can meet the needs of the trackingrequirement of the large telescope.

High-orbit optical surveillance platform has advantages accepting state information of space-objects. The orbit determination theory for space objects and its precision assessment methods by high-orbit optics surveillance platform were analyzed. The accuracy of three kinds of space objects based on high-orbit optics surveillance platform was studied.The orbit determination accuracy was assessed by two methods of observation residuals and results’ covariance and the reliability of results was demonstrated and the effects of orbit determination caused by the error in the measurement were studied. Simulation experiment shows that the orbit determination precision for those space objects reaches kilometer level. System error is the main source of orbit determination errors.

To solve the problems of dynamic focusing, correction of magnification, and vibration with noise in volumetric 3D display technique by reciprocating scanning, a method of volumetric 3D display by scanning with multi-screen in parallel motion is proposed. In this method, multi-screen and multi-mirror in circle-curvilinear motion are adopted to maintain the optical path constant between fixed projector and moving screen, so as to form a size-invariable and sharp focused 2D image sequence, avoiding the problem of dynamic focusing and correction of magnification. With the persistence of vision, the 2D image sequence is perceived as a 3D image with physical depth in which uniform voxels are allowed to be arranged uniformly without physical dead zone. By arranging all the moving components symmetrically in corresponding circles and driving them in even velocity, the inertia forces are balanced automatically so as to solve the problem of vibration with noise. A prototype is manufactured and the method is verified by experiments.

To meet the requirement of online measurement of large forgings in thermal state, a method based on structured light stereo vision technique was presented. To solve the problem of the acquirement of structured light strip on the condition of thermal state, the method combined physical filtering and digital filtering to overcome the influence of red thermal radiation，and temporal phase unwrapping structured light technology was also used to improve the optical noise immunity significantly. Multi-view auto registration techniques based on mark balls were adopted to measure large objects. An experimental system has achieved non-contacting 3D measurement of large forging in 15 m distance. The experiment proves that the method has advantage of high-speed and high- precision, and it is also simple and convenient,which can meet the requirements of online measurement of large forging in thermal state.

For big noises tire image, a method of character truncation was presented to get distinguishing outer contourfeatures. Artificial neural network and features of relations was used to recognize tire specifications. First, the samplefeatures of distinguishing specifications were gotten: two-thirds characters were intercepted to get run-lengthcharacteristics of its external contour. Then, one recognition result was gotten by BP neural network training andrecognition. Finally, character features of relations processed fore results were applied to do the second recognition.Experimental results show that the characteristics of acquisition and the second recognition algorithms can effectivelyidentify the tire specifications. Moreover, BP neural network has better fault-tolerance and recognition rate fordeformation characters caused by noises, and feature gotten and secondary identification algorithm can effectivelyidentify tire specifications.

In optical test for manufacturing large aspherical mirrors, horizontal test arrangements are used traditionally.Because of the environmental factors such as air turbulence, temperature variation and vibration ect., it is difficult to obtain the test results reliably and repeatedly. Online or offline vertical test towers are commonly applied in many countries. A vertical test tower used online was designed in our workshop for testing an aspherical mirror Φ1.07 m, F/1.5.In order to fit for adjusting all freedom, a four-dimensional and another two five-dimensional adjustable tables are designed to adjust a mirror, a null lens and an interferometer. In our jobs, the stress, the deformation and the vibration modules of the tower are formatted by finite element methods to avoid resonance. After our works, the stable interferograms and testing repeatability are got by the test tower in the process of manufacturing aspherical mirror Φ1.07 m, F/1.5. The test result for nearly finishing is PV=0.313λ and RMS=0.024λ.

As to detection, recognition and simulation for infrared characteristic of target, a method based on the physical model of the targets which are irradiated through infrared illuminator is proposed. The normal circumstance and artificial infrared illuminator were combined to affect targets. The calculation model of surface temperature and infrared radiation character was built. Analysis of spectra characteristic based on carbon-silicon rod infrared illuminator, taking the Hefei’s building for example, 3~5 μm and 8~14 μm wave-band’s infrared radiation was calculated and analyzed. The results show that target’s infrared radiation character is changed when irradiated by using differential power. In the same condition, the power usage of 8~14 μm wave-band is greater than that of 3~5 μm wave-band.

In order to address small sample size (3S) problem, a new algorithm for implementing linear extension of graph embedding (LGE) based on QR decomposition is proposed, which could be used in face recognition. Different from the traditional approach of dimension reduction by Principle Component Analysis (PCA), the new algorithm applies QR decomposition to implement dimension reduction. Then the LGE is followed and employed for the second feature extraction in the transformed space. Finally, the nearest neighbor classifier is used for classification and recognition. The new algorithm not only can effectively solve 3S problem, but also hold the discriminant information. Experimental results on YALE and PIE face databases show that the algorithm outperforms the traditional method in recognition rates.decomposition

A new band selection method for hyperspectral images, namely Maximum Band Screening (MBS), is proposed. Considering a band image as a band vector, a spectral similarity measure is used by MBS to measure the dissimilarity between a band and the bands selected, and the most distinctive and informative bands are selected form the original bands. In order to demonstrate its utility, an anomaly detection algorithm is developed, which first extracts the anomalous target spectrum from the image, followed by noise whitening, then implements MBS to select bands that participate in the subsequent Adaptive Cosine Estimator (ACE) target detection. As shown in the experimental result on the AVIRIS dataset,only 8% of the full bands are selected by MBS, with which the detection performance of ACE has been improved greatly over that by other two band selection methods.

Aiming at the problem that the matching between phosphors and chips has significantly influence on white LEDs, characterization and measurement of LED phosphors were presented. According to the measurement principle of phosphors, a double splitting system was presented. Phosphor’s spectral characteristics were analyzed, and three-dimensional spectral distribution was achieved by combination. Luminous efficacy, quantum efficiency and energy efficiency were brought forward to evaluate phosphor’s performance. By evaluating the matching between kinds of phosphors and LED chips with different peak wavelengths but the same power, the optimal matching was acquired. Results show that the system and evaluation method can not only determine LED phosphors’ characteristics, but also has significant meaning in the research of achieving high-efficiency white LEDs.

The solar blind region provides a good background for the detection systems that work in the wavelength from 240 to 280 nanometers on the near earth surface. A set of solar blind ultraviolet imaging detection experiment system is proposed. From the point of view of radiation energy transfer, referring to the analysis methods of staring infrared search and track system, the detection range model of ultraviolet detection system was derived. With a deuterium lamp as the detection target, the experiment system and model were analyzed and validated in experiment under conditions of three kinds of typical weather. The result suggests that the calculated values are consistent with the practically measured values.It provides an important theoretical and experimental basis for the design and performance evaluation of solar blind ultraviolet detection system.

A new concept of Approximate Non-diffracting Grating Structured Light (ANGSL) and its implementation method are proposed. ANGSL has characteristics of long focal-depth, narrow strip-width, steady light field, and constant light intensity in certain rang etc. An optics system with a precise triangular-section prism is set up to produce ANGSL.Applying Huygens-Fresnel principle and Kirchhoff theory, theory analysis and calculation are completed. The distribution of the light field is simulated in the non-diffracting range. Experimental results show that experimental data is in good agreement with the theory calculation. According to system parameters, ANGSL is produced with 18 μm strip-width and 570 cm non-diffracting range, whose strip-width and intensity distribution do not change with transmitting distance. Compared with traditional light source, ANGSL has advantages of high resolution and precision in the application of 3D profile measurement.

In contourlet transform, a drawback of Laplacian Pyramid (LP) is oversampling and non-orthogonal decomposition. Wavelet transform is a critically sampled scheme but it only generated three directions. Taking advantage of wavelet and contourlet, the proposed method was completed by wavelet transform in conjunction with directional filter banks. First，the proposed method derived wavelet coefficients of entry image through wavelet transform, then reconstructed the image only by lowpass coefficients. Finally, after processing, a highpass subband was derived. The proposed method kept critical sampling and it was convenient for direction filter banks to partition the highpass subbands.The proposed method can eliminate the noise effectively and provide the improved Peak Signal Noise Ratio (PSNR) over wavelet and contourlet transform.