Mode converter, achieving the mode conversion from fundamental mode to higher-order mode, is a key component for the on-chip multimode transmission and mode division multiplexing transmission. Here, a nanowire-loaded mode converter based on the thin film lithium niobate (TFLN) waveguide is proposed. The mode conversion structure is consisted of a nanowire grating and a phase change material based rectangular nanowire, where both of them are deposited atop the TFLN waveguide. Based on such structure, detailed structural analyses and optimizations are conducted, where the required conversion length is only 6μm and the central wavelength is 1.55μm for the mode conversion from input TE0 mode to output TE1 mode. The mode conversion efficiency, crosstalk, and insertion loss are 97.6%, -19.7dB, and 0.31dB, respectively. The device structure is further extended and the mode conversion is botained from input TE0 mode to output TE2 mode, TE3 mode and TE4 mode in the same length. If the device structure is obtained, other higher-order modes can also be obtained. The proposed device structure and scheme could benefit the multimode transmission for the TFLN waveguide and boost the development of photonic integrated components and circuits based on the TFLN platform.

Conformal metasurfaces can break the limitation between an object’s geometric shape and its optical function, and significantly improve the optical properties of any curved object. This extends the capability of metasurfaces to components with arbitrary shapes. In previous research on conformal metasurfaces, polarization multiplexing technology was not applied, limiting the multi-functionality of these metasurfaces for different polarization channels. A polarization-sensitive conformal metasurface on a curved substrate was proposed by us.. By designing the unit structures of the conformal metasurfaces based on propagation phase, such metasurface can achieve different optical functions for incident light with different polarization states, and achieve functions such as curved holography and optical stealth. With high flexibility, the conformal metasurface design can be embedded in various non-planar systems to achieve multi-functionality, and has broad application prospects in areas such as military security and wearable electronic devices.

The existing super-resolution microscopy technologies based on iterative algorithm to realize super-resolution imaging of fluorescence microscopy and obtain the accurate value of each point center in the fluorescence image by convergence to extreme values are faced with problems such as complex image data fitting process and high computing power requirements. In order to improve the computation speed, a fast point spread function parameter extraction algorithm based on linear regression is proposed, which does not require an iterative process. Experimental results show that the compared with the existing comparison algorithm that can accurately calculate the center position and FWHM (Full Width at Half Maximum) of the image point, although the calculation accuracy is slightly lower, the calculation time of this algorithm is less than 20% of the comparison algorithm. Moreover, the calculation results obtained by this algorithm can be used as the initial parameters of more accurate comparison algorithms, which can reduce the overall calculation time of the comparison algorithm by 30%. This algorithm can also be used as a real-time point spread function FWHM (Full Width at Half Maximum) calculation algorithm, which can be applied to microscope autofocus.

Rapid and accurate three-dimensional positioning of particles has important applications in life science, material science, industrial detection, and other fields. The traditional particle location methods cannot meet the requirement of fast and accurate because of heavy computation. A new 3D particle locating method based on phase mask is proposed. Fienup optimization algorithm was used to design the phase mask on the spectrum plane of a 4f imaging system, and a series of point spread functions were obtained for different depth of point light sources. By training the neural network based on the particle positions and generated images, the target particle can be located in 3D space with high precision by a single measurement within the axial range of 8 μm. Numerical simulation results of two sets of design show that the proposed method can quickly and efficiently complete the three-dimensional locating of sparse particles, and may have important applications in the locating of cell particles.

Aiming at the demodulation technology of coherent OAM multiplexed beam in OAM communication system, an amplitude-only diffractive deep neural network (D2NN)-based implementation method for OAM coherent multiplexing is proposed. The performance of the demodulator is investigated via numerical experiments and characterized by the bit error rate (BER). An advanced OAM selection strategy is proposed to reduce the BER of D2NN. And compared with the phase-only D2NN demodulator, simulation experiments verify that the amplitude-only D2NN has obvious advantages in high demultiplexing and demodulation accuracy for quad-OAM coherently multiplexed beam, which provides a flexible robust real-time demodulation approach for OAM coherent multiplexing communication.

When the beam is transmitted in turbulent atmosphere, the wavefront of the beam will be destroyed with the increase of the transmission distance due to the existence of atmospheric turbulence, which is not conducive to the extraction of the information carried by the beam at the terminal. Based on the generalized Huygens-Fresnel principle, this paper investigates the evolution behaviors of one-side limited extended edge dislocation and optical vortex in turbulent atmospheric transmission by taking Gaussian beams with the two dislocations as the research object. The results show that because of the different bending degree of the edge dislocation, with the increase of the beam transmission distance, the one-side limited extended edge dislocation will disappear or evolve into an optical vortex after disappearing. As the transmission distance continues to increase, optical vortices induced by atmospheric turbulence will appear in the wavefront of the beam. When the beam is transmitted far enough, the optical vortex induced by atmospheric turbulence and the optical vortex evolved by edge dislocation will annihilate, or annihilation will occur between the optical vortices induced by atmospheric turbulence. The optical vortex carried by the beam itself is stably transmitted throughout the transmission process. The research results of the paper have important applications in the field of optical communication.

Liquid crystal materials are an important component of display devices, and their high birefringence under the action of external fields is a crucial condition for their use in display. Using a strong terahertz wave as an external field to excite the birefringence effect of liquid crystal 5CB material, the birefringence effect of terahertz field induced liquid crystal in the terahertz band was studied. With a terahertz time domain spectroscopy system, we measured the transmittance of terahertz waves with different polarization states. The relationship between the birefringence effect of liquid crystals and the external field strength, as well as the frequency response characteristics, were analyzed. The results show that the birefringence of liquid crystal increases with the increase of field strength provided by strong terahertz waves, and the birefringence value is proportional to the square of field strength, and there is a threshold value for the birefringence effect of liquid crystal 5CB excited by terahertz waves.

Aiming at the problems of weak Raman spectrum signal and long detection time of weakly scattered samples, a fast laser confocal Raman spectrum detection method based on Crop Mode is proposed. The method reduces the exposure time by reducing the signal transmission unit, and only collects the effective Raman spot area of the signal target unit. On the basis of guaranteeing the intensity of Raman signal, the signal-to-noise ratio of Raman spectral is also improved to reduce the required detection time and realize fast Raman spectral imaging of weakly scattered samples. Experiments show that compared with the traditional confocal Raman spectral detection system, the Crop mode adopted in this paper can improve the imaging speed of confocal Raman spectrum by more than 60 times, which provides a new technical way for the rapid detection of confocal Raman spectral microscopy technology.

A method is proposed to evaluate the laser irradiation accuracy of helicopter optoelectronic systems by using a shortwave infrared camera to monitor the laser irradiation spot and calculate the irradiation accuracy. Infrared detector with appropriate pixel size, band and sensitivity, lens with appropriate focal length and field angle, 3m ×3m cross target and camera set up the test environment to realize the laser spot camera monitoring and data recording of the airborne photoelectric system. According to the laser spot monitoring video image captured by the monitoring equipment, the laser irradiation accuracy of the airborne photoelectric system is calculated by using the data processing method, and is applied in the flight test, which is feasible.

To solve the problems of improving retrieved speed accuracy and optimal parameters design in spatial filtering velocimeter (SFV), a weighted average of logarithm of speed retrieval algorithm is researched based on computer simulation. With well designed simulation experiments, the moving targets’ images by a linear COMS image sensor and retrieved speed accuracy are both evaluated. Within the targets’ speed range of 0.1~100m/s, the designed SFV could retrieve targets’ speed correctly. The relative mean speed and standard deviation errors are -0.250% and 0.623%, respectively. The weighted average of logarithm speed retrieval algorithm with different power thresholds is researched as well based on the computer Monte Carlo simulations. The simulation results show that a suitable power threshold in SFV would improve the retrieved speed accuracy. Take a moving target’s speed of 100m/s for example, the retrieved speed accuracy improves from 1.065m/s to 0.381m/s (or relative speed error improves from 1.065% to 0.381%) with the power threshold decreases from 0 dB to -5dB.

A kind of multi-parameter tracking effect evaluation system based on trajectory is presented. Firstly, the principle and mode of target miss distance interpretation are expounded. Then the structure of tracking effect evaluation system is designed, which consists of image interpretation module, tracking effect evaluation module and result display module. Then the evaluation model of target tracking effect is established, which includes five parameters: trajectory difficulty, tracking error, tracking stability, equipment field of view, and tracking time. The system interprets the task video image acquired by the optical measuring equipment, and then transmits the interpreted data to the tracking effect evaluation module for processing and analysis, and evaluates the tracking effect according to the score of the evaluation model. Finally, the system test is carried out, which shows that the system can scientifically evaluate the tracking effect.

Laser backscattering imaging is the imaging of scattered light generated by the interaction between laser and biological tissue. It is widely used in the quality classification of agricultural products. By using laser backscatter imaging and deep learning, potato quality classification under different storage conditions was realized. The laser backscattering imaging data collection system was established based on the theoretical analysis of laser backscattering imaging. The laser backscattering image collection was carried out on potato samples, and the laser backscattering imaging data sets of fresh potatoes, refrigerator storage and room temperature storage potatoes were obtained. The data set is trained using the improved VGG16 network, and the training results are compared with the DenseNet121 network and the original VGG16 network. The results show that the classification accuracy of the improved VGG16 network is 95.33%. The results show that the combination of laser backscatter imaging and deep learning can achieve intelligent classification of potato quality.

In order to solve the problem of low efficiency of massive capsules screening in laser inertial confinement fusion experiments, a rapid capsules screening method based on improved YOLO-V5 deep learning model was proposed. In this method, the capsules were imaged in different scene depths, and the images were spliced together to obtain the clear images; At the same time, the channel attention mechanism was introduced to enhance the feature extraction ability of the model, and the SE-YOLOV5s deep learning capsule surface defects recognition model is established, and the capsule defects are classified and evaluated according to the defect types to achieve the screening of massive capsules. The accuracy of capsule surface defect detection is 94.4%, with fifty capsule images (resolution 3072×4096) detected per second, providing a fast and accurate method for screening massive targets for laser inertial confinement fusion test.

The three-dimensional (3D) measurement speed of optical-sectioning structured illumination microscopy (SIM) has always been an important factor in the application of this technology. Present 3D measurement methods require at least two exposures at the same axial position to obtain the sectioned image. A 3D measurement technique based on single-exposure SIM is proposed. Firstly, one SIM image is taken at each axial position, and there is a certain phase shifting between the fringes of adjacent axial positions. Then the axial light intensity curve corresponding to each pixel is analyzed, the axial modulation curve is calculated, and the peak value is located. Finally, the 3D reconstruction result of the sample can be obtained through calibration and scaling. Experiment proves that our method can obtain 3D result as accurate as the traditional optical sectioning method, with the measurement efficiency and image processing efficiency greatly improved.

Power generation from coal-fired boilers generates a large amount of polluting gases. This is based on tunable laser absorption spectroscopy combined with chromatographic imaging to reconstruct the temperature and concentration of premixed flames, and use these parameters to regulate the operating conditions of boilers as a means to reduce polluting gas emissions and improve energy efficiency. After the spectral analysis, two absorption spectral lines (7149.058cm-1 and 7150.4716cm-1) near 7148.8~7151cm-1 were selected as suitable for high temperature reconstruction of H2O, and different temperature and concentration fields were simulated and reconstructed by adopting adaptive iterative algorithm, BP-neural network algorithm and convolutional neural network algorithm. It was found that the convolutional neural network algorithm outperformed the other two algorithms in terms of reconstruction accuracy and stability. To investigate the effect of error on the reconstruction results, it was found that the error had less effect on the convolutional neural network algorithm by adding random errors, and the temperature and concentration reconstructions were highly accurate. In order to verify the feasibility of the convolutional neural network algorithm, different combustion conditions were selected for reconstruction comparison. The study shows that the reconstructed images of the convolutional neural network algorithm tend to be flatter and more consistent with the actual combustion conditions. The study also demonstrates the advantages and feasibility of the convolutional neural network algorithm in the reconstruction of combustion fields.

In order to overcome the issues of clock synchronization required in time division multiplexing (TDM) technology and the inflexibility of frequency division multiplexing (FDM) systems,an accurate visible light localization algorithm based on code division multiplexing is proposed.The algorithm realized symbol synchronization in asynchronous conditions and performs visible light positioning.The simulation results show that when covering more than 80% of the area of a room, the average positioning error of the algorithm is 6.9cm, and the maximum positioning error is about 11cm, which is suitable for most indoor location-based services. In addition,the relationship between the positioning accuracy of the system and the field of view of the receiver and the height of the receiver above the ground is analyzed.It is found that in a certain range,the larger the field of view of the receiver,the poorer the positioning performance of the system;while the larger the height of the receiver above the ground,the better the positioning performance of the system.

Image quality assessment (IQA) method is designed to measure the image quality in consistent with subjective ratings by computational methods. In this research, a valid no reference IQA (NR-IQA) method for color blurred image quality assessment is proposed based on local color appearance, e.g., clarity, in CIELAB color space. In the proposed method, the maximum local clarity and the variability of local clarity are combined to evaluate blurry level. The sharpest spot of an image is represented by the maximum clarity and the variability of clarity expresses the variation in the image content. Massive experiments are performed on five publicly available benchmark databases between proposed method and other state-of-the-art NR-IQA method, for the accuracy, complexity, and generalization performance of IQA. The results show that the weight average accuracy SROCC and PLCC of the proposed method can achieve 0.9345 and 0.9379, the direct average accuracy SROCC and PLCC of the proposed method can achieve 0.9331and 0.9357. The commonly evaluation criteria results prove that the proposed method work better than the state-of-the-art and newly NR-IQA methods for the overall performance on blurry images. These results of test and comparison above show that the proposed method is effective and feasible, and the corresponding method has an excellent overall performance.

Aiming at the problem of low imaging quality of infrared night vision remote sensing system, a multi-modal image fusion method with separation of the target and background, to improve the imaging quality of infrared night vision. On the one hand, the attention U-Net is adopted to segment and fuse the target region of the infrared image and the visible light image, the powerful leaning ability of U-Net is taken advantage to preserve the target information of the source images; on the other hand, the background regions of the infrared image and the visible light image are decomposed through the guided filter, different fusion strategies are used to fuse base layer information and detail layer information, so that the salient regions of the background are enhanced. Compared experiments are carried on the TNO dataset, the results show that the proposed method outperforms the other compared methods on both subjective visual evaluationand objective quantitative evaluation.

To address the problems of existing medical image segmentation models, such as high computational complexity and large number of parameters, which make it difficult to deploy the models into real-time medical-aided diagnosis systems, and the existing lightweight models, such as the degradation of segmentation performance due to parameter reduction, an improved lightweight U-Net segmentation model is proposed. The model consists of three main components: encoder, decoder and hop connection. First, the encoder uses a multi-scale fusion module formed by the combination of standard convolution and depth-separable convolution, based on which a bottleneck layer structure is introduced to enhance the learning ability of the neural network, and a lightweight cross-level partial network module is designed as a feature extractor for feature extraction of the input image using an aggregation method. Secondly, the lightweight module is continued to be used in the decoder to further optimize the model, reduce the computational complexity of the model and the number of parameters, and produce better segmentation effects. Finally, the fusion of feature information at different resolutions between the encoder and decoder is achieved by means of hopping connections. Experiments were carried out on abdominal organ CHAOS and Chest X-ray data sets. The results showed that the number of parameters and computational complexity of the improved U-Net segmentation model were reduced to varying degrees. When the number of parameters was only 1.28M, the DSC was 87.53% and 95.85%, respectively. The IOU values are 85.25% and 92.21%, respectively, and the segmentation performance is not inferior to that of other networks.