Obtaining three-dimensional ice shape data on the surface of the model in a low-temperature icing wind tunnel test is beneficial to analyze the effect of icing on the aerodynamic characteristics of the model and provides a basis for optimizing the model's anti-icing design. Due to the low-temperature environment of the low-temperature icing wind tunnel site, the core components of the existing visual measurement methods are difficult to work stably, and after the test, the temperature of the wind tunnel site will rise and the effective measurement time will be short. Therefore, ice shape measurement is facing great challenges. To this end, a low temperature protection system combined with thermal balance and external insulation was first designed to ensure that the core components in the equipment work at an appropriate temperature. Based on this, an ice type combining projection acceleration and low temperature protection was proposed. The fast three-dimensional measurement method, through the cooperation of the light source and the micromirror, to complete the two-stage exposure to achieve gray-scale recombination, which reduces the total flip period of the micromirror when projecting the image and improves the measurement speed. The experimental results prove that this method can realize stable and fast three-dimensional measurement of ice shape under the low temperature environment of -20℃．

Narrow line-width lasers were widely used in optical sensing. The frequency drift and noise of lasers directly affect the measurement accuracy of optical sensors. The frequency drift and noise characteristics of narrow line-width fiber lasers and semiconductor lasers were measured by self-timed optical fiber delay method. Firstly, the theoretical calculation of frequency drift measurement was carried out. The relationship between the length of delay fiber and frequency drift was simulated and analyzed, and the length of delay heterodyne fiber was established. Then, a frequency drift test device for narrow line-width laser based on self-heterodyne structure was built, and the composition of the test device was described. Finally, the frequency drift and noise test of narrow line-width laser was carried out based on the test device. The test results show that there is obvious frequency drift in both fiber laser and semiconductor laser. The drift speed is about 2MHz/10s, and the frequency noise of the fiber laser is also lower than that of the semiconductor laser, which is consistent with the measurement results of the frequency noise of the two types of lasers.

Compared with LED lamp, laser lamp has the advantages of small energy consumption, small volume and high brightness. It is a new direction of automobile lamp development. An optical system structure of low power and high brightness laser lamp is proposed. By building an optical test platform, the characteristic parameters of laser-driven white source are studied, which include the shape and size of the light spot and luminous flux. The light source model is established. The mapping relationship between the energy of point source and receiving surface is solved, and the initial surface shape of lens is solved. According to the source model studied, the initial lens is optimized, and the final design meets the requirements well. The designed aspheric lens is realized by ultra-precision machining, and the laser lamp is assembled and tested. The illumination at 25m is about 106lux driven by 12.16W electric power.

The scheme of emitting lens direct collimating semiconductor laser source can make the linear scanning lidar more compact and cheaper, but the high power semiconductor laser source heating seriously will lead to the thermal deformation of optical elements, which will lead to the sharp reduction of the optical power received by the detector and make it undetectable. A thermal integration optimization design method for the optical path of line-scanning laser radar with a detection range of 30 meters is proposed. This method firstly takes the preset working temperature from 40~80℃ as the initial condition, and optimizes the optical path system of the transmitting lens and receiving lens based on Zemax software, so that the optical performance of the optical path system operating at 60℃ is the best. Secondly, the finite element method was used to analyze the thermal deformation of optical elements when the optical path and the corresponding mechanical structure changed with temperature, and the mechanical structure of the optical path system was optimized by adding SiO2 aerogel as thermal insulation material. Machine heat integration optimization results show that the optimization design method, the optimized optical path and mechanical structure in the working temperature of 40~80℃ range detector receives the light power in 10-4w level all the time, compared with only using Zemax software optimization design method of emission and receiving lens (detector receives the light power 10-6~ 10-4w) has significant improvement.

The optical properties of polymer solar cells based on MoO3/Ag/MoO3 (MAM) transparent conductive electrode and ITO electrode were studied by transfer matrix method. The thickness of coupling layer, activation layer and metal electrode were optimized respectively, and the structure with excellent efficiency was obtained. The results show that MAM and ITO organic solar cells have obvious differences in optical properties under different active layer thickness and optical spacer thickness. The results show that the maximum short-circuit current density can reach 16.85mA/cm2 for thin active layer (100nm), which is 7.3mA/cm2 higher than that of ITO device. For thick active layer MAM electrode (270nm), the optical performance of ITO device is obviously better than that of MAM device. In addition, by changing the thickness of LIF, the critical thickness of optical spacer 30nm is obtained in our simulation.

In the covid-19 epidemic period, in order to investigate the research and application of various ultraviolet disinfection technologies, low-pressure mercury lamp and ultraviolet light-emitting diode ( UV-LED ) as the mainstream ultraviolet disinfection light sources in the market are investigated. The principle, application and main characteristics of the two light sources are briefly introduced and the research progress, application sites and potential of UV-LED single wavelength irradiation, pulsed irradiation and multi-wavelength synergistic irradiation are introduced and evaluated. The 222nm safety ultraviolet disinfection technology was introduced and its application was prospected. Finally, the problems of UV-LED at the present stage are discussed and the prospect of the development of UV-LED disinfection market is presented. It is expected that the future UV disinfection market will form mercury lamp, LED as a supplement, the two complementary pattern.

With increasing peak voltages, surface discharge patterns are formed successively on a dielectric target, which include diffuse spots, single rings and concentric triple rings in a system where an argon column hit the dielectric vertically. During the evolution of these patterns, the number of positive discharge increases while the number of negative discharges remains one per voltage cycle. The mechanism of pattern formation is studied using fast photography. It is shown that?the positive and negative discharges in each pattern are caused by streamer mechanism, and the formation of these patterns is originated from the temporal superposition of positive and negative surface discharges. In each voltage cycle, the first positive streamer results in a central point of each pattern; the last positive streamer leads to the formation of diffuse background of each pattern; other positive streamers finally forms rings in patterns. The negative streamers corresponding to each pattern are always diffuse. The propagation characteristics of these positive streamers are qualitatively explained by analyzing the influence of applied electric fields, residual charges and air diffusions.

Due to the development of visible light communication, indoor visible light positioning technology has received extensive attention。In the visible light positioning system based on the image sensor, single frequency modulation ID scheme is simple and easy to implement. However, it is limited by the upper limit of the collection frequency of the terminal, and the number of IDs that can be provided is too small, which is difficult to meet the application requirements of the actual larger area scene. For the scenario where hundreds of IDs need to be provided, the result of OR operation of two square waves with different frequencies is proposed as the LED ID. Using ordinary commercial mobile phones to carry out experimental research. The results show that, compared with the use of a single frequency as the LED ID, the number of IDs that can be provided in this system has increased by 600%, 153 types of IDs can be provided, the identification ID accuracy rate is 100%, when the area of the LED lamp being blocked reaches 1/3, the ID can still be correctly identified.

As the precision manufacturing industry's requirements for the shape of optical components increase, the calibration of interferometer system errors is becoming more and more important. Among them, using the liquid level reference method as the interferometer system error calibration scheme has the advantages of high precision and low cost. Without considering the influence of external factors, theoretically the curvature of a liquid flat is the same as the radius of the earth, which can be regarded as an ideal plane. However, because the liquid is more sensitive to the environment and not easy to control, how to establish a high-precision liquid level reference is the most important link in the whole process, which directly affects the system accuracy. Due to the fluidity characteristics of the liquid, the previous time-phase shifting method is no longer applicable, so a point source ectopic vertical Fizeau interferometer is used for dynamic measurement, and comprehensively analyze the properties of the liquid and the liquid plate to complete the establishment of the liquid reference.

Generally, simplified calibration models are preferred in existing reconstruction algorithms for fringe projection profilometry (FPP) to improve numerical stability. However, the lens distortions of the camera and the projector of a FPP system cannot be compensated effectively with such a simplified calibration model and the measurement accuracy of FPP cannot be assured as a result. Therefore, a two-step calibration method, in which lens distortion compensation and system calibration are performed successively, is presented. The lens distortion compensation is performed by correcting the values of input variables with calibrated parameters of the camera and the projector. Therefore, lens distortions of the camera and the projector can be compensated effectively, and 3D shape data with higher accuracy can be retrieved accurately with a concise calibration model. The validity and practicality of the proposed method are verified with experimental results.

Aiming at the problem of low spatial resolution in conventional optical frequency domain reflectometry(OFDR) systems utilizing external modulation laser source, a method of an OFDR system based on IQ modulation tunable laser source(TLS) is proposed. The principles of the basic OFDR system and the IQ modulated frequency sweeping laser source are analysed. The frequency stabled incident laser is modulated utilizing IQ modulator to generate frequency sweeping light signal. The simulation model of OFDR system based on IQ modulation is established. The temperature or strain changes are generated on the fiber under test(FUT) by altering the phase on the local backscattering light. A differential FFT demodulation method is proposed to detect the position of parameter variation with a high speed. The simulation results demostrate that the system is able to sense and locate the simulated parameter changes on FUT with high accuracy, and a 0.23 m spatial resolution with 100 m sensing range is achieved. The research demostrates that the system optimizes the spatial resolution, reduces the complexity of the demodulation algorithm and improves the parameter sensing accuracy, which provides theoretical basis and measurement methods for OFDR applications in short range strain and temperature sensing with high spatial resolution.

Terahertz time-domain spectroscopy is a powerful tool for spectral research in terahertz band. But its detection speed and signal quality need to be further improved. An all fiber direct acquisition terahertz time domain spectrum system based on fast optical delay line is proposed. The experimental results show that the scanning frequency of the system can reach 16Hz under 64ps time window, and the dynamic range of frequency domain is more than 85dB. Compared with the traditional phase-locked scanning system, not only the signal quality of single frame THz time-domain pulse is better, but also the detection time is shortened to 1/40 of the original, which provides a feasible solution for the practical application of THz time-domain spectrum system.

Haze reduces the clarity and details of the image, and thus the impact on subsequent visual information processing is a challenging issue. Although the existing image dehazing algorithm can remove the haze, it is less effective for processing outdoor dense fog scenes. In order to break through the limitations of the existing defogging algorithm, this research combined with the atmospheric scattering physical model, proposed an end-to-end multi-scale parallel fusion defogging network. The network uses multi-scale convolution to extract features of different scales from the whole to the part, and fuse these features in parallel multiple times. In addition, by introducing a residual module to carry out in-depth learning of detailed features, more image details can be recovered. Experimental results and data analysis show that the proposed method can exhibit good defogging performance on both synthetic and real images, with PSNR and SSIM indicators increasing by an average of 3% year-on-year.

In recent years, the research of facial expression recognition has achieved good recognition accuracy, but in the actual environment, due to the influence of posture, occlusion, light and other factors, its detection accuracy has a little weakening effect. To solve these problems, a new FER system based on the dual-stream convolutional networks is proposed. A dual stream CNN from two aspects of appearance and geometric characteristics is established . The network based on appearance features is to extract the local derivative pattern features of the preprocessed image as the input, whereas the geometric feature-based network learns the coordinate change of action units’ landmark, which is a muscle that moves mainly when making facial expressions. In addition, a technique to generate facial images with neutral emotion using the autoencoder technique is be proposed. By this technique, the dynamic facial features between the neutral and emotional can be extracted images without sequence data. The detection accuracy of the algorithm is 98.81% and 96.05% respectively on CK+ and Jaffe datasets, which shows better results compared with other latest methods.

In order to solve the problem that the spatial feature of the fusion image is not ideal because the significant content of the image is ignored in the process of information fusion, a remote sensing image fusion algorithm based on the second generation curvelet transform coupled with the significant content determination mechanism is proposed . With the help of HSV transform, the lightness component of multispectral image is calculated; then the panchromatic image and V component are calculated by the second generation curvelet transform, and the corresponding frequency-domain subband of them is output. Based on the amplitude spectrum characteristics of the image, the salient information of the image is calculated. Through the segmentation of the image, the salient content determination mechanism is established based on the salient information of the segmented image. According to the salient value of the segmented image, different methods are used to fuse the low-frequency subband. Finally, using the gradient value of the image, the detail measure factor is constructed to calculate the detail information of the image and realize the fusion of high-frequency coefficients. The experimental results show that compared with the existing remote sensing fusion scheme, the fusion image of the proposed algorithm has better spectral characteristics, showing higher standard deviation and correlation coefficient values.

In order to make the fusion image of visible and infrared better express the target information, the non-subsampled Shearlet transform and split strategy are combined to fuse them. First of all, the high and low frequency components of visible and infrared images are obtained by using the non-sampling shearlet transform. Then, the Otsu thresholding segmentation method is used to develop the stripping strategy for ping the target layer and its background layer of infrared image. Based on the target layer, the low-frequency coefficients of the background layer of the visible image and the infrared image are weighted by using the regional energy characteristics of the image, and the result is combined with the target layer of the infrared image to obtain the fusion low-frequency coefficients which are rich in the target content and background content. The regional variance function is introduced to measure the detail features of the image, and by constructing the variance weighting factor, the fusion high-frequency coefficient which riched in detail features is obtained. Finally, the non-subsampled Shearlet inverse transform is applied to the two fusion coefficients for outputting the fused image. The experimental results show that the fusion image of this algorithm has better target and detail expression ability than the existing fusion image, which can be used to obtain high-quality visible and infrared fusion image.

Segmentation of whole brain regions has important clinical significance for the diagnosis and treatment of brain diseases. In order to improve the accuracy of segmentation, a multi-atlas segmentation algorithm for polynomial expansion registration is proposed. Firstly, by using the model of linear polynomial expansion and combining affine transformation and non-rigid body transformation, the target image to be segmented and the map image are registered one by one to obtain the displacement field. Then, the similarity between the target image and the atlas is calculated by normalized mutual information, in which the atlas with high similarity to the target image is screened out. Then, the marked image of the atlas is mapped by displacement field, and the rough segmentation result is obtained. Finally, the global weighted voting method is used to fuse the obtained coarse segmentation results to obtain the final fine segmentation results. 35 T1-Weighted MRI images from the MICCAI 2012 multi-atlas Labeling challenge and 10 T1-Weighted MRI images from Shanghai Ninth People’s Hospital, Shanghai Jiao Tong University School of Medicine are selected to verify the method. Finally, DSC values of the whole brain region are 0.7585 and 0.7351. Experimental results show that the algorithm has high accuracy and robustness in brain segmentation, which is expected to assist clinical diagnosis and treatment of brain-related diseases.

In recent years, deep learning algorithm develops rapidly and is widely used in the task of target detection. However, real-time target detection cannot be carried out on devices with limited memory and computing power. To solve this problem, a fast pedestrian detection method is proposed in the surveillance system with limited memory and processing unit. Firstly, aiming at the problem of low detection efficiency caused by extracting high-dimensional pedestrian features in general pedestrian detection, an improved directional gradient histogram (HOG) and Sobel edge local binary pattern (Sobel LBP) are fused as features. Secondly, a model compression technique based on teacher-student framework is proposed, which is applied to random forest (RF) classifier without deep network, because the compressed deep network still needs a lot of memory to process parameter multiplication. Students' random forest (S-RF) (born again random forest, BARF) is trained to imitate the performance of teachers' random forest by using the soft target of teachers' random forest output. Then the pedestrian detection is carried out by BARF classifier, and finally the pedestrian detection is carried out by sliding window method. In experiments, the proposed method achieved up to a 2.05 times faster speed and a 5.39 times higher compression rate than T-RF and its detection performance is also ideal.

Aiming at the problem of low dynamic gesture recognition accuracy in complex environment, a dynamic gesture recognition technique in complex environment based on attention mechanism convolutional neural network is proposed. First of all, the long short term memory network is adopted to learn the weight of each filter, and predict human appearance correlated filter banks; then, the convolutional neural network is used to extract the trajectory images, and construct a color trajectory image; finally, trajectory images are delivered to attention convolutional neural network to train, the trained neural network is taken advantage to recognize the target gesture in complex environment. Experimental results indicate that the proposed gesture recognition algorithm can detect and track the dynamic gestures, at the same time, it realizes a good gesture recognition accuracy.

KDP crystal plays an important role in inertial confinement fusion (ICF) optical system. However, how to obtain KDP crystal elements with the application requirements is still a difficult problem. The processing technology of KDP crystal flat element is studied using fly-cutting technology. Firstly, the technical principle of fly-cutting and the influence factors of the surface roughness are explained. And then, the influence factors in processing are analyzed with single variable method through a mass of processing experiments. The experimental results show that the roughness of KDP surface is affected by the diamond tool parameters, the processing parameters and the cleaning method, but the diamond tool parameters have the greatest influence on the roughness. The ultra-smooth surface can be obtained by using diamond tool with rake angle of -45 °, tool radius of 5.0mm and optimum processing parameters, and its corresponding Sa is less than 1 nm. The research results provide an effective technological scheme for the processing of KDP crystal flat elements by fly-cutting, and have extensive engineering application value.