Aiming at the problems of near-eye display system such as small field of view, bulky, poor flexibility, and poor mass production, the design method of adding free-form surfaces and polydimethylsiloxane (PDMS) groups to the system is adopted. First, the design requirements and working principles of the flexible near-eye display optical system are discussed, and the initial structure is established. Then analyze the reasons for the stray light of the system. Then use XY polynomial free-form surface to replace the traditional aspheric surface to correct system aberrations. Finally, a cylindrical waveguide was used to simulate the bending of the waveguide, and the optimization analysis of the bending of the flexible waveguide was carried out, and a near-eye display optical system with flexible bending ability, large field of view and high imaging quality was designed. The designed working wavelength of the near-eye display system is 400~700nm and a pupil diameter of 8 mm, field of view is 30°, and a thickness of 1.5mm. At a cut-off frequency of 71lp/mm, the field of view Modulation Transfer Function (MTF) value is greater than 0.1 and system distortion is less than 5%, in cases where the waveguide deformation bending radius is greater than 500mm, the modulation transfer function value of the full field of view is greater than 0.1 at the cut-off frequency of 15lp/mm, and the system weight is about 22.8g.The design results show that the design achieves the miniaturization, light weight and flexibility of the near-eye display optical system, and can be applied to a new generation of helmet display systems.

Photoelastic modulator uses the birefringence effect produced by the forced vibration of isotropic crystal to modulate the phase of incident polarized light, which has important applications in polarization modulation and measurement. The phase delay of photoelastic modulator is its key index parameter, and the driving voltage and phase delay should be calibrated in practical application. However, the components in the traditional calibration system of photoelastic modulator are bulky, and the calibration is completed before use, resulting in large errors. In order to make real-time calibration of the photoelastic modulation system available, achieve the purpose of precise calibration, and improve the convenience of use, a micro-system for precise calibration of photoelastic modulator is proposed. By reducing the size of the calibration system components, and making them integrated at the edge of the transparent aperture of the photoelastic modulator to realize real-time calibration of phase delay. It is verified by experiments that the calibration microsystem of photoelastic modulator can accurately calibrate the phase delay of photoelastic modulator in real time without affecting the central light path. The calibration results show that the phase delay deviation is less than 1% after one hour's testing.

In order to reduce the influence of temperature on the measurement accuracy in photogrammetry, based on the physical experiment, the influence mechanism of camera's autothermal image drift was studied, and a compensation method for camera's autothermal image drift was proposed. Through the long-term continuous monitoring of the stationary mark points by the camera measurement technology, it is found that the temperature change of the measurement system will indeed cause the drift measurement error of the image points, of which the horizontal drift measurement error can reach 1.11Pixel and the vertical drift measurement error can reach 2.22Pixel. Based on the test field calibration method, the camera was calibrated continuously, and the relationship between the camera temperature variation and the value of internal parameters was quantitatively analyzed. The statistical model was used to fit the parameters of the camera's participating temperatures, and the temperature error compensation model was established to realize the compensation of internal parameters. The experimental data show that the image drift error compensated by the model decreases by about 0.4Pixel in the horizontal direction and 0.9Pixel in the vertical direction, and the image drift quantity is significantly reduced and the measurement accuracy is improved. The temperature compensation model proposed in this study can take a series of corresponding temperature error avoidance measures to control the measurement error within the allowable range, which provides a corresponding method and theoretical basis for the application and practice of high-precision photogrammetry engineering.

Absolute inspection is a high-precision inspection method for optical components, in which the absolute inspection of two flat crystals cannot accurately measure the tilt error in the translation process, which leads to deviations in the quadratic term in the restored surface shape. The liquid level can be kept level before and after the position changes, and there will be no tilt error during the rotation and translation process. The relationship between the tilt error in the translation process and the defocus term in the restored surface shape is deduced and verified by simulation analysis. When there is no tilt error, the residual surface PV value is less than 1nm. The liquid surface rotation and translation interference experiment was carried out on a 300mm vertical Fizeau interferometer. The results were compared with the traditional three-sided mutual inspection method. The surface shape was basically the same, the deviation of PV value was less than 1.3nm, and the deviation of RMS value was less than 0.65nm, it proves that the liquid surface rotation and translation method will not introduce tilt error in the interferometric measurement process.

Product inspection of industrial parts is the most important link to ensure that the quality of the parts is qualified. The traditional contact detection method is difficult to meet the requirements of high efficiency and high precision in the industrial field. The image measurement system based on machine vision has been widely used to detect the geometric parameters of the product. A lot of work has been done on the measurement of high-contrast images formed by non-transparent objects, and relatively little researches have been done on low-contrast images formed by transparent objects. Right-angle prisms were used as the object, and an image measurement system for measuring the size of low-contrast products was developed. First, the images collected under different light intensities are averaged to obtain a smooth image. Then the contrast-limited histogram equalization algorithm is used to enhance the contrast of the image and the Zernike moment edge detection algorithm is used to determine the precise sub-pixel edge. Several comparative experiments have verified the rationality and superiority of the improved algorithm. The average error of prism thickness is less than 0.003mm, and the standard deviation is less than 0.0015mm. The solution proposed provides a feasible solution for the high-precision measurement of relatively transparent objects.

A polarization calibration method based on adjustable polarization source is proposed, and the calibration and correction of measurement results of degree of linear polarization of spectropolarimeters based on polarization spectral intensity modulation (PSIM) technique are completed by using this method. The linearly polarized light with different degrees of polarization from the adjustable polarization source is used as the source to be measured, and the original data is obtained from the output of the source to be measured by the PSIM spectropolarimeter. The correction coefficient is obtained by linearly fitting the polarization degree spectrum obtained by analyzing the original data and the theoretical polarization degree spectrum output by the light source to be measured. The results of the PSIM spectropolarimeter were calibrated and corrected by using the coefficients. The results show that in the effective measuring band (500~650nm), the measurement accuracy of the PSIM spectropolarimeter is significantly improved after calibration and correction, and the maximum error between the PSIM and the output standard polarization degree decreases from 0.017 to about 0.003, which indicates that the polarization calibration and correction method based on the adjustable polarization degree source is feasible.

The laser scanning projection system assists workers in assembling by projecting a contour frame on the target surface. Using this system for auxiliary assembly can effectively solve the problems of poor efficiency and low accuracy caused by traditional assembly methods. In order to make the self-developed laser scanning projection system directly read and project the data of complex curves in the digital model. The B-spline curve entity type is analyzed in the IGES (Initial Graphic Exchange Specification) file, and then a method for extracting scanning projection coordinates based on the IGES file and generating a 2D galvanometer scan drive file is proposed. The self-developed laser scanning projection system can be realized to directly read the data of the complex curve in the theoretical model, and the projection accuracy can reach 0.5mm. This method can effectively improve the production efficiency of workshop assembly and make industrial production more intelligent and efficient.

Three-dimensional (3-D) shape reconstruction based on the phase shift method has high accuracy and is not sensitive to environmental noise and shadows. However, it is difficult to apply to the 3-D measurement of dynamic objects due to the phase-resolving of multiple fringes. To this end, a new algorithm is proposed. First, the corner points of the rigid moving checkerboard are extracted based on the Harris algorithm, the pixel deviation between the captured fringe images of two adjacent frames is determined and the captured fringe image is corrected. Then, the spacial transform matrix between the projection image and the captured image are established according to the calibration parameters of the projector and the camera. According to the transform matrix, the corrected projection fringe image is obtained. Third, high-speed projection technology is used to reduce the pixel deviation between frames. Finally, the corrected projection fringe images are projected and the captured images are transformed, so as to obtain the captured fringes similar to the static objects. Experimental results have proved that the proposed method can reconstruct the three-dimensional shape of the flexible and rigid moving objects with a determined trajectory and speed with high accuracy.

As a kind of aspherical surface, off-axis aspherical surface is an indispensable optical device for space optical systems, astronomy and high-precision measurement system. In response to the significant demand of manufacturing technology of large-scale off-axis aspheric optical components in space optics, researches on precision grinding technology of off-axis aspheric mirrors have been carried out. The principle of grinding off-axis aspheric surface using generating method in a five-axis machine tool is analyzed. The coaxial parent mirror where the off-axis aspheric surface locates is divided into a series of spherical rings with different radii, and the workpiece coordinate system is established at the center of the coaxial parent mirror of the off-axis aspheric surface to be machined. The cup wheel grinds each ring on the coaxial parent mirror in turn, and the off-axis aspheric surface is formed by splicing some of the rings. Based on this method, an off-axis parabolic mirror with the aperture of 100mm is sucessfully machined.The width of the divided rings is set to 7.28mm. After one process which takes only 7 min, the PV value of the surface error reaches 7μm, and the RMS value of the surface error reaches 0.622μm.The experiment proves the feasibility of this method and provides an effective solution for off-axis aspheric rapid grinding.

The Silica microsphere lens with low refractive index and BaTiO3 microsphere lenses with high refractive index are used to image the sub-wavelength surface structure of data-recorded Blu-ray disc. The imaging properties of the microsphere lenses when they are spatially separated from the surface of the sample are observed in the experiments. The experimental results illustrate that the sub-wavelength surface patterns can still be resolved by the microsphere lens with a significant magnification, when the distance between the microsphere lens and the sample surface is about 0~6μm. The comparative experiment shows that the high-index microsphere perform different imaging properties in immersing mode, magnification and imaging scale, et al. The simulation results of electric filed distribution of the microsphere lenses illuminate that the imaging properties of microspher lenses are greatly effected by the electric filed distribution, which provide basic for the revealing of the mechanism of super resolution of microsphere lenses.

According to the band-pass filtering characteristics of chirped fiber Bragg gratings, a fiber core/cladding composite chirped fiber Bragg gratings filter is designed. Firstly, the spectral characteristics are simulated and analyzed based on the fiber coupling mode, transmission matrix theory and coordinate transformation method, and the double band-pass filtering characteristics of the grating are verified. Then, the spectra of the fiber core/cladding composite structures with periods of 536, 537 and 538nm, grating lengths of 8, 10 and 12mm, refractive index modulation depth of 0.0002, 0.0003 and 0.0004, chirp coefficients of 0.05, 0.10 and 0.15nm/cm were simulated. The half-peak width increases with the grating length, refractive index modulation depth and chirp coefficient. The peak reflectance increases with the increase of refractive index modulation depth and decreases with the increase of chirp coefficient. It is proved that this grating has double band-pass filtering characteristics and its amplitude bandwidth is adjustable.

The photoinduced optical non-linear effect in nematic liquid crystals is analyzed. On this basis, an optical-optical device is obtained based on the principle of liquid crystal displays. First, the beam propagation in a simple model which is obtained based on liquid crystal displays is given by Jones matrix. Secondly, the external voltage is replaced by an oblique incidence control beam to reorientate the molecule of the liquid crystals. Base on this principle, the phase and power of the out-put beam is modulated by the control beam. Finally, the influence of temperature on optical-optical control devices is investigated based on the temperature characteristics of 5CB liquid crystal. Numerically results indicate that the phase difference of out-put light increases monotonically with the raise of the control beam intensity, so the power of the out-put beam appears periodic oscillation with the increases of the control beam intensity. The out-put beam intensity of the signal beam can be linear modulated by the adjustment of the control beam intensity when the control beam intensity changes from 0.5(MW·m-2) to 1.1(MW·m-2). It is found that the power of the out-put beam shows periodic oscillation with the increases of the temperature when the control light intensity is fixed. This discovery shows a available scheme for the achievement of optical device.

As the current pedestrian detection method has the problems of large computation and low detection accuracy, an improved pedestrian detection method based on YOLOv4-Tiny was proposed. This method introduces Convolutional Block Attention Module into CSPDarknet53-tiny network to get richer features by learning the position information and channel information of the image, adds the spatial pyramid pooling module following CSPDarknet53-tiny, which can greatly increase the receptive field and isolate the most significant context features, and uses CIoU loss function to optimize the combined loss of multiple tasks. In the experiment, the training set in INRIA and WiderPerson are used as the training model, and the test set in INRIA and WiderPerson are used as the test set to verify the model. Compared with YOLOv4-Tiny, the precision, recall and average accuracy of the improved YOLOv4-Tiny network in INRIA test set are increased by 6.23%, 3.15% and 6.12%, respectively, and the improved network increased the precision, recall, and average accuracy in the WiderPerson test set by 3.65%, 3.28%, and 4.41%, respectively. It is found that this improved model can extract pedestrian features more easily and improve the detection accuracy on the premise that the real-time detection is hardly affected.

The current cable insulation layer thickness detection algorithm mainly uses image processing technology to extract the edge contour of the insulation layer. Such algorithms have problems such as excessively wide insulation layer edges and discontinuous edges, which affect the subsequent detection accuracy. In order to improve the measurement accuracy of the insulation layer, new algorithm is based on the RCF (Richer Convolutional Features) algorithm to improve, in the 4th and 5th stages of the model, the cavity convolution is used to increase the receptive field of the model; and the scale enhancement module (SEM module) is added to the side output network. And the cascade network from shallow to deep to increase the detailed information of the side output image. The model was trained through the self-made cable insulation data set. The results show that the improved model has 0.821 and 0.842 in the optimal scale of the data set (ODS) and the optimal scale of a single picture (OIS), respectively, and the average accuracy is 0.799. Compared with the RCF model ODS and OIS, the algorithm is improved by 0.008 and 0.01 respectively, and the detection accuracy is improved by 0.021. The performance of the model is further verified on the Berkeley University Data Set (BSD500) data set, where ODS and OIS are 0.810 and 0.825, respectively. Compared with the RCF model, the ODS and OIS of this algorithm are improved by 0.009 and 0.006, respectively.

To improve the spatial resolution of hyperspectral image, a super-resolution (SR) method based on GoogLeNet and spatial spectrum transform is proposed. Firstly, the spectral SR framework of remote sensing image is designed to extract different reflection spectra from the image. Then, the coarse pixel spectrum is amplified by using the sparse coding of GoogLeNet, and projected into the high-resolution dictionary to invert the potential SR representation to obtain the super-resolution spectrum. Finally, in order to improve the fidelity of image reconstruction, a coding and decoding structure based on GoogLeNet network is proposed to realize spatial spectral prior transformation. The proposed method is demonstrated experimentally on KSC and other dataset. The results show that the proposed method can effectively reconstruct the image details and texture structure, and the average peak signal-to-noise ratio (APSNR), average structural similarity (ASSIM) and spectral angle mapping (SAM) are better than other comparison methods, and the spectral information is better preserved. Taking KSC data set as an example, APSNR, ASSIM and SAM are 25.643db, 0.789 and 0.084, respectively.

Speckle noise exists in Optical Coherence Tomography (OCT) and affects the quality of OCT images. In the diagnosis of various common eye diseases by OCT equipment, high quality OCT images are extremely important. Deep neural network is used to reduce the noise of OCT images, so that the images can show more information on the basis of retaining the details of spatial structure. A novel OCT image denoising network, CMCNN, based on residual learning network, is proposed. It has the characteristics of multi-scale, multi-weight and multi-level feature fusion, and reduces image noise while preserving the spatial structure of image details. Then the proposed model is compared with traditional denoising algorithm and deep learning denoising model. Experimental results show that the peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) of CMCNN are improved by about 2.5% compared with other deep learning methods. It is verified that the proposed method can effectively retain the details of OCT images, suppress the noise and improve the image quality.

The detection range of the PCF-SPR refractive index sensor reported in the past is generally narrow, and the detection of low refractive index cannot be realized, and the wavelength is mostly concentrated in the visible waveband or communication waveband, which limits the application range of the sensor. In view of this, an SPR sensor based on D-shaped dual core PCF structure is proposed. Indium tin oxide is deposited on the polished surface of D-shaped PCF as plasma material, and the theoretical model of the sensor is analyzed, including the influence of metal parameters on sensing performance and the influence of PCF structure parameters on sensing performance. The results show that the sensor can achieve an ultra-wide detection range with a refractive index of 1.27~1.39. Its working wavelength is in the near-infrared region of 1300~2600nm, the highest sensitivity can reach 35000nm/RIU, the resolution is 2.85×10-6 RIU.The ultra-high sensitivity and ultra-wide refractive index detection range of the sensor can be applied to the detection of plasma, white blood cells［WBC］, Hemoglobin［Hb］, intestinal mucosa of huma, human liver, acetone and other biological tissues, as well as in the fields of medicine and chemistry.

Plaque is an unobservable biofilm on the tooth surface, which is a direct cause of a series of diseases such as dental caries and gingivitis. Early quantitative nondestructive detection of dental plaque is of great clinical importance. The bacteria of dental plaque and their metabolites can produce autofluorescence under short wavelength light excitation. The red fluorescence generated by a large amount of plaque under the excitation of 405nm blue light was collected based on the previous imaging system; the more mature the plaque, the higher the intensity of the red fluorescence; a modified U-net network was used to segment the red fluorescence, and the contour of the segmented plaque was extracted to obtain the core, and the teeth to which the plaque was attached were segmented using the area growth algorithm, and the plaque maturity and area were integrated to evaluate the plaque. The plaque maturity and area were combined to evaluate the plaque index. The results showed that the segmentation accuracy of the improved U-net network was better than that of the traditional method. The combination of plaque area and maturity degree to quantify dental plaque can eliminate the variability of human diagnosis to some extent.

In order to objectively evaluate the wearing comfort of the lens.Eye movement data is collected using the eye tracker in experiments and the "changing percentage of eyelid spacing" is proposed as one of the evaluation indexes to evaluate the degree of wearing comfort. A new model to evaluate the degree of lenses comfort based upon the deep neural network is designed by combining the new index with other traditional eye movement indexes. The experiment results reveal that the proposed index of "changing percentage of eyelid spacing" is a good indicator in evaluating the degree of lenses comfort, and the deep neural network model has a high prediction accuracy and generalization ability. The research overcomes the disadvantages of the traditional fitting process which mainly relies on the subjective feedback from the patients, and therefore can effectively help optometrists and opticians to adjust the fitting plan objectively and accurately.