Based on the localized surface plasmon resonance (LSPR) effect of the large-sized gold nanoshells (AuNS),the principle and characteristics of fiber Bragg grating (FBG) LSPR refractive index (RI) sensor was studied.The physical mechanism of the RI sensitivity enhancement of the AuNS for the sensor was studied by the finite difference time domain method.The LSPR RI sensor based on the etched fiber Bragg grating (eFBG) was constructed,and the response spectra under different surrounding refractive index (SRI) environments were obtained.The experimental results show that the intensity-based RI sensitivity of the eFBG-LSPR sensor was －72.33 dB/RIU,which was 61.3 times higher than that of eFBG,verifying the validation of the theoretical analysis.The proposed sensor has good application prospects in the fields of biochemical sensing and environmental monitoring.

A medium and high frequency fiber Bragg grating (FBG) acceleration detector based on lever amplification is proposed.The acceleration sensitivity and amplitude-frequency characteristics of the detector are theoretically analyzed and experimentally studied.The experimental results show that the change of central wavelength of the detector has a good linear relationship with the change of external acceleration.The natural frequency of the detector is 612 Hz and the flat area range is 20—250 Hz.The linearity is 0.995,the acceleration sensitivity is 106.7 pm/g,and the cross sensitivity is less than 4.9%.It has good anti-interference ability.

Metamaterials and metasurfaces are widely used in electromagnetic wave control due to their special properties that natural materials do not possess.Aiming at the application requirements of broadband,high-efficiency and miniaturization of polarization converters,a dual-polarization converter based on broadband and high-efficiency reflective metasurface is designed.Based on the principle of polarization control of electromagnetic wave,a periodic array metasurface of circularly truncated square patch is designed to achieve efficient linear polarization and circular polarization conversion in a single broadband on the basis of the two-angle truncated square patch.The results show that the polarization conversion ratio (PCR) of the polarization converter exceeds 90% in the frequency band range from 6.850 GHz to 11.050 GHz,and performance can be maintained within the range of incident angle less than or equal to 45°,covering the whole X-band to realize the linear polarization conversion with high efficiency and wide incident angle.The axial ratio (AR) is less than 3 dB in the frequency range from 6.225 GHz to 6.402 GHz and range from 12.562 GHz to 13.472 GHz,which realizes high-efficiency circular polarization conversion.In addition,the metasurface has the advantages of simple structure,wide incident angle and convenient processing,which can meet the application needs in the fields of microwave communication,microscopic imaging and so on.

A dual-parameter optical fiber sensor with a ball-fine core-ball structure is designed,and then the refractive index and axial strain characteristics of the optical fiber sensor are compared before and after the optical fiber is tapered.Through experimental and theoretical analysis,it is known that,after the fine core between the two-ball structure is drawn,it has a greater impact on the strain of the sensor and the sensitivity of the refractive index in the tapered area.The experimental structure is simple,convenient to manufacture and easy to repeat.After the tapering of the dual-parameter optical fiber sensor fabricated by this structure,the refractive index sensitivity of the tapered area is more than twice that of before the tapering,reaching 50.53 nm/RIU,and the axial strain sensitivity is also twice as high as that before the tapering.At about 2.908 pm/με,the experimental results are basically in line with the theoretical derivation.

As the absorption and scattering of light in underwater propagation,problems such as fuzzy,low contrast,color deviation and uneven illumination appear in underwater images taken.In view of the above problems,an improved underwater enhancement algorithm based on Gamma correction and multi-scale fusion is proposed.Firstly,Red and Blue channels are compensated based on Green channel,and the correction image is obtained by Gray World algorithm after histogram stretching of RGB three channels.Then,the color corrected image is corrected for uneven illumination by using the improved Gamma function to obtain the uneven illumination corrected image,and the illumination uniformity image is normalized.Then,the contrast of the image after uneven illumination correction is improved by contrast limited adaptive histogram equalization (CLAHE)algorithm to obtain the contrast improved image.Finally,the multi-scale fusion algorithm is used to fuse the above three images to obtain the enhanced image.Experimental results show that the proposed algorithm has good processing effects on images in different underwater environments,and the image quality evaluation index is significantly improved.

In remote sensing image scene classification,a classification algorithm based on convolutional neural network (CNN) has the dependence on training data,and the classification effect is poor in the absence of training data,and a classification algorithm based on transfer learning is proposed.Firstly,the existing pre-training model of multiple CNN is selected,and the model is fine-tuned by using the advantages of transfer learning to extract the different high-level features of the image,then,the fusion of the image′s many high-level features makes the feature information more abundant,and finally,the merged high-level features are input into the remote sensing image classifier based on logical regression,and the classification results of remote sensing images are obtained.Experiments are carried out in remote sensing data sets of UCMerced~~LandUse,and the existing algorithms are compared and analyzed,and the proposed algorithms are significantly improved in three evaluation indicators.By analyzing the experimental results,it is shown that the algorithm can achieve 92.01% classification accuracy and 91.61% Kappa coefficient under only 10% of the training data.

Aiming at the current lack of multi-scale detail information in the original image based on wavelet transform image fusion enhancement algorithm,an improved image enhancement algorithm combining multi-scale wavelet transform and depth residual selection is proposed.After the original image is decomposed and extracted by wavelet transform to obtain its multi-level decomposition coefficients,different rules are used to reconstruct different levels of wavelet coefficients.At the same time,the idea of deep residual algorithm is introduced to make residuals for subband coefficients.For the high frequency subband coefficients,the proposed algorithm will calculate the coefficients of the subband residuals and the coefficients of the gradient feature fusion method,and select the maximum value of the two for fusion enhancement,while for the low frequency subband coefficients,the algorithm uses the method of averaging the gradient feature fusion enhancement coefficient and the subband residual coefficient for fusion.The algorithm is verified through experiments on MATLAB platform,compared with the comparison method,the peak signal-to-noise ratio has been improved,and the root mean square error has also been reduced,and the structural similarity has been improved.The experimental results show that the method can enhance the multi-scale detail information of the image,improve the signal-to-noise ratio of the image,and has a better image enhancement effect.

In order to overcome the limitations of existing headdress segmentation methods in portrait Thangka images and the high cost of fully supervised semantic segmentation with pixel level annotation,we propose a weakly supervised semantic segmentation method with frame level annotation.Firstly,the proposed method uses Canny algorithm to obtain the rough edge of headdress.Secondly,the improved EDLines algorithm is used to extract the key points of headwear.Finally,we use Polygons processing to generate feature masks according to the characteristics of headwear.Experiments show that the mean intersection over union,mean intersection over union (mIoU) index of this method is 7.56% higher than semantic segmentation instance (SDI) and is 6.11% higher than weakly-supervised instance segmentation~~bounding box prior (WSIS~~BBTP).It is effective.

In view of practical problems for the difficulty of obtaining large number of fault samples in the field of the intelligent fault diagnosis and problems of the real-time and so on for the need of a complete retraining period in the new fault categories,the new incremental 2D principal component analysis (I2DPCA) method of fault diagnoses is applied in the nonlinear cracked rotor system.Firstly,dynamics equations of a horizontally supported nonlinear rotor system with transverse cracks are established to investigate vibration varying characteristics of the system with different crack depths and mass eccentricity.Secondly,vibration signals in the time domain are normalized to image samples,and low dimension fault features with high discrimination are extracted by the I2DPCA algorithm.Based on the above treatment,the k-nearest neighbor (KNN) classification algorithm is used to calculate the recognition rate.The results of numerical simulations and related experiments show that the fault diagnosis method based on the I2DPCA can effectively distinguish signals of different fault conditions in high rotating speed zone and small samples situation,and provide a new detection strategy for the early diagnosis of cracked rotor systems.

A strategy for locating and tracking pipeline inspection gauge (PIG) in the pipeline via phase-sensitive optical time domain reflectometry (φ-OTDR) and you only look once v5 (YOLOv5) target detection algorithm is proposed.Vibration will be generated when the cups at both ends of the PIG collide with the welding seam of the pipeline.The φ〖ＷＴＢＺ〗-OTDR technology can be used to collect the vibration signal and present an "inverted V" feature that distinguishes it from other background noises on the space-time map.A training set and a test set are constructed by obtaining a large number of space-time maps containing "inverted V" features.The training set is used to train the YOLOv5 network model,and the test set is used to test the trained YOLOv5 network.The trained model is proved to be able to accurately capture the "inverted-V" feature in the space-time map,thereby inverting the real-time position and path of the PIG.The distributed optical fiber sensor and neural network algorithm are combined to further improve the convenience and accuracy of PIG positioning and tracking,which is conducive to the realization of PIG online and automatic tracking.

Three-dimensional (3D) emission computerized tomography (ECT) is a simple,efficient and accurate technology for 3D imaging and measuring of combustion field,in which the calculation accuracy of weight matrix determines the accuracy and quality of tomographic reconstruction.In this paper,a weight matrix calculation algorithm based on ray tracing of high-density subgrids is studied.The measured zone is divided into high-density subgrids,and ray tracing is performed according to the camera imaging model to determine the weight factors of discrete grid with projection pixels.Numerical simulation and reconstruction experiment for combustion flame prove that the algorithm has high accuracy and computational efficiency.The research has important theoretical reference for the practical application of 3D ECT.

The 65TeO2-15ZnO-10Na2O-10WO3 series tellurite doped with 0.2 mol% Er2O3,1 mol% Yb2O3,0.1 mol% Tm2O3,and x mol% Pr6O11 (x〖ＷＴＢＺ〗=0.25,0.3,0.35 and 0.4) are prepared by melt quenching technology.The glass sample is characterized by X-ray diffraction (XRD) and differential scanning calorimetry (DSC) curve to characterize the crystallization resistance and thermal stability.The results show that the glass sample has good crystallization resistance,the difference between crystallization temperature and transition temperature is 140 ℃,and has good thermal stability.The absorption spectrum results show that the Er3+/Yb3+/Tm3+/Pr3+ codoped tellurite glass has a strong absorption peak at 980 nm,so the glass sample can be excited by a 980 nm pump source.In the near-infrared range of 1 200—2 000 nm,the glass sample has three emission peaks of 1.35 μm,1.53 μm and 1.8 μm,and the full width at half maxima (FWHM) of the three emission peaks is greater than 100 nm,which covers the four wavebands of E,S,C and C+L for optical signal transmission,and greatly increased the amplification bandwidth of erbium doped fiber amplifier (EDFA).

Aiming at the problems of high gray value of intervertebral disc image and uneven imaging image,it is difficult to capture spatial information and lack of semantic information.Taking the recognition of lumbar intervertebral disc in T2 sagittal position by magnetic resonance as the object,this paper proposes an intervertebral disc detection algorithm,TCA~~CenterNet (top coordinate attention CenterNet),based on improved CenterNet model,firstly,the coordinate attention (CA) mechanism is added to the top of the backbone feature extraction network to strengthen the network′s attention to the intervertebral disc and enhance the sensitivity of the model to the target position;Secondly,deep and shallow feature fusion is used to enhance the ability of CenterNet to extract effective features,and the generalization performance of the model is improved through data enhancement.The experimental results show that the final mean average precision (mAP) 〖ＷＴＢＺ〗of the model is 81.15% and the average frame rate is 14.2 frame/s.Compared with other comparison algorithms,the improved algorithm has better accuracy and robustness.

In this paper,aiming at the robustness and accuracy of classifying a large number of unlabeled samples with only a small number of labeled samples,we propose an improved semi-supervised generative adversarial networks (SGAN) method for breast cancer image classification.This method uses Softmax function instead of Sigmoid function to realize multi-classification in the output layer.Firstly,the random vector is input into the generation network to generate pseudo samples and be labeled as pseudo sample class for training.Then the real labeled samples,real unlabeled samples and pseudo samples are input into the discrimination network and output as different kinds of probability values.Then the semi-supervised training method is used to update the parameters by back propagation.Finally,the classification of breast cancer pathological images is realized.The number of labeled samples is 25,50,100 and 200 respectively.The final accuracy rate is 95.5%.The experimental results show that the accuracy rate of this algorithm has good robustness when the labeled samples are limited.Compared with the classification methods such as convolution neural networks and transfer learning (TL),the accuracy of this algorithm is significantly improved.

Electroencephalography (EEG) has become the most widely used tool for doctors to diagnose nervous system diseases.It is of great significance to realize automatic recognition of epileptic EEG signals of the clinical diagnosis and treatment of epilepsy patients.In order to improve the recognition precision,this paper proposes a kind of automatic recognition model based on multi-scale convolution feature fusion of epileptic EEG signals.First of all,the multi-scale convolution features fusion method is used to extract more granularity data and solve the problem of information complementation at different levels in convolutional neural network (CNN).Then,the temporal features are extracted by long short-term memory network (LSTM),and the final recognition results are given by softmax classifier.The experiment is completed on the Epilepsy Research Center at the University of Bonn experiment data set.The proposed model is compared with CNN-LSTM model, the single LSTM model,et al.The experimental results show that the recognition precision of the proposed method is higher than other method,the average accuracy is 99.19%.The model could recognize epileptic EEG category,has excellent recognition performance and clinical application potential.