In view of the special requirements of the beam parallelism deviation of the Space-borne corner cube prism, a aperture of 10mm corner cube prism used in the Pina satellite laser ranging cooperation target is introduced as an example: Firstly, through the traditional process to process the moderate size of the corner cube prism; Then, the precise control of the beam parallelism deviation is realized through the Counterweight separator processing; Finally, the small-diameter corner cube prism is processed through the "big to small" process, The final product qualification rate can reach 83%。Practice has proved that this process can overcome the problem that the beam parallelism deviation cannot be accurately controlled and is suitable for the processing of small caliber star-borne angular cone prism.

The working principle of the four-quadrant detector used in laser guidance is introduced. According to the technical requirements of the laser seeker and the working characteristics of the four-quadrant detector, the principle and characteristics of the optical receiving system in the seeker based on the four-quadrant detector for space positioning are analyzed. With the sequential ray tracing and non-sequential ray tracing,an optical lens is designed. Its illumination uniformity is more than 75% in the ±4°linear receiving field. Its illumination uniformity is 90% in the ±1.5° receiving field. At the same time, the spot diameter deviation of the lens in the 4 linear receiving field is less than 0.05mm. The lens design result not only meet the design requirements but also have a compact structure.The optical design method of this kind of lens is explored.

Digital holographic microscopy can measure quantitative light field information, but holographic phase reconstruction usually requires spectral filtering, analog diffraction, phase unwrapping, distortion compensation and other steps, and the size error of the manually selected filter window will greatly affect the filtering, image quality. A one-step digital holographic phase reconstruction method based on DeepLabV3+ network is proposed. The method introduces MobileNetV2 structure on the basis of DeepLabV3+ network for improvement. Firstly, MobileNetV2 is used to extract hologram features, and then multi-scale features are fused through hole space pyramid pooling. Finally, bilinear interpolation method is used for upsampling to obtain high-precision quantitative phase reconstruction results. The experimental results show that compared with the reconstruction using PhaseNet, the method improves the structural similarity index by 6.5%, and can accurately and efficiently realize the high-precision quantitative phase reconstruction of digital holography.

With the increasing demand for spatial resolution performance of astronomical observation, it is urgent to develop telescopes with optical synthetic aperture of 10 meters, 100 meters or even kilometers. Traditional optical synthetic aperture imaging technology faces the challenge of high precision phase technology, it is difficult to extend the long baseline SAR imaging to optical band. The measurement method of intensity interference is relatively insensitive to phase, which provides a possible technical scheme for optical synthetic aperture with long baseline. The development history and key scientific problems of the stellar intensity interferometer and the speckle interferometer are reviewed respectively, and the technical characteristics and development potential of the space intensity autocorrelation synthetic aperture telescope are analyzed. By summarizing the characteristics of the three techniques, it can provide reference for the development of long baseline optical synthetic aperture imaging system.

Aiming at the problems of under-segmentation and over-segmentation in the road clustering method of point cloud extraction, a road point cloud extraction method based on slope filtering algorithm and improved European distance region growth algorithm is proposed. First, the original point cloud data is large and complex, and statistical analysis is used for preprocessing to remove a certain amount of dangling noise and reduce the amount of data; secondly, in order to prevent the loss of accuracy of point cloud data, the preprocessed point cloud is divided into grids, and combined with the slope filtering algorithm, the non surface point interference is removed to obtain the ground point cloud; finally, taking the angle of normal vector and Euclidean distance as constraints, the region growth algorithm is used to extract road point cloud. Two sets of vehicle point cloud data in different scenarios are used to test the effectiveness of this method.

Laser paint stripping technology has the advantages of no pollution, no mechanical contact and real-time monitoring. It is expected to become the main technical way of aircraft paint removal. The main working mechanism of laser paint stripping technology was analyzed. The influence law of process parameters on paint removal effect was summarized, and the selection idea of process parameters in engineering paint removal was put forward. The evaluation methods of laser paint stripping effect were summarized. The application examples of aircraft skin engineering laser paint stripping were introduced, and the future research focuses of laser paint stripping technology were prospected.

Based on the matrix optics theory of the Total Reflection Prism Ring Laser (TRPRL) and the working theory of the ring laser in the Ring He-Ne Laser, mode-selection techniques for ring laser with the Total Reflection Prisms are studied. The clockwise and counterclockwise laser in the ring laser can work in the state of single spectral line, single transverse mode and single longitudinal mode, so as to form a stable standing wave in the ring laser, thus a stable interference fringe signal waveform is output to the photodetector. By shortening the capillary diameter of TRPRL, 3.39μm spectral lines can be suppressed; Because of the special structure of ring laser of TRPRL,the Frequency-selecting Diaphragm designed with the special split structure can be placed at any position of the optical path of the ring laser, and multiple Frequency-selecting Diaphragm can effectively suppress 1.15μm spectral line; Based on the different beam waist spot radius of laser with different wavelengths, a special elliptical small aperture diaphragm can be used at the beam waist to suppress 0.63μm Higher Order Transverse Modes of laser. It is of great significance for the application of TRPRL in Total Reflection Prism Laser Gyro(TRPLG).

Aiming at the requirement of high precision, fast and real-time polarization measurement, a polarization parameter measurement method based on double elastic optical difference frequency modulation is proposed. The principle of polarization measurement is analyzed in detail, and a multi-channel digital phase-locked data processing scheme based on Field-Programmable Gate Array (FPGA) is designed for the work control and signal demodulation requirements of the dual photoelastic modulator. The FPGA provides a Pulse Width Modulation (PWM) wave with a certain frequency duty cycle, and outputs a sinusoidal high voltage through the high voltage resonant circuit to drive the elastic optical modulator to work., at the same time, the modulated optical signal detected by the detector is converted by analog to digital, the Analog to Digital Converter (ADC) enters into the FPGA after acquisition. The FPGA provides a local reference signal to complete the simultaneous demodulation of multiple frequency signals, and then a single measurement the four parameters of the Stokes vector can be obtained. An experimental system was built for experimental verification, and a polarization generating device was established by rotating 1/4 wave plate method, which realized the measurement of linearly polarized light, circularly polarized light and elliptically polarized light. The experimental results show that the relative error of the system's measurement is less than 0.8%, the repeatability standard deviation is less than 0.2%, and the single measurement time is less than 200ms, realizing high-precision, repeatable, real-time polarization measurement.

In order to guide the processing for grinding to polishing large aperture aspheric mirrors, Ronchi testing system is designed. Firstly, the null compensator is used to replace the compensation grating, which not only for detection of rough machining process by Ronchi test, but also for the testing of interferometer. Secondly, the relationship between the fringe shape and tested surface aberration in the Ronchi system is simulated, and polish process is guided by judging the fringe shape. Then, for a 600mm diameter hyperboloid mirror, Ronchi grating test system, null compensation system and imaging system are designed. The influence of residual reflection of components in test system is analyzed, and the resolution methods are proposed. The result shows that in the Ronchi test system, 5% residual reflection on the plitting prisms surface relative to 0.1%, the illumination of the dark stripes on the image plane increases more than 30 times, and the contrast of the brightness and shade stripes decreases, which is difficult for resolution. When the residual reflection of the grating is greater than 3%, there are bright spots in the center of the stripe, and illumination of the bright spots is 5 times more that of the stripe. By optimizing the structure of the system, the residual reflection on the surface of the component is reduced, and contrast of the image stripes are improved. Ronchi test system was used to detect the polishing process of hyperboloid mirror, and the RMS value after rough polishing of hyperboloid was measured to be 0.042λ, PV is 0.291λ. It can guide process of fine grinding and polishing of aspheric surfaces effectively.

Birefringence measurement methods such as interferometry and compensation need to rotate the sample or polarization element to be measured, which is inconvenient to operate, slow to measure and limited in accuracy.A birefringence measurement system composed of double photoelastic modulation and PC is proposed. The modulation signal is generated by the photoelastic modulation, the information of the sample to be tested is loaded, and the modulation signal is collected to the PC for data processing, so as to further complete the solution of the retardation and the fast axis azimuth angle.The principle of the test scheme is analyzed, and the upper computer design and implementation of the system data processing are emphasized, and the software and hardware system is built.The experimental results show that the relative error of delay measurement is 2%,and the relative error of fast axis azimuth measurement is 0.4%. The standard deviations of the retardation and fast axis azimuth angle are 0.056nm and 0.022°.

In order to achieve high and flatting spontaneous radiation output, a flatting broadband light source based on 974nm pump source and Er-doped fiber with high doping density is proposed and designed. Firstly, the pump light is divided into two parts to pump Er-doped fibers. Then, the two parts of light are combined and output by wavelength division multiplexing. Finally, adjusting the length of Er-doped fiber and pump power to optimizing the output spectra. According to the result in simulation, when the pump power is 160mW, the length of Er-doped fibers are 7m and 2m, the spectra with output power of 20.348mW and bandwidth of 80nm can be obtained from 1530nm to 1610nm. The flatness of spectra from 1530nm to 1610nm is ±1.268dB. The system can output flatting wide spectra without adding filters and the length of Er-doped fiber less than 10m, which is expected to be used in optical fiber sensing system.

The temperature of high-voltage cable will seriously affect the use of cable. It is of great significance to monitor the internal temperature of wire core through skin temperature. By analyzing its working principle and demodulation mechanism, it is proved that there is an obvious linear relationship between the temperature and the central wavelength of the fiber Bragg grating. The standard relationship of three core cable is improved by shape factor, the equation between core temperature and skin temperature is constructed, and its accuracy is proved. The fiber Bragg grating real-time temperature measurement system is built. Through 20 ~ 100 ℃ calibration experiment and error analysis, the effectiveness of the temperature measurement system is proved, and the linear relationship between temperature and central wavelength is verified. Through experimental verification, it is proved that the temperature measurement system can effectively realize the temperature measurement of high-voltage cable and distinguish the abnormal cable.

In response to the problem of low efficiency of manual verification of medicines, a method for recognizing tablet position with improved MobilenetV3-Small network structure is proposed. The acquired original images are converted to the LAB color space, and the L channel under the LAB color space is employed with canny edge segmentation, morphological processing, image filling and single pill image frame selection, so as to position a single pill; the network structure of the MobilenetV3-Small is improved by loading pre-trained weights, using the K-means clustering algorithm to optimize the input size, modifying the network layer according to the feature map, etc.; by combining with the positioning method of image processing and the recognition method of improved MobilenetV3-Small network, the rapid and accurate detection of tablets is realized. Finally, on the 523 single pill images processed, the Loss value of the pill recognition is 0.000035, which verified the effectiveness of the algorithm.

A multi task gait recognition method based on multi-scale convolutional neural network and human posture estimation model is studied, explains the recognition results of neural network, and improves its recognition effect in the face of covariate changing scenes. In this method, the gait spatial features extracted by convolutional neural network and the human posture estimation model are fused with the temporal features of human joints for identity recognition. Experiments were carried out using normal and synthetic walking sequence data in gait dataset CASIA-B and TUM-GAID gait dataset. The results show that the recognition rates of the three scenes T1, T2 and T3 are 95.2%, 72.4% and 84.5% respectively. In the experiment of CASIA-B gait dataset, for normal walking sequences and two kinds of synthetic walking sequences, this method has good performance in recognition accuracy, which reflects that the model has strong robustness.

Precise liver segmentation is crucial for the localization and treatment of liver cancer, and in view of the problems of different liver shapes and sizes, as well as the difficulty of segmentation of edges and lesion areas, a liver segmentation network based on multilayer sensor and multi-scale feature extraction (M2U-Net) is proposed. The network is divided into convolutional phases and multilayer perceptron phases. First, an extrusion excitation module is added to the encoder portion of the convolution phase to highlight specific liver segmentation tasks and inhibit irrelevant background areas. Secondly, a tokenized multilayer perceptron module is added to the multilayer perceptron stage to reduce the complexity of the model. The transition layer adds a multi-scale feature extraction module to adapt to the segmentation of livers at different scales and the segmentation of detail areas. Finally, experimental results on the LiTS dataset and the dataset provided by Oriental Hepatobiliary Hospital show that the segmentation network is better than the segmentation networks such as U-Net, U-Net++ and CE-Net on three evaluation indicators.

Aiming at the defects of existing infrared image denoising algorithms in edge restoration and preservation, an infrared image denoising method based on complex wavelet and maximum a posteriori estimation is proposed. This method takes full advantage of the excellent characteristics of dual tree complex wavelet transform, such as multi-resolution analysis, translation invariance and multi-directional selectivity, performs dual tree complex wavelet transformation for noisy infrared image; and based on the assumption of probability density distribution of Gaussian noise and noiseless image, the maximum a posteriori estimation of wavelet coefficients of noiseless image is made in the wavelet domain to realize the denoising and restoration of infrared images. Infrared image denoising experiments have proven the effectiveness of the proposed method, this method not only removes noise effectively, but also maintains and recovers the edge details perfectly, the image quality indexes PSNR and SSIM of denoising are 1dB more and 2% higher than the existing methods, respectively.

Lung lesion detection and disease diagnosis based on chest X-ray images(DR) is a routine clinical operation. For pulmonary tuberculosis patients, the lesion area of tuberculosis in the DR image is highly compatible with the background, the target diffusion is serious, and the edge shape is extremely irregular, which seriously interferes with the accuracy of diagnosis. To solve the above problems, a Tuberculosis Deep Transfer Net(TDT-Net)integrating the imaging characteristics of pneumonia is proposed. Using the characteristics that tuberculosis and COVID-19 are respiratory infectious diseases and have similar imaging manifestations on DR images, with the help of a large number of pneumonia DR images, strong correlation features are introduced through transfer learning to improve the detection accuracy of tuberculosis lesions. TDT-Net combines transformer and extended residual technology, and proposes a context-aware enhancement module to strengthen the modeling ability of the migration model for global context information; The feature refinement module is used to reduce the redundant information introduced in the transfer process and highlight the representation of strongly related features. The experimental results show that on the TBK11k dataset, the Average Precision (AP) of the proposed detection method reaches 87.5%, and the Recall reaches 80.7%. Compared with the networks such as YOLOV5 and RetinaNet, the detection accuracy of tuberculosis lesions is effectively improved, and more accurate localization and classification of tuberculosis lesions are achieved.