The velocity fluctuation of the interferometer is one of the important factors affecting the spectral signal-to-noise ratio.To further improve the spectral signal-to-noise ratio,the relationship between the velocity fluctuation of the interferometer and the spectral signal-to-noise ratio is studied.In this paper,the causes of optical path difference velocity fluctuation of the tilt compensation interferometer are quantitatively analyzed from the aspects of rotation angle,velocity feedback testing,and circuit noise.On this basis,the influence law of optical path difference velocity instability on the spectral signal-to-noise ratio is verified through theoretical analysis and experiments.The experimental results show that in the same wavenumber range,the higher the velocity stability of optical path difference,the better the spectral signal-to-noise ratio.When the velocity stability is increased by 5%,the spectral signal-to-noise ratio is increased by about 14%—16%.When the optical path difference velocity is constant,the spectral signal-to-noise ratio in the low wavenumber band is significantly better than that in the high wavenumber band,and the signal-to-noise ratio in the 2 100—2 〖ＫＧ－１／６〗200 cm－1 band is about 20% higher than that in the 2 500—2 600 cm－1 band.The research results provide a theoretical basis for the design of a high signal-to-noise ratio spectrometer.

Birefringence is an important parameter in polarization sensitive fiber systems.Piezoelectric transducer based single-mode fiber (SMF) stretcher is a common device in optical fiber systems to introduce strain,optical path or phase changes,but people barely pay attention to the birefringence introduced in the fiber stretcher.In this paper,a characterization method of birefringence characteristics of SMF stretcher is proposed to obtain the birefringence distribution along the fiber wound in the stretcher,based on a distributed polarization analyzer achieved by introducing the full Muller matrix analyzing into the optical frequency domain reflectometry.From the experimental results we draw the following important conclusions:the birefringence of SMF wound in the fiber stretcher increases with the increase of driving voltage;when the surface of stretcher for SMF winding is uneven,a higher level background birefringence can be introduced,compared to a smooth surface,and the background birefringence increases more significantly when the driving voltage is applied;designing an appropriate stretching structure and fiber winding method can effectively avoid the change of birefringence during running the fiber stretcher,while a strong background birefringence may be introduced.This work will provide important guide for the evaluation and optimization of the polarization sensitive fiber systems in which the SMF stretchers are used.

A miniature Fabry Perot (F-P) optical fiber humidity sensor based on agar film is proposed.The sensor forms a double F-P structure by inserting a standard single-mode optical fiber into an empty core glass tube and dipping agar film on the end face of the glass tube.When the ambient relative humidity changes,the volume and refractive index of agar film change,resulting in wavelength shift of interference light.The humidity sensing experimental system is built to characterize the humidity sensing characteristics of the sensor.The relative humidity measurement sensitivity of up to 1.232 4 nm/% RH is realized in the relative humidity range of 50%RH—80%RH.The sensor has the advantages of compact size,low cost,good linear sensitivity and simple preparation method.

Organic light emitting diode (OLED) devices possess the characteristics of light weight,low power consumption,high response speed and flexibility.They have shown the great superiority in the flat panel display and solid-state lighting.In order to enhance the performance of OLED devices,the Expert OLED software was used to simulate the actual OLED devices with different grating structures.The electroluminescent (EL) spectrum,the luminous angle and the electric field distribution were analyzed in the OLED devices with bimetal electrodes.The optimized micro/nano grating structures was obtained.And for OLEDs with different grating periods,the varied grating height has different effects on the luminescence performance.We can also improve the luminescence performance by adjusting the molecular dipole orientation of emitting layer and optimizing the electric field distribution in the OLEDs.This work provides a valuable reference on improving the performance of OLEDs by using the optimized periodic micro/nano grating structures.

Aiming at the problem that existing scene text recognition methods only focus on the classification of local sequence characters and ignore the global information of the entire word,a multilevel feature selection scene text recognition (MFSSTR) algorithm is proposed.The algorithm uses a stacked block architecture and applies a multilevel feature selection module to capture contextual and semantic features in visual features. In the process of character prediction,a novel multilevel attention selection decoder (MASD) is proposed,which combines visual features,context features and semantic features into a new feature space,and re-weights the new feature space through a self-attention mechanism.While paying attention to the internal relations of the feature sequence,select more valuable features and participate in decoding prediction.At the same time,intermediate supervision is introduced in the training process to gradually refine the text prediction.The experimental results show that the algorithm in this paper can reach a high level of recognition accuracy on multiple public scene text data sets.In particular,the accuracy rate can reach 87.1% on the irregular text data set SVTP,which is improved compared with the current popular algorithms by about 2%.

With the continuous development of deep learning,machine vision methods based on deep learning are widely used,and the convolutional neural network (CNN) has remarkable effect on hyperspectral imagery (HSI) classification.The sampling position of the convolutional kernel in traditional convolutional networks is fixed and cannot be changed according to the complex spatial structure in HSI,ignoring the features of the data on spatial distribution.To improve the performance of hyperspectral image classification in practical applications,this paper proposes a deformable convolution-based hyperspectral image classification method,which extends the deformable convolution from 2D to 3D considering the high dimensionality of HSI,so as to better extract the features on 3D space.This paper combines the double-branch dual-attention mechanism network (DBDA) structure and 3D deformable convolution,and experiments are conducted on two datasets,Indian Pines (IP) and Botswana (BS),and the experimental results show that the method of this paper achieves better classification accuracy on overall accu-racy (OA),average accuracy (AA) and KAPPA evaluation criteria,and improves OA by 0.15%— 0.23%,AA by 0.21%,and KAPPA by 0.000 〖ＫＧ－１／６〗3—0.001 4 compared with the suboptimal algorithm.

Atmospheric turbulence has random effects on the resolution of optical target images.The lucky-region fusion （LRF）is an image synthesis technology aimed at atmospheric turbulence that affects images,it synthesizes a clear image by selecting high-resolution parts from a series of short-exposure images.The LRF algorithm is relatively simple to implement on desktop computer,but it is only a method of post-processing,without real-time performance.This paper introduces an LRF algorithm and its system implementation technology for real-time extraction,partition processing and synthesis of gray-scale image streams.According to the characteristics of field programmable gate array (FPGA) digital signal processing,a real-time LRF algorithm suitable for FPGA processing is proposed.The algorithm is logically designed with a hardware description language,and it is embedded in a small and medium-sized FPGA,so as to form a compact LRF processing system of pure hardware.The system was tested through simulated sequence images and short-exposure sequence images taken in the laboratory.The results show that the proposed real-time LRF algorithm is feasible,and the implemented FPGA system can realize the real-time dynamic fusion of lucky regions of the input gray-scale image sequence,and finally obtain a high-resolutionfusion image.

With the advancement of computer technology,the existing Transformer has been expanded into a network structure for processing computer vision tasks.In order to improve the early diagnosis rate of melanoma and the cure rate of skin disease patients,this paper proposes an improved network model based on PiT (pyramid pooling transformer) to realize automatic classification of dermoscopic images of seven skin lesions.The model of this paper is mainly composed of the PiT module and the anti-interference module.Pit inherits the advantages of ViT and uses pooling to perform spatial size conversion to improve the robustness of the model.The pre-trained PiT network has a large number of natural image features,and the PiT part of the network can provide the required image features for downstream classification tasks.In this paper,an anti-interference module is designed to resist the influence of interference factors (such as hair and foreign object occlusion) in the dermoscopic image,thereby improving the performance of the model.Improve classification accuracy.Experimental results show that the classification accuracy,precision,recall,and F1-score values of this model on the ISIC 2018 verification set are as high as 91.58%,83.59%,89.92%,86.34%,and the number of frames per second (FPS) reaches 85 Hz.Compared with several existing advanced classification networks,the classification performance and model efficiency have been improved,and it has relative advantages,which proves that the model in this paper has certain practical value.

The state detection of railway catenary insulators is of great significance to the safety of railway traffic.To solve the uncertainty of manual inspection on insulator inspection results,a detection method combining deep learning and gray texture features are proposed.First,the Faster R-CNN (faster region-based convolutional neural network) algorithm is used to accurately identify the insulators in the image,and then the texture features of the insulators are analyzed and extracted through the gray-level co-occurrence matrix.Then,the support vector machine is used to divide the insulators into normal insulators and abnormal insulators.The result of the experimental data proves that the classification accuracy of the normal insulators in the experimental data can reach 100%,and the classification accuracy of the abnormal insulators can reach 97.5% when the three texture features of energy,entropy and correlation are used to classify the insulator state.Finally,according to the periodic characteristics of the gray distribution of the insulator image,the abnormal insulators are divided into damaged insulators and foreign matter insulators by gray-level integration projection.Experimental results have showed that the proposed method can effectively detect and classify the state of insulators.

In the block diagonal (BD) precoding multi-user multiple input multiple output (MIMO) indoor visible light communication (VLC) system based on singular value decomposition (SVD),the bit error rate performance of different receivers of the same user terminal is quite different,and the performance of the user terminal is limited to the worst-performing receiver.To solve above problem,this paper uses nonlinear receivers of decision feedback equalization (DFE),and proposes a BD precoding multi-user MIMO indoor VLC system based on geometric mean decomposition (GMD).For the same user terminal,different field of view (FOV) receivers to improve the bit error rate performance of user terminal,and analyze the impact of user terminal location on the bit error rate performance of user terminals.Finally,the simulation results show that the proposed system in this paper enables different receivers of the same user terminal to obtain similar bit error rate performance,improves the performance of the worst receiver and reduces the bit error rate of the user terminal.When the emission power of a single light emitting diode (LED) is 10 mW and the system transmission rate is 100 Mbit/s,user terminals can achieve the bit error rate performance of about 10－6 in 95% of the indoor area.

The carrier and spin dynamics in solution-processed PbI2 film on quartz substrate have been systematically studied by transient reflectance (TR) spectroscopy and femtosecond resolved Kerr rotation spectroscopy. The results show that TR spectroscopy includes one band of photo-induced bleach (PB) centered at 502 nm and two bands of photo-induced absorption (PIA) at 487 and 522 nm,which are interpreted by band filling effect and band gap renormalization effect.The Kerr magneto-optical signals excited by left-handed and right-handed circularly polarized light are completely opposite in sign and almost equal in size.With the increase of pump fluence,the Kerr rotation angle and ellipticity increase linearly at first,then slowly,finally reach maximum value,~10 degree and 0.12 per micro,respectively.At the same time,the spin relaxation life decreases a minimum value ~1.6 ps.According to our experimental data,the spin relaxation mechanism in PbI2 thin films is attributed to the Elliott-Yafet process owing to the strong spin-orbit coupling caused by the heavy atom lead.In addition,we find that Kerr effect has a large signal around the band gap, which suggests that the spintronic devices based on PbI2 have high sensitivity around the band gap.These experimental results obtained in this paper are of great significance for exploring the potential applications of PbI2 thin films in spintronic devices.

Aiming at the problem of large force fluctuations of each agent in the process of group movement,which leads to group movement oscillations,a control algorithm based on the combination of distributed artificial potential field method and fuzzy control is proposed.By analyzing the force of the agents in the process of group movement,establish the dissipation force to optimize the oscillation during the movement,reduce the negative impact of the oscillation during the movement of the group,and improve the consistency of the group movement.Establish retention to optimize the stability and maintenance of the formation,and improve the stability of the formation.Use the optimized resultant force as the input of the fuzzy controller,adapt to environmental changes through intensive learning,adjust the parameters to control the output,and realize the group follow-up movement control.In addition,a variety of formation control generation models that can be automatically adjusted are given to realize group formation movement control.The simulation results show that the group can be stably and effectively controlled to follow the movement,and the formation can effectively avoid obstacles during the movement,which can improve the movement efficiency of the formation and maintain the topological stability.

After prostate magnetic resonance (MR) image slices,it is found that some images do not have effective edge information,which makes it impossible to clearly locate the edge position,and thus cannot segment the prostate.At the same time,the traditional convolutional network requires a large amount of parameters and takes up too much storage space of the model.This paper proposes a method to segment the prostate using U-Net that combines multi-scale dilated separable convolution and channel attention.First,slice 50 three-dimensional (3D) prostate samples and perform contrast enhancement on the sliced images.Subsequently,the processed data is input into the residual U-Net,and the multi-scale dilated convolution and channel attention are used as the encoding-decoding unit to extract the feature information.Finally,the Dice coefficient and Hausdorff distance (HD) are used to evaluate the segmentation results.The experiment was verified on the PROMISE12 challenge dataset,and the final Dice coefficient and HD were 88.13% and 14.17 mm,respectively,and the parameter amount and storage space were reduced by 57%.The results show that this method can not only segment the prostate area without effective edges to improve its segmentation accuracy,but also effectively reduce the parameter amount and storage space,and can be applied to medical images with blurred edges.