In order to investigate the influence of terahertz polarization-maintaining waveguides on the transmission signal of the terahertz communication systems,we propose the power budget model of waveguide links.The effects of the parameters (length,loss,dispersion,and birefringence) of polarization-maintaining subwavelength dielectric waveguides on the communication performance of amplitude shift keying (ASK) modulation systems are investigated.We calculate the received power,signal-to-interference-plus-noise ratio (SINR),coherence and incoherent demodulation bit error rate (BER) of two orthogonal polarization state signals for different transmitting power,waveguide lengths,modulated signal characteristic time widths,and modulated signal periods.The results show that the communication system can achieve high-speed (3.33 Gbit/s) and low BER (less than forward error correction limitation,10-3) by designing the waveguide structure.This work reveals the influence of length,loss,group velocity dispersion (GVD) and birefringence of terahertz subwavelength waveguides on communication performance,which has positive significance for the development of broadband communication systems based on terahertz waveguide links.

The introduction of partial coherence theory has enriched the research in the field of optical imaging resolution.Compared with the two-point imaging system ideally illuminated by completely coherent and completely incoherent light,the resolution of the partially coherent imaging system has significantly improved.The two-point imaging system under illumination of partially coherent light is designed and the mathematical model of image intensity distribution through the imaging system is constructed.MATLAB is used to simulate the effects of the radiance characteristics of different spatial coherent light source and two-point separation on two-point resolution.The resolution of the partial coherent imaging system is determined:when the peak and valley values of the image intensity distribution are equal,the two overlapping points can be determined resolvable.The calculation methods and differences of imaging system resolution under partially coherent illumination are also presented.We design an imaging system experimental platform and the experimental results show that the imaging resolution are 1.0 mm, 1.4 mm,and 1.5 mm when the coherence degree are 0,0.9,and 1.0,respectively,and the error range is less than 5%.The experimental results are consistent with the theoretical results.

A hydrogen sensing model with subwavelength circular hole array structure based on Pd/Au composite film was proposed to detect hydrogen leakage and hydrogen concentration.Finite-difference time-domain (FDTD) method was used to simulate the subwavelength circular hole array structure with the Pd/Au composite film.The electric field energy distribution and transmission characteristics of the composite film cross and vertical section were calculated,and the extraordinary optical transmission (EOT) characteristics and sensing characteristics of the sensor structure were further analyzed under different structural parameters.The simulation results show that the subwavelength circular hole array structure with Pd/Au composite film can produce obvious EOT effect,and there are two prominent abnormal transmission peaks,denoted as peak Ⅰ and peak Ⅱ.When the hydrogen in the environment,the palladium (Pd) film will absorb hydrogen and cause the change of Pd dielectric constant,which leads to the change of EOT characteristics,especially the change of wavelength and transmittance.When the hydrogen concentration increases from 0% to 4%,the transmittance of peak Ⅰ and peak Ⅱ can reach 0.16 and 0.24,respectively.Because of the obvious change of its transmittance intensity,the hydrogen concentration can be monitored and detected by the intensity change of the transmission peak.The proposed sensor is simple to process.The thin Pd film makes the sensor respond quickly,and the EOT effect makes the signal-to-noise ratio (SNR) of the structure high.The photoelectric integration technology can be used for mass production,which has strong practical value.

In the laser radar (Lidar) application scenarios,high power vertical cavity surface emitting laser (VCSEL) attractes more and more attention.However,in the long wavelength band,such as 1 550 nm lasers,there are problems such as unsatisfactory output power and low power conversion efficiency.At present,multi-junction VCSEL is the key technology to obtain high power,high power conversion efficiency (PCE),high optical power density and high slope efficiency (SE).This paper combines 1 550 nm VCSEL and multi-junction technology,which aims to design a high power 1 550 nm device.Finally,1 550 nm multi-junction (1—4) VCSEL devices with different diameters (15 μm,20 μm,30 μm,50 μm,100 μm) are successfully designed and simulated.By comparing the output power of devices with different parameters,it is found that the output power of devices increases with the junctions and diameter,and the maximum power can reach 2 217 mW（4 juctions,100 μm,VCSEL）.And the results show that the output characteristics of 30 μm VCSEL are relatively good:the SE of single junction structure is better than that of 50 μm and 100 μm,is 0.39 W/A,the SE of multi-junction structure is better than those of 15 μm and 20 μm,is 0.89 W/A,is 1.55 W/A and 1.79 W/A respectively.Finally,the PCE and far field distribution of 30 μm VCSEL are analyzed,the paper found that PCE significantly increases with the increase of junctions,up to 40.4% of three junctions,only 1 A driving current,this is 6.2 times the PCE of single junction device.The power output of three junctions VCSEL is up to 1 549 mW,and there is almost no difference in beam divergence between single junction and multi-junction devices,is about 5°.These results provide a good device research basis for the application of 1 550 nm VCSEL as Lidar system and 3D sensing.

Due to the limitation of the receptive field of the shallow convolutional neural network (CNN) model,long distance features cannot be captured,and the spatial-spectral information of the image cannot be fully utilized in hyperspectral image classification,so it is difficult to obtain high precision classification results.To solve these problems,this paper proposes a model based on convolutional neural network and attention mechanism (CNNAM).The model uses coordinate attention (CA) to encode the position of the image channel data,and uses the Transformer module with self-attention mechanism as the core architecture to extract the long distance features to solve the CNN receptive field limitation problem.The overall classification accuracy of the proposed model on Indian Pines and Salinas datasets is 97.63% and 99.34%.Compared with other models,the proposed model shows better classification performance.In addition,the ablation experiments are carried out based on whether CA is combined with CNNAM,and it is proved that CA playes an important role in CNNAM.Experiments show that the combination of traditional CNN and attention mechanism can achieve higher classification accuracy in HSI classification.

The segmentation accuracy of retinal vessels has an important impact on the early diagnosis of ophthalmic diseases and diabetes.Facing the problem of poor segmentation performance of existing methods in microvascular and lesion regions,a segmentation model with enhanced extraction of vessel features is proposed in this paper.The model introduces a multi-scale feature extraction residual module (MFE-residual) and a multi-level residual null convolution layer at the encoding site,which is used to extend the perceptual field,learn multilevel image features and improve the utilization of vessel information by the model; the downsampling and short connection sites are incorporated into the lightweight attention mechanisms and multi-channel attention module to increase the recognition of vessels by the model and reduce the possibility of mis-segmentation.Experiments are conducted based on two publicly available datasets,DRIVE and STARE,to verify the segmentation capability of the improved model.The results show that the accuracy is 0.965 2 and 0.971 5 on both data,and the sensitivity is 0.820 5 and 0.825 6,respectively,which are more advantageous than other algorithms in terms of segmentation performance.

The calculation of a series of biomechanical parameters based on corneal deformation is the data foundation for training early keratoconus classification models,so the accuracy of keratoconus contour segmentation directly affects the accuracy of early keratoconus classification models.In this paper,we propose an unsupervised corneal video segmentation method based on residual networks.A set of anchor points are extracted by uniformly sampling the video frames in the same sequence,which reduces the computational complexity of the network model learning feature representation and improves computational efficiency.At the same time,a regularization branch is designed to transform the original video set for similarity to solve possible degenerate solutions.Compared with existing unsupervised video segmentation tasks,our experimental model uses a small amount of training data but achieves higher segmentation accuracy and computational efficiency.

Aimed at the defect of printed circuit board (PCB) in the process of actual production style variety and small defects,difficult located defect position,and a huge model is difficult to achieve the requirements of real-time detection,and a large number of the depth of the separable convolution layer established lightweight model can't achieve enough accuracy,this paper proposes a PCB defect detection algorithm based on YOLOv5s.Therefore,the original Backbone Conv module and C3 module are replaced by GhostConv.In the Neck part,a new lightweight convolution technology GSConv is introduced to reduce the size of the model while maintaining the accuracy.GSConv achieves an excellent trade-off between the accuracy and speed of the model.Aiming at the problem that many attention modules cannot pay attention to global information while the model is large,a multi-scale lightweight double channel depthwise attention module (DWAM) is proposed to further improve the model accuracy.The experimental results show that,the average mAP of all categories of the improved algorithm is 99.14%,and the GFLOPS of the model is 7.194 G,and the Params is 7.175.The average mAP of the original YOLOv5s is 96.86%,and the GFLOPs is 6.89 G,and the Params is 6.596.Although Params and GFLOPs have increased,they still meet the requirements of lightweight network,and the accuracy is improved by 2.25% compared with YOLOv5s,and the defect recognition accuracy of each category has been improved,which greatly reduces the computation amount and model parameters while ensuring the accuracy.It can meet the demand of industrial testing and production and facilitate mobile terminal deployment.

Phase measuring deflectometry (PMD) is a technique for measuring optical three-dimensional profiles of specular and mirror-like objects.However,when measuring the transparent object,the reflection light from its bottom surface mixes with that created by top surface.This phenomenon produces the superimposed fringe patterns which leads to decrease the measurement accuracy.An effective method based on the principle of muti-frequency and phase-shifting is proposed to solve this problem.By analyzing the relationship between the phase errors and fringe frequencies,and utilizing the characteristics of the modulation curves and introducing an auxiliary function,we can find a special frequency at which the unwrapped continuous phase is equal to the phase of the top surface in theory.Due to the complicacy of a lens,the continuous phase order calculated by temporal phase-unwrapping algorithm frequency-by-frequency may be error.So the order correction method is proposed to solve this matter.And then the screen coordination of the top surface can be obtained by the correction algorithm.The top surface shape can be reconstructed subsequently.Experimental results of measurement of a spectacle lens demonstrate that the proposed method can successfully untangle the phase of the front surface from superimposed reflections.

A text recognition algorithm based on multimodal iteration and correction is proposed to address the problems that scene text recognition is prone to information loss when modeling over long distances and weak characterization for low-resolution text images.The visual model of the algorithm in this paper is a combination of contextual transformer networks for visual recognition (CoTNet),a dynamic convolutional attention module (DCAM),an external attention encoder (EA-Encoder),and a positional attention mechanism.The CoTNet can effectively alleviate the information loss problem arising from long-distance modeling.The DCAM enhances representation by focusing on the essential features while passing the critical components to the EA-Encoder,improving the connection between CoTNet and EA-Encoder.EA-Encoder learns the best distinguishing features on the entire dataset,capturing the most semantic information parts and thus enhancing representation.After the visual model,the text correction and fusion modules obtain the final recognition results.According to the experimental data,the algorithm proposed in this paper performs well on several public scene text datasets,especially on the irregular dataset ICDAR2015 with an accuracy of 85.9%.

In view of the problem of low recognition accuracy caused by 3D palmprint recognition due to noise interference and neglect of adjacent depth information,a 3D palmprint recognition method combining potential texture and surface consistency is proposed.First,the energy local edge binary code (ELEBC) is used to extract the potential texture direction information from the energy map to eliminate the noise.Then,surface consistency is obtained by mean block pattern surface type (MBST).Finally,the principal component analysis (PCA) is used to reduce the data dimension,and the final recognition result is obtained after decision-level fusion.Relevant experiments are carried out using the 3D palmprint database of Hong Kong Polytechnic University,and the results show that it has significant advantages over other novel algorithms.The maximum correct recognition rate can reach 99.71%,and the recognition classification time is kept below 0.5 s.Therefore,the proposed scheme not only has accurate recognition effect,but also can meet the real-time requirements,and its application value is obvious.

The geodesic laser gyroscope is an important tool for measuring the angular velocity of the earth's rotation and provides the basis for the solution of universal time.Environmental changes will affect the geometry of the optical cavity of the geodesic laser gyroscope,which in turn affects the measurement accuracy of the gyroscope.This paper mainly focuses on the error of gyroscope Sagnac frequency difference brought by the tilt of the geodesic laser gyroscope plane.Based on the basic error theory and coordinate system theory,the theoretical model of the influence of tilt on the rotation angular velocity of gyroscope is derived.Allan deviation (AD) method is used to analyze the tilt data from three Laboratories.The raw data is denoised according to the analysis results.Based on this,a tilt error compensation model is developed by using the time series method.It reduces the angular velocity error from tilt by an order of magnitude,bringing the Sagnac output frequency difference closer to the theoretical value.It shows that the noise level of the gyroscope output data can be reduced by constructing a reasonable tilt error compensation model,and it provides a reference for further optimizing the operating environment of the gyroscope and the processing method of related data.

Ultra-fast laser technology is now widely used in the processing of micro and nano structures.The extremely small size produces heat transfer characteristics different from macroscopic conditions and it has gradually become a research focus.An improved dual-phase-lagging (DPL) model is utilized to investigate the two-dimensional thermal conduction of nano-silicon films irradiated by ultra-fast laser.The modified DPL equation reflecting the size effect is solved by integral variation method.The variation of the internal temperature along the thickness direction and the radial direction of the thin film is analyzed.The result shows the temperature increases rapidly in the heated region of the film,and the energy is gradually transferred to the inside of the film in the form of waves.Although both the results based on the improved DPL model and the DPL model can obtain similar thermal wave temperature fields,the temperature distribution in the film obtained by the improved DPL model is relatively flat.Under the same Knudsen number,the temperature obtained by the two-dimensional improved DPL model is higher than that obtained by the one-dimensional model,and the temperature difference becomes larger and larger as time goes by,which implies that the radial heat conduction has an important effect on the temperature distribution inside the film in the two-dimensional heat transfer process.