To meet the high speed,wide bandwidth,and large capacity requirements proposed by the next-generation 6G communication networks for optical communication networks,the optimal design of pumping parameters for Raman fiber amplifier (RFA) has become the focus of current research in optical communication systems.In this paper,using erbium-doped tellurium-based fiber as the transmission medium,a modified marine predator algorithm (MMPA) is proposed to optimize the design of the multi-pump RFA,which effectively achieves the flat optical amplification for each channel in the C+L band.Compared with the existing optimization algorithms for RFA,the MMPA has the characteristics of excellent performance and strong robustness,which can effectively solve the problems of nonlinearity and combinatorial optimization in the Raman coupled wave equation and ensure the high gain and low gain flatness of the RFA.The simulation results show that the average gain of the amplifier is 42.36 dB and the gain flatness is 0.67 dB in the gain spectral width range of 1 530—1 630 nm.

An achromatic lens based on radial gradient index (GRIN) is proposed,which effectively suppresses the chromatic aberration effect of optical systems.Firstly,the radial GRIN lens has the advantages of the lower spherical aberration,better-focusing imaging, and higher optical coupling efficiency.Considering the above advantages,an achromatic optical system of the radial GRIN lens is designed to eliminate the high-order aberration.Moreover,the optical power and spherical aberration of the optical system are optimized by controlling the operands POWR and SPHA in ZEMAX software.Finally,the optical power and dispersion model of the achromatic optical system based on radical GRIN lens is derived and established.At the same time,the achromatic effect comparative experiments for the radial GRIN lens and homogeneous doublet lens are conducted.The experimental results show that the radial GRIN lens can achieve good achromatic function in the wavelength range of 486—656 nm.The achromatic effect is significantly better than the homogeneous doublet lens.The blur spots of the radial GRIN lens are all within the Airy spots,and the lens is well focused.The aberration is basically corrected and the diffraction limit is reached,which meets the imaging requirements of the achromatic lens well.

For rigid dielectric substrate materials,it is difficult to bend conformally.In order to meet the application requirements of broadband,high polarization conversion ratio (PCR) and miniaturization of polarization converter,a dual broadband,high PCR flexible polarization conversion metasurface is proposed in this paper.The designed metasurface is composed of metal pattern layer,flexible dielectric layer and metal ground plate.The simulation results show that the metasurface can rotate the polarization direction of the linear polarization wave by 90° in the frequency bands of 8.52—13.35 GHz and 15.12—19.54 GHz. PCRs are 96.50% and 98.43%,and the corresponding relative bandwidths are 44.21% and 25.51%,respectively.The dual-band polarization conversion performance is manipulated by changing the thickness of flexible dielectric layer.The surface current at the resonant requency is simulated,and the mechanisms of dual-band and high PCR are analyzed. In addition,the dependence of PCR on incident angle and polarization mode of the metasurface are studied.The results show that the proposed metasurface has good dual-band polarization conversion performance and also has polarization insensitivity within 0° to 30°.The proposed dual-band flexible polarization conversion metasurface has great potentials in the application of electromagnetic wave polarization regulation.

In order to realize simultaneous measurement of magnetic field and temperature,a composite sensing structure based on the magnetic-volume effect of magnetic fluid (MF),a Fabry-Perot (FP) cavity and fiber Bragg grating (FBG) cascaded structure is proposed and fabricated.The FP cavity formed by the liquid interface of the MF injected into the hollow-core fiber (HF) and the end face of the single-mode fiber are both sensitive to magnetic field and temperature,and the cascaded FBG is only sensitive to temperature.By simultaneously monitoring the wavelength drift of FP cavity characteristic peak and FBG Bragg reflection peak,the magnetic field and temperature sensitivity coefficient matrix is used to obtain the temperature and magnetic field information simultaneously.An FBG-FP temperature and magnetic field dual parameter sensor with an initial cavity length of 56.1 μm is successfully fabricated,of which the magnetic field and temperature sensitivity of are 16.21 pm/Oe and 9.96 pm/℃,respectively.The sensor has the characteristics of small volume,simple structure and low cost,which can be applied to solve the magnetic field strength and temperature cross-sensitivity problem of conventiaonl fiber-optic magnetic field sensors.

Monocular depth prediction plays a crucial role in 3D scenes understanding and perception.At present,deep learning methods have achieved great performance,however,their performance is dependent on training data.In complex scenes,the assumption of global light invariance based on depth cues is not effective.To this end,we propose a depth prediction method based on cross Bayesian fusion of global and local atmospheric light.Specifically,we use the global and the designed local atmospheric light scattering model to predict the initial depth map respectively.And then we construct a cross Bayesian model to fuse depth maps,in order to better realize the complementarity between the two models.Finally,we optimize the depth map by using the edge filtering mechanism.A series of experiments can prove that the model designed can achieve good results in the depth prediction datasets,and help to improve the accuracy of salient object detection task.

As an effective description of 3D scenes and objects,color point clouds (CPCs) are widely used in many fields such as virtual reality and augmented reality.However,distortions will be introduced to CPC in the process of its collection, compression,transmission and reconstruction,so it is necessary to design an effective assessment method to evaluate the quality of distorted CPC.In this paper,a no-reference quality assessment method is proposed for CPC based on guided modulation.Considering the joint distortion of geometric information and color texture information,the guided modulation is used to combine them to comprehensively consider geometric distortion,color texture distortion and the joint distortion.Combined with the multi-channel characteristic of human eyes,Shearlet transform is used to extract features.Finally,the feature vector composed of all extracted features is inputted into the support vector regression (SVR) model to learn and predict the quality of point cloud.Experimental results show that the proposed method is well consistent with human subjective perception.

Surface electromyography (sEMG) gesture recognition algorithms based on convolutional neural network (CNN) usually convert one-dimensional sEMG to two-dimensional electromyogram (EMG) as the input of CNN.In order to solve the problems such as the lack of instantaneous samples of sEMG and the loss of local timing features caused by converting one-dimensional sEMG to two-dimensional EMG images,a processing method which fuses the multivariate empirical mode decomposition (MEMD) algorithm and the Hilbert space-filling curve is proposed to improve the accuracy of the gesture recognition algorithm.The open-source dataset NinaPro-DB1 is applied.Firstly,the sEMG is decomposed by the MEMD algorithm.Secondly,the decomposed intrinsic mode functions (IMFs) are used as the filled domain (Hilb-IMFs) of the Hilbert curve for mapping them to a two-dimensional EMG image.Finally,DenseNet is chosen as the basic network for gesture recognition.The experimental results show that the proposed method has a performance improvement of about 4% in gesture recognition accuracy compared with traditional signal dimensionalization method,which verifies the effectiveness of the method.

In view of difficulties in extracting effective vibration characteristics from complex vibration phenomena that are occurred when coupled with high pressure and low pressure rotors of a dual-rotor runs high-speed operation,and there aren′t corresponding researches.So,this paper proposes a method that combines multilinear principal component analysis of tensor objects （MPCA) and K-nearest neighbor （KNN） classification and applies it to fault diagnoses of nonlinear dual-rotor systems.Firstly,a nonlinear cracked dual-rotor model and its dynamic equations are created using the concentrated mass method,and the vibration characteristics of high pressure and low pressure rotors are analyzed based on the changes of crack angles.Then,the vibration energy signal and the vibration signal are normalized into color image samples,and the MPCA algorithm is used to compress and extract the fault features.Lastly,the KNN classification algorithm is used to classify the features of different crack angles, and the corresponding classification rates are calculated.The experimental results show that,in the high-speed region of the rotor,MPCA can effectively distinguish different degrees of cracked characteristic signals in the case of low noise,and provides a new detection method for fault diagnoses of nonlinear cracked dual-rotor systems.

Defect detection is an indispensable process in the strip steel production process,and existing inspection methods generally have problems such as low detection accuracy and poor real-time performance.To solve the above problems,a fast defect detection method based on lightweight YOLOv3 is proposed in this paper.MobileNetv2 is used as the backbone network and output with two scales of feature maps,so that the lightweight of the network model is guaranteed;the improved attention module is fused into the feature pyramid network (FPN) and the network is combined with the spatial pyramid pooling (SPP) to improve the learning ability of the algorithm for defects;the K-means mean clustering algorithm is used to obtain a better anchor box,and the complete-intersection over union (CIoU) is used to optimize the loss function to further improve the network performance.The proposed method has a detection speed of 70.8 FPS on the strip steel defect dataset; the number of model parameters is 7.1 MB,which is only 3.02% of YOLOv3.Experiments show that the proposed method can achieve rapid detection of defects while ensuring accuracy,and has good production line deployment capabilities.

At present,the defect detection of key components of electric multiple units (EMUs) has the problems of complex model, high missed detection rate of small targets and low detection efficiency.To solve the existing problems,a defect detection method based on improved YOLOv5 is proposed.On the basis of using generative adversarial network (GAN) to expand the dataset,the YOLOv5m backbone extraction network is changed to the MobileNetV3-large network structure,and the neck 3×3 convolution layer is optimized by using depthwise separable convolution to further reduce the model complexity.Then,the coordinate attention (CA) is introduced into the improved backbone network to capture the location information and channel information of small targets, thereby enhancing the feature expression ability of the network.The non-max suppression (NMS) algorithm is optimized by integrating the position information of the center point of the overlapping detection box to improve the accuracy of the prediction box location.The experimental results on the EMUs defect dataset show that,compared with YOLOv5m,the improved model reduces the amount of parameters by 77%,the amount of computation by 80.9%,the detection time of a single image by 31.7%, and the mean average precision (mAP) can reach 0.804.In addition,the experimental results on the NEU-DET dataset show that the improved model also has a strong generalization ability.

An unsupervised person re-dentification (ReID) method was proposed to solve the insufficient representation of person features and the noisy labels generated by the clustering process in the process of unsupervised ReID,which jointed feature refinement and noise-tolerant comparative learning.Firstly,a non-local channel refinement module (NCRM) was designed to enrich the unlabeled person representation by weighted reinforcement of key feature information,which fused the important features of non-local channel to capture the discriminative representation between classes of unlabeled data to form more discriminative feature descriptors.Secondly,generalized mean (GEM) pooling adaptive adjustment parameters were used to enhance the ability of extract information from different fine-grained regions to accomplish full expression of characteristics.Then,a noise-tolerant dynamic contrastive equalization (DCE) loss was designed for unsupervised associated learning to mitigate the negative impact of noisy label on the network.Finally,the experimental results on two public datasets verify the effectiveness and advancement of the proposed method.The mAPreaches 83.1 % and 71.9 % respectively,which is superior to other advanced methods.

When performing single-photon emission computed tomography (SPECT) bone imaging examination,physicians often reduce radiation damage by reducing radiation dose,resulting in low signal-to-noise ratio and resolution of bone imaging,which seriously affects the diagnosis and automatic detection of lesions.In order to improve the quality of bone imaging,a two-stage SPECT bone imaging noise reduction method based on U-Net is proposed.Firstly,a U-Net noise estimation network is designed to quickly estimate the noise level of each bone image,providing noise prior knowledge for the backbone noise reduction network. Secondly,the backbone noise reduction network also uses U-Net as the basic framework,and combines multi-scale feature fusion and channel-spatial attention mechanism structure to enhance the noise feature extraction ability of the network and predict the noise map.Finally,denoised bone imaging is obtained through residual learning.At the same time,in order to solve the problem that the reconstructed image using the mean square error (MSE) loss function is too smooth,a composite loss function is designed to retain the original detailed information of bone imaging.In the experiments,different noise levels are applied to the bone images in the training set for data augmentation,and a transfer strategy is used to solve the problem of model overfitting.The results show that,compared with the current mainstream algorithms,the proposed noise reduction method can effectively reduce the noise of bone imaging and preserve the detailed features of the lesions. In addition,blind noise reduction can improve the imaging quality of the original bone imaging and improve the automatic segmentation effect of lesions.