In order to obtain a high uniformity of chlorophyll fluorescence active induced excitation source and more contrasting chlorophyll fluorescence parameters, the ratio of the minimum irradiance to the mean irradiance in the sample area was proposed as the index of excitation uniformity for the irradiance distribution of the sample area under different numbers of light sources and at different distances. Three kinds of basic LED arrays A0,A1 and A2 were simulated by spatial simulation. The results showed that the irradiance uniformity was positively correlated with the LED array height and LED spacing, and the irradiance uniformity increases with the increase of the number of LEDs. In the research area of 0.04m2, when the LED spacing is 0.04m and the array height is 0.5m, the irradiance uniformity of A2 array is the largest, which is about 0.920. The uniformity of excitation light distribution of LED arrays was verified using a handheld illuminance meter, and the experimental results were consistent with the simulation results. The fluorescence images were collected by CCD camera, and the leaf fluorescence distribution under different excitation array arrangement was analyzed, which provides an idea for future analysis of the difference of leaf regional emission fluorescence under high uniformity excitation light.

In order to design high stop-band loss photonic crystal stop-band filter (PCSBF) structure, performances of the structure are optimized by adjusting coupling length between the waveguide and rectangular resonator, the size of the rectangular cavity and inner m×n rods radii etc. Operational features of the PCSBF based on m×n rectangular resonator are qualitatively analyzed by using time domain CMT (Coupled-mode theory), and strong coupling stop-band loss transmission conditions of the PCSBF are obtained. Output performances of the PCSBF are studied by using FDTD method between 1440nm wavelength and 1625nm wavelength. Normalized stop-band spectra curves show that this kind of micro PCSBF possesses characteristics of high normalized stop-band loss, tunable stop-bands, flat top stop-bands and so on. It possesses potential application value in the removal of unneeded wavelengths in the fields of optical fiber communications and integration of optical paths.

The thickness measurement sensor of alkali metal gas chamber based on spectral confocal is a non-contact thickness measurement sensor, which can obtain the thickness value of transparent material by directly measuring the wavelength information of light wave. The dispersive objective is the core component of the thickness measurement sensor. The linearity and dispersion range of the dispersive objective lens determine the accuracy and resolution of the thickness measurement sensor. The principle of thickness measurement sensor based on spectral confocal is introduced. The relationship between wavelength information and axial dispersion range is analyzed. The dispersive objective lens is designed by using ZEAMX optical design software, and optimizes it with multiple structures. The axial dispersion range of 0.801mm is achieved in the wavelength range of 420~620nm. The linear fitting of wavelength and axial dispersion by least square method is 0.9975. With a high resolution spectrometer of 0.001nm, the measurement accuracy of the sensor can reach nanometer level, which can meet the needs of high-precision measurement of alkali metal gas chamber wall thickness.

A 64-channel APD array LiDAR system based on the full-waveform sampling technology was constructed. The signal processing circuit is designed using discrete components, and the output signal of the circuit is sampled in full-waveform using a data acquisition card. The acquired waveform data is processed by a least-squares Gaussian decomposition method, and the distance and characteristics of the target are analyzed by optimizing the parametric calculations. A common optical path experimental platform was built to test the circuit output waveforms and range accuracy of all channels. The results show that the signal processing circuit meets the design requirements of low noise, wide bandwidth, and distortion-free amplification; the maximum RMSE of the system is 16.10 mm and the minimum is 8.47 mm. Finally, step imaging experiments were conducted to provide a reference for future development of larger APD arrays of non-scanning LiDAR.

Based on the principle of reflective electro-optic phase modulation combined with polarizer and Faraday rotator, a non-reciprocal phase modulation scheme for fiber ring interferometer is proposed. The structure makes use of the different tangential electro-optic characteristics of lithium niobate crystal to form two interference signals with adjustable phase on the ring interferometer, especially by introducing a total reflection prism whose energy loss is small, so as to realize non-reciprocal phase modulation, the half-wave voltage of the phase modulator is reduced effectively. The measured reflection (modulate twice) structure phase modulation half-wave voltage is half lower than transmission (modulate once). and used in the debugging of fiber-optic ring interferometer to obtain the linear correlation between the detected signal and modulation signal phase. It provides a feasible reference for the development of the next principle prototype.

Image measurement technology based on linear CCD is an important field in current engineering applications. In view of the contradiction between the current high-precision large dynamic range measurement and the standard linear CCD measurement range and linear CCD geometric structure, a high Linear array CCD measurement splicing scheme with large precision was proposed. The scheme uses the principle of half-reflective half-lens plane reflection to design the optical-mechanical structure principle of the optical splicing system of double-line = array CCD high-precision splicing. The calibration principle of overlapping pixels is given, and the calibration and splicing errors are analyzed. The long-term laboratory experiment results of the splicing system show that the splicing scheme is simple, practical and reliable. The overall splicing error of the system is about 0.019mm, and there is no leakage in the splicing, and the splicing accuracy meets the requirements of high-precision measurement. The stitching scheme has certain reference significance for the practical application of high-precision, large dynamic range CCD measurement.

In order to meet the imaging requirements of OLED defect monitoring, a linear scanning lens of high-resolution is designed. The optical system is designed on the basis of double Gauss optical structure, the focal length of the lens is 116 mm, the magnification is 0.5, the object side numerical aperture is 0.038 and the linear object field of view is120mm.The modulation transfer function (MTF) of the len can reach more than 0.2 at Nyquist frequency of 100lpmm which can be obtained by Zemax simulation, and the diffraction limit can be reached. All the field of view points are basically within the Airy patch, and the RMS radius is small. The symmetry characteristic of double gauss is used to reduce the distortion, and the relative distortion is less than 0.5%. According to imaging requirements for adjustable depth of field and exposure( that is the switching requirements of F# ) and structural arrangement of optical components, the opto-mechanical structure design with variable aperture diaphragmof lensis provide. Finally, the imaging quality of the lensis detected by MTF tester. The experimental results show that thelens can meet the industrial imaging needs of OLED screen defect detection function.

Due to the small object distance and large divergence angle of light source in the existing laboratory microscope, it is impossible to observe the cells in the culture tank in real time. A micro system for external observation of cell culture tank with different capacity is designed. The working distance of the system is 45mm, NA=0.19, and the image resolution δ=10.8um. The design results show that the meridian coincides well with the Hu vector line, which is close to the theoretical diffraction limit. At the space frequency of 88lp / mm, the MTF of each field of view is above 0.05, which has high resolution. The imaging size of each field of view is smaller than the Airy spot radius, and the imaging quality is good.

The bidirectional reflection distribution function (BRDF) of the machined surface morphology of optical elements was analyzed based on C-T model, a simulation model of the surface morphology was established for the precision cutting surface of optical components, the influence of incident light parameters and surface microstructure parameters on BRDF was simulated and analyzed by Matlab software. The results show that the peak value of BRDF function increases with the increase of incident Angle and incident wavelength under the same micromorphological parameters. When the parameters of the incident light are the same, the change of the parameters of the microscopic morphology will cause multiple scattering and shielding effects of the incident light on the surface, which will change the peak value of the BRDF function. With the increase of root mean square height, the scattering and shielding effects increase, and the peak value of BRDF function decreases. The peak value of the BRDF function increases with the increase of the contour width and decreases with the increase of the root mean square slope.

Detecting the concealed object from millimeter wave images is one of the key techniques to construct an intelligent millimeter wave based security inspection. To address the issue that the concealed objects are inspected hardly due to their locality and low identifiability in the millimeter wave images, a dynamic self-attentive bilinear convolutional neural network (DSA-BCNN) is proposed to train with image-level labels to detect the concealed objects. Self-attention mechanism is utilized to guide network to extract the features from concealed object regions, which enhances the network capability to depict the global information. Simultaneously, the second order features are constructed by bilinear pooling to enrich the representation of subtle differences between concealed objects and non-detected regions. Experimental results verify the propose method effectiveness, which is superior than others in terms of each evaluation metric, and the accuracy is 93.6%.

The traditional Convolutional Neural Network (CNN) algorithm-based abnormal crowd behavior detection methods adopt two-dimensional convolution kernel to extract image features, which causes such methods impossible to accurately capture the dynamic features of video streams in time series. To solve this problem, a detection method based on the combination of improved C3D network and random forest (RF) algorithm is proposed. Firstly, the C3D network with temporal feature capture capability is used to extract the Histogram of Oriented Gradient (HOG) feature of the video stream, which is used as the input of the three-dimensional convolution kernel to realize the extraction of the temporal and spatial features of the video. Secondly, a Random Forest (RF) classifier is used to replace the softmax fully connected layer to avoid cumbersome gradient calculation operations during the training process, and to reduce the requirement on the sample size of the training data set. Finally, calculation results of the examples based on the benchmark data set show that the proposed improved C3D-RF scheme maintains an accuracy rate of over 90% for the detection of abnormal crowd behaviors. Compared with traditional C3D network, Support Vector Data Description model (SVDD), Convolutional Auto-Encode (CAE) and other machine learning classifiers, the training time of the proposed scheme is shortened by more than 15.34%.

Body actions automatic detection techniques are usually influenced by imaging angle and imaging environment, these factors lead to detection accuracy reduction, in view of this, a new action detection technique combined monocular vision and depth sensor is proposed. Firstly, the key frames corresponding to body actions are filtered, this process helps to fix the problem of action speed difference; then, the convolutional neural network features and magnitude histogram are combined as action descriptors; finally, adversarial networks are adopted to realize domain adaption transfer learning for body action, in order to solve the problem of imaging anger difference of monocular vision. Validation experiments are carried on a public action detection dataset, and the results of which show that the proposed technique improves action detection performance, and the technique overcomes the detection performance reduction caused by monocular vision imaging anger difference, effectively.

In the task of pose measurement based on monocular vision, traditional convolutional neural networks suffer from the problem that the measurement accuracy reduces dramatically in both fuzzy background and complicated background, a deep learning model based on capsule network and Bayesian network is proposed, further a pose measurement method is proposed based on moving robot and monocular vision. First of all, the new Capsules Network is adopted to locating the important joint points of monocular vision objectives; then, a simple learning algorithm for Bayesian networks is designed, and the attitudes of joint points are inferred by Bayesian networks. Finally, validation experiments are carried on complicated human pose measurement datasets, the experimental results show that the proposed method realizes a good measurement accuracy, it remains a high-level accuracy in complicated background too, the result F1-measure values equal 0.9 and 0.78 for indoors and outdoors scenarios respectively.

Due to the high noise of Positron Emission Tomography (PET), the existing image denoising effect is not ideal. Therefore, a new method combining residual U-Net neural network and Deep Image Prior (DIP) is proposed. Firstly, residual learning is introduced into the U-Net network to improve the network expression ability and convergence speed. Then, a DIP algorithm without training data is proposed. The neural network is interpreted as the parameterization of the image, and the noise is removed by using the high impedance characteristics of the parameterized noise to achieve the purpose of noise reduction. Finally, the real data obtained from the brains of living monkeys injected with 18f-2-fluorodeoxyglucose (18F-FDG) were used for simulation analysis. The results show that the proposed method can get a clear and smooth image. In different noise levels and time frames, the denoising effect of the proposed method is better than other contrast methods, and can obtain high-quality images.

In order to solve the lack of multi-scale detail information in the original single source image and the noise problem in image fusion, a multi-scale image fusion enhancement algorithm based on wavelet transform is proposed. According to the idea of using different fusion rules to the sub-band components of different frequencies, three kinds of fusion methods for high frequency sub-bands are proposed, and then, a novel multi-scale residual pyramid space which participates in the fusion process to enhance the multi-scale information of the image and reduces the fusion noise are built. The results of various wavelet decomposition and comparison experiments show that the wavelet multi-scale fusion enhancement algorithm proposed can effectively reduce the fusion image noise and enhance the multi-scale detail information of the image to a certain extent.

An optical watermarking reconstruction method based on deep learning is proposed. The watermark is encrypted by double random phase encryption and the encrypted image is embedded in the host image. Then the physical relationship between watermark image and watermarked host image is used to train an improved neural network. Using the model of FC-DenseNets-BC can reconstruct the watermark image. In the traditional optical watermarking technology, the quality of watermarked host image and decrypted watermark image depends on the selection of the embedding strength. However, using deep learning to reconstruct the watermark image can get rid of this dependency. The simulation results show that the method can reconstruct high-quality watermark image with peak signal-to-noise ratio over 35dB even when the embedding strength is as low as 0.05. It also has a certain generalization, safety and ability to resist noise and shear. The feasibility and efficiency of this method are further verified by the experiment of optical system.

The thermal microscope imaging technology can obtain non-contact imaging for fine target objects according to temperature changes, which is an important research direction in the field of photoelectric imaging technology. Due to the limited accuracy of the processing technology at present, the scanning position error will occur during the operation of the optical micro-scanning thermal microscope imaging system developed in the early stage. It will have a certain effect on the subsequent image embedding reconstruction, and the resolution of the reconstructed image is decreased. To solve the problem, a method of micro-scanning error correction by combining image pre-processing thought, micro-scanning principles, and calculating pixel-correlation is proposed. The results of simulation and experiment show that this method can improve the quality of image reconstruction and the spatial resolution of the system. The method can be directly applied to other photoelectric imaging systems to improve its performance.

Aiming at the ghost problem in multi-exposure image fusion, a multi-scale exposure image fusion algorithm based on improved intensity mapping function is proposed. Firstly, the high and low contrast of the input frame is determined based on the reference frame image. Secondly, the high contrast region is used to detect the structure consistency to get the ghost area and the improved energy function based on the intensity mapping relationship is used to further detect the ghost information in the low contrast area. Finally, the multi-scale patch matching algorithm is used for fast fusion. Experimental results show that compared with the existing representative methods, the proposed method can effectively remove ghost region and retain the color and detail information of the image.

In order to overcome the shortcomings as spectral distortion of fused images induced by only using the single energy feature to fuse image and ignoring the spectral characteristics of images coefficients in many remote sensing image fusion algorithms, a remote sensing image fusion algorithm based on second-generation curved wave transform coupled definition weighted model is proposed. Firstly, the multispectral image is computed by IHS transform, and the intensity components of the multispectral image are segmented. The high and low frequency coefficients of panchromatic image and I component are calculated by second generation curved wave transform. Then, the spectral features of the image are measured by means of the image features, and the regional energy features of the image are combined as the fusion rules of the low frequency coefficients. The definition of the image is calculated by using Laplace feature of the image, and the definition weighted model is constructed by using the calculated results to fuse the high frequency coefficients. Finally, the fusion coefficients are fused by the second generation Curvelet and IHS inverse transformation. The experimental results show that, compared with the current fusion algorithm, the proposed algorithm has higher general image quality index value and lower spectral difference index value, which indicates that the proposed algorithm has better fusion performance.

In order to improve the accuracy of magnetic resonance image segmentation, a magnetic resonance image segmentation method based on residual network and wavelet transformation is proposed. First of all, the discrete wavelet transformation is adopted to fuse different sequences of magnetic resonance images, it leads the fusion image contains more texture information and structure information; then, a residual network including channel attention model and spatial attention model is proposed, thus the network can focus on the target region, and the residual block is included to reduce the vanishing gradient problem of deep neural networks. Finally, validation experiments are carried on the public Brain Tumor Segmentation Challenge 2015 dataset, the results show that the proposed method achieves good effect of average Dice score for whole tumor area, core tumor region and enhanced tumor region.