Based on geometry optical, the numerical simulation model of segmented telescope system with ray tracing is built, and the relation between curvature radius error of individual mirrors of segmented primary mirror and the image quality of segmented telescope system are analyzed. As monoliphic primary mirror, the error of curvature radius primary can be compensated by adjusting the focus of telescope system, but wavefront aberration induce by the random error of curvature radius of segmented primary mirror cannot be compensated and should zoom the wavefront aberration. When cophased the segmented primary mirror (wipe off the piston and tip-tilt induce by error of curvature radius of individual mirror), the image quality of segmented telescope system will be improved and the effect of error of curvature radius of individual mirror will be reduced.

The stroke of the wave-front corrector is a key factor that determines its correction ability in the adaptive optics systems. The stroke of the current wave-front correctors, i.e. solid deformable mirrors, is small and far from enough to meet the requirement of large aberration corrections. A large stroke Magnetic Fluid Deformable Mirror (MFDM) was presented, where the linearization of the response of the MFDM was obtained by superimposing a large uniform magnetic field generated by Maxwell coil to the small magnetic field generated by the micro-electromagnetic coils. The proposed fluid mirror can easily supply a large stroke of the surface deflection more than 100 μm. The linearized surface analytical model of the MFDM was first built. Then the magnetic fields produced by Maxwell coil and micro-electromagnetic coils and the linearization response characteristic of the MFDM were simulated in COMSOL multiphysics EFA environment combined by MATLAB software. Finally, the simulation results were verified experimentally based on a designed prototype MFDM. The results show that the developed analytical model can correctly represent the characteristics of the MFDM, and the large stroke more than 100 μm can be obtained with small current inputs to the micro-electromagnetic coils under the large uniform magnetic field generated by the Maxwell coil.

A new tracker based on the multiple instance learning framework is proposed in this work, which introduces modified distribution fields features and a temporal fusion framework. The new distribution field features can describe the spatial information of the object more efficiently, and gain robustness to motion blur, minor occlusion and deformation of the object. Our temporal information fusion framework can contain the previous information of the object, and respond to the appearance variation of the object simultaneously, which improves the tracker's ability of recovering from tracking outliers. This new algorithm obtains a better performance on several test sequences compared with other state-of-the-art methods, and can track the object stably under various complicated situations.

In order to solve the problem of complex environmental impact and serious occlusion during the object tracking in the sequence images, a hybrid particle filter tracking method based on the global and local information was proposed. The block texture histogram was imported into the traditional particle filter algorithm and made the algorithm be able to improve the robustness of the tracking algorithm which included the spatial information of target. The adaptability and occlusion resistance of the tracking algorithm was improved since adjusting adaptively the contribution of the global and local information by the degree of the object occlusion. Experimental results show that the proposed tracking algorithm exhibits good result in the presence of partial occlusion and serious occlusion.

High-resolution remote sensors and equipments collected far more images than transmitted through limited-band downlink channels. There is a critical need for special target detection algorithm to select subset of the image collection effectively. An adaptive non-uniform rectangular segmentation and merging algorithm (ANURSM) is proposed to meet the requirement. The algorithm extracts coefficients corresponding to target-area from background in LL3 subband of wavelet-transformed images, and can be integrated with CCSDS-IDC (The Consultative Committee for Space Data Systems-Image Data Compression Recommended Standard) easily and seamlessly. Experiments have demonstrated that the algorithm can effectively reduce data volume and improve the quality of target areas in images.

Referring to the limited field of view of the high resolution camera in the visual precision test of industrial parts, an algorithm using the triangular geometric constraint relation and the partial autocorrelation analysis of the feature points is proposed. According to the correlation between the overlapping images, the improved algorithm realize the mosaic of high resolution images of small gear. The simulation experiment results show that the image similarity between the processed image and the original one exceeds 0.98. Compared with the existing regional matching algorithm and the feature points matching algorithm, the proposed algorithm makes a good tradeoff between performance and complexity. Therefore, it is suitable for applications in fast and precisely visual inspection of industrial parts.

In order to achieve the real-time detection of Coding characters in the process of filling cans, a real-time detection method based on convolutional neural network is proposed. This method initially adopts the histogram equalization and OSTU to deal with the images and then operates the images by the morphological inflation method. Besides, the region of the printed code characters is extracted by the area method of connected domain and then rotates and corrects this region. By using the projection method, the region is divided into single characters which will be trained by the convolutional neural network under the offline state. All above procedures are done in order to recognize the characters while doing the online detection. Experiments show that the average time of every detected image is 46 ms and its accuracy achieves 98.97% which show high instantaneity and accuracy. Thus, it can meet the demand of the real-time detection of industrial cans characters.

In the visual-based aerial refueling assistance system, how to separate the refueling drogue from the complex background and do rapid detection of drogue area is an urgent problem. How to detect and locate refueling drogue over a long distance in real-time, a refueling drogue detection and location over a long distance method based on Hough-LS was presented. First, did pre-detection on refueling drogue based on Fast Hough Gradient Method (FHGM), and then extracted taper sleeve edge based on the method of Search in eight directions from the center. Finally, realized refueling taper sleeve detection over a long distance by the least squares fitting ellipse. Experimental results show that, the method has good detection effect for long distance (7 to 25 meters) taper sleeve target and it can locate the precise area boundary.

The pickup of 3D model is importantly applied in terms of 3D human-computer interaction, and the graphics pickup of interactive graphics system. Currently, the picking algorithm must count on mouse. There are a series of inverse matrix problems in the process of picking algorithm, such as too much simple pickup-surrounded of the structure of 3D model, which influence the effectiveness, convenience and veracity of pickup. A 3D model packing algorithm is presented based on spatial orientation vector and bounding volume strategy. Firstly, spatial attitude information of interactive pen is solved in the slave computer space using entity interactive pen instead of a mouse, then map it into the virtual scene space to make a pick-ray. Secondly, adopt the strategy of shape determination threshold, for 3D model, according to the shape characteristics of different modules, which establish box-shaped bounding and spherical bounding real-time. At last, we judge if it pickup the specific modules or not through the intersection detection between pick-ray and specific bounding. We make specific 3D Model as an example in our experiment, with the using of this algorithm which is pointed out in this paper, the rate of pickup rises 16.98%. What’s more, the pickup misdiagnosis rate of module edges of 3D model is lowed down to 12.80% from 71.26% when it uses the single bounding. The operation is effective and convenience.

When we apply constant values of SOS and DC in reconstruction, the in-focus region will be in a certain depth, causing artifacts and displacement at other depths which deteriorates the image quality seriously. In this paper, the detailed reasons are analyzed for this phenomenon and. Furthermore, a reconstruction approach based on Fitted Delay Compensation (FDC) is proposed. It uses the various fitted delay compensation in image reconstruction, other than apply fixed delay and can significantly reduce the artifact. The experiment is carried out to validate its efficiency, which proves that, compared with the traditional reconstruction algorithm like delay and sum back projection (DNSBP), it shows pretty good effect in improving the image quality.

To make full use of object targeting information in infrared image, the algorithm based on fractal feature and guided filtering is presented to fuse visible and infrared image. The proposed algorithm adopts fractal feature to perform man-made object enhancement in infrared image first, and then gets object distribution picture through threshold segmentation. The approximation image and detail image are obtained by one-level decomposition of images to be fused, and the fusion coefficients are obtained by guided filtering based on object distribution picture. The experimental results demonstrate that, the fused image combines background information from visible image and object information from infrared image fully, and can be beneficial for object recognition tasks further.

An algorithm of image registration based on Hough transform and SURF features is proposed, to solve the problem that the accuracy is low in image registration when large angle rotation and translation occur, and the problem that traditional method is sensitive about local deformation and uneven illumination. Firstly, Hough transform is used to extract straight line features to estimate the rotation and translation parameters to achieve a rough image registration. Then, the image is divided into a few parts, and the SURF is used to do the precisely match sub-regionally. Finally, matching parameters are calculated and the registration is accomplished. Experiments show that the method can be used in the registration of IR and VI images, compared with traditional methods, and this method can achieve higher accuracy in the presence of a large angle rotation and translation and get a better result when localized deformation or uneven illumination is occurring.

To solve the problem that fusion algorithm is not optimal selection along with the change of difference characteristics when dealing with infrared polarization and infrared intensity of dual-mode image, a method of set-valued mapping fusion is presented. Multi-group of polarization and intensity of infrared images as training samples, different classification of low-level features and fusion algorithms are used to construct difference characteristics class set and fusion algorithm class set. Fusion effective measure distribution of each characteristic corresponding to fusion algorithms is constructed. Mean aggregating for multi-group images fusion effective measure distributions in order to establish the set-valued mapping relationship between difference characteristics and fusion algorithms of the global image. Difference characteristics of test images are extracted. Pick out corresponding algorithms according to the set-valued mapping. Then, get the fusion result by the combination of the algorithms. Experimental results show that the fusion method can select the optimal fusion algorithm, and have the highly complementary characteristics fused effectively.

To solve the shortcomings of low energy efficiency and contrast of traditional LCDs, and considering the advantages of mainstream backlight construct and local dimming technology, a local dimming algorithm for edge-lit LED backlight is proposed. The method consists of three parts: backlight level dimming, LED allocation and pixel compensation algorithm. In backlight level dimming algorithm, the input image is divided into several blocks and each block image luminance is calculated. Then a resource allocation and transport model is introduced to establish the mapping relationship between block luminance and LED strings in allocation algorithm. In pixel compensation, first, the backlight diffusion effect of the light guide plate (LGP) and diffusion film is acquired from simulated with low-pass filters. Then, the LC pixel is compensated based on original image pixel and backlight effect. The algorithm is simulated, and a TV prototype is developed to test actual display effect. The actual test results show that the average power reduction reaches 20.57% and the Structural Similarity (SSIM) of original and dimmed image is 0.985. The proposed algorithm presents a good power saving and display effect as well as light and thin backlight structure.

For the heavy pollution and high content of trace soluble heavy metals Cd2+、Cu2+、Zn2+、Ni2+ in waters, a determination method of trace soluble heavy metals in seawater is proposed based on flow injection analysis and spectrophotometry. Firstly, the absorption spectrums of four heavy metal ions are analyzed at different wavelengths. Then, an orthogonal correction set with 4 levels and 4 factors is established as the multiple linear regression analysis model. Finally, the measuring equipment of four heavy metal ions is set up, the concentration of soluble metals Cd2+、Cu2+、Zn2+、 Ni2+ in waters is measured, and the detection limit at 10-9 (ppb) can be achieved.