A sensor based on Mach-Zehnder interference structure is designed to measure refractive index and temperature simultaneously.The sensor structure is single mode-multi-mode-thin core-multi-mode-single mode.BeamPROP module of RSoft optical simulation software was used to simulate and analyze the optical field inside the sensor structure.The optimal length of the multi-mode fiber (MMF) and the thin core fiber (TCF) was determined.The sensor structure was made and the experimental system was set up to observe the refractive index and temperature response.Dual-parameter measuring simultaneously was realized combined with sensitivity matrix.The experimental results show that the sensitivity of the sensor is -44.944 nm/RIU in the refractive index range of 1.333—1.380 and 0.082 9 nm/℃ in the temperature range of 30—65 ℃.The proposed sensor has the advantages of simple structure, small volume and high sensitivity,which can provide reference for the design of dual-parameter sensor of refractive index and temperature.

The minimum solution calibration method of 2D lidar and camera has shortcomings such as poor accuracy and missing solution.Therefore,a new reliable minimum solution calibration method is proposed in this paper.Firstly,the perspective similar triangle (PST) algorithm is used to solve the perspective three points (P3P) problem constructed by three checkerboards,and the missing solution is obtained with the polynomial extreme points, which improves the anti-noise interference ability of the algorithm.Secondly,an error measurement model based on two types of laser point constraints is proposed to evaluate the error degree of multiple solutions,so as to select the optimal solution from the multiple solutions of the calibration results.Experimental results show that the proposed algorithm can significantly improve the valid solution probability and the calibration accuracy.Under different noise levels,compared with FRANCISCO method and HU method,the probability of valid solution is improved by 5%—20% and 5%—13%,the rotation matrix accuracy is improved by 46%—63% and 41%—47%,the translation vector accuracy is improved by 170—430 mm and 120—170 mm,so the performance is improved obviously.

In order to solve the shortcomings of mutual interference between spectra and slow modeling speed,a multi-band weighted combination model combined with the partial least squares (PLS) method was used for quantitative analysis to improve the measurement accuracy.In this paper,a gas detection system based on spectrum laser absorption spectrum (SCLAS) was built to perform a weighted combination measurement of CO2 in different wavelengths of the near-infrared based on PLS.The absorption spectra of different concentrations of CO2 in the bands of 1425—1443 nm,1565—1587 nm,and 1595—1616 nm were measured at room temperature and pressure.The single-band regression model based on PLS was established,and the coefficient of determination (R2) were 0.9897,0.9486 and 0.9497,respectively.The weights of the single band models are determined based on R2 and the root mean square error (RMSE).A new PLS combination model is established using the multi-band weighted combination model algorithm,and the obtained R2 are 0.9852 and 0.9912,respectively.The experimental results show that the PLS-based weighted combination model can improve the accuracy and stability of CO2 concentration prediction and effectively avoid the slow modeling speed and interference problems.

The output accuracy and stability of the optical fiber voltage sensor can be seriously affected under the disturbance of external ambient temperature.Therefore,this paper proposes a sensor head optimization method using polarization maintaining photonic crystal fiber (PM-PCF) as the sensing fiber to suppress the system′s nonreciprocal error under temperature disturbance.Consideration of the fusion loss,transmission loss between the traditional panda polarization maintaining fiber (panda-PMF) pigtail and PM-PCF,and the fiber fabrication problem under the existing technological conditions,a PM-PCF with the fusion loss as low as 0.14 dB between it and the traditional panda-PMF is optimized.The birefringent temperature dependence characteristics of the two kinds of polarization maintaining fibers are compared and analyzed.And the thermal system error output model of the fiber optic voltage sensor (FOVS) is established.Then,the feasibility of PM-PCF in solving the thermal nonreciprocity problem of reflective inverse piezoelectric FOVS is analyzed and verified.The results show that the thermal nonreciprocity error of the FOVS optimized by PM-PCF is about 10-2 of the traditional PCF voltage sensor.It has significant advantages in improving the output accuracy and stability of the system.At the same time,this low-loss optimization method has important value in fiber sensing and communication application area.

Aiming at the small target characteristic of rail surface crack and the low precision and slow speed of traditional detection methods,we propose an object detection method based on improved YOLOV4 network for cracks on the surface of rails in this paper.Firstly,in order to obtain the larger effective receptive field area of the feature map and improve the detection accuracy,we use the improved receptive field block (RFB) module to replace the spatial pyramid pooling (SPP) structure;Secondly,we use the deep separable convolution structure to replace the common convolution structure in the network model,so that the network is lightweight and the detection speed is improved;At the same time,we use K-means + + algorithm to reacquire the anchor frame, and then change the linear scale of the anchor frame to solve the problem that the original anchor frames are not suitable for small target detection.The results show that the mean average precision (mAP) of the improved YOLOV4 is 84.8%,which is 3.4% higher than that of the original YOLOV4 algorithm;The detection speed (FPS) is 62.39 frame/s,which increases by 4.07 frame/s.

An underwater image enhancement method combining deep learning and multi-scale orientation filter Retinex is proposed to tackle the problems of blurry texture and serious color distortion.Firstly,the land image is degraded by texture and histogram matching method to establish a dataset which simulates the underwater image distortion,and an end-to-end convolutional neural network (CNN) model is built.By using the model,color correction is performed on original underwater images to obtain color-restored underwater images.Then,the multi-scale Retinex (MSR) method is used for the brightness channel of the color restoration images to generate texture-enhanced images.Finally,chrominance of the color-restored images and the texture-enhanced images are fused to eventually get the enhanced underwater images.The proposed method is tested on the simulated underwater image dataset and real underwater images individually.The experimental results show that root mean square error, peak signal-to-noise ratio,CIEDE2000,and underwater image quality measurement are 0.302 0,17.239 2 dB,16.878 4 and 4.960 0 and prevail to five comparison methods. The enhanced underwater images are more real and natural.In conclusion,the proposed method can effectively improve the clarity and contrast while accurately correcting the color distortion of the underwater images.

Aiming at the problem that the current network is difficult to deal with various corrupted pedestrian images and easily loses cross-dimensional information,a pedestrian re-identification (ReID) method based on style normalization and global attention is proposed for corrupted images.The method filters out style changes in the domain by smooth maximum unit-style normalization and restitution (SM-SNR) module in the instance normalization (IN), and at the same time smooth maximum unit (SMU) enables the module to more fully extract pedestrian-related features from the deleted information and restore them to the network,so as to alleviate the style difference caused by corrupted images.In addition,the global attention mechanism (GAM) captures the salient pedestrian features in three dimensions by focusing on the interaction between the channel and the space,reducing the loss of cross-dimensional information.Finally,the recognition ability of the model in recognizing pedestrian corrupted images is effectively improved,and the competitiveness on clean datasets is retained.The experimental results show that the indicators of the algorithm on the corrupted test set has significant advantages compared with the current mainstream algorithms.Among these algorithms,the result of comparison with the 2021 CIL model using the CUHK03 dataset is that:On Corrupted Eval,R-1,mAP and mINP increase by 15.18%,15.75% and 11.65% respectively;on Clean Eval,R-1 and mINP only decrease by 0.24%, 0.75%, and mAP increased by 0.25%.

Fractional vegetation coverage (FVC) is one of the important indicators for grassland degradation evaluation,and real-time,fast and accurate FVC acquisition is the basis for grassland degradation evaluation.This paper proposes a 3D-ResNet18 deep learning coverage extraction method using unmanned aerial vehicle (UAV) hyperspectral remote sensing images as the data source,compares this method with the regression model method and the ResNet18 classical deep learning method,and validates the extraction accuracy.The results show that the proposed 3D-ResNet18 method shows a better extraction effect on desert grassland FVC,with an overall estimation accuracy of 97.56%,which is 8.32%,5.92%,2.20%,2.14% and 1.87% higher compared to NDVI,SAVI,G~~CR~~NDVI,G~~CR~~ SAVI and ResNet18, respectively.The foundation for high-precision and efficient statistics of desert grassland FVC information is laid.

This paper proposes a double-beam complementary fiber Bragg grating (FBG) displacement sensor to achieve the measurement of positive and negative bidirectional displacement.The double "cantilever beam + wedge-shaped slider" structure is adopted,when one cantilever beam is deformed by displacement,the other cantilever beam is not deformed and provides temperature compensation function.The two cantilever beams are in zero bend when the sensor is in zero value measurement point state,and the twin beams are temperature compensated for each other,eliminating temperature effects.The performance test experiments prove that the sensitivity of the sensor is 29.369 pm/mm within a range of ±50 and the measurement repeatability is good.Eight sensors are manufactured for deformation safety monitoring of a city metro,and deformation measurements are carried out in three areas:track bed settlement,track bed annulus and segment annulus.During the long-term monitoring period,the sensor workes stably and the deformation state of the monitored structure is stable,indicating that the sensor has good measurement performance and it is applicable to long-term structural health monitoring.

Aiming at the problems of large measurement error, low accuracy,and poor anti-interference ability of non-cooperative target laser methane telemetry system,a set of methane integrated concentration telemetry simulation systems is established by using simulink software based on tunable diode laser absorption spectroscopy (TDLAS) and wavelength modulation spectroscopy (WMS) technology.The normalized second harmonic detection technology is used to remove environmental interference.The effects of sine wave frequency and modulation coefficient m on the normalized second harmonic signal are simulated and analyzed.The results show that m=2.2 is the best modulation coefficient,and the sine wave frequency has no effect on the normalized second harmonic signal.It is determined that the effective range of methane integral concentration detection is 0—2 000 ppm·m,within this range,the error of system simulation is -3.57%—4.06%.According to the simulation results,the relevant parameters of the telemeter are optimized and experimental research and analysis are carried out. The results show that the measurement error of the system is -4.68%—2.45%.The Allan variance analysis method is used to evaluate the stability and detection limit of the system.The lower detection limit of the system under the 5 s integration time is 37.85 ppm·m,the optimal integration time is 345 s,and the corresponding lower detection limit is 6.27 ppm·m.The research results are of great significance to the research and development of a high-precision laser methane telemetry system.

Aiming at the problems of weak reconstruction performance and robustness in the traditional two-dimensional principal component analysis (2DPCA) algorithm applied to weld surface defect detection,maximizing the projection distance and minimizing the reconstruction error are introduced into the objective function as optimization objectives.And a non-greedy two-dimensional principal component analysis algorithm based on F-norm (non-greedy 2DPCA with F-norm, NG-2DPCA-F) is proposed.This algorithm has good robustness and low reconstruction error.In order to further extract the structural information of the image and obtain the feature matrix with smaller dimension,this paper proposes a bidirectional two-dimensional principal component analysis algorithm based on F-norm (non-greedy bilateral 2DPCA with F-norm,NG-B2DPCA-F).The experiments are carried out with weld surface images with different noise blocks as datasets.The results demonstrate that the proposed algorithm has good robustness in the average reconstruction error,reconstruction image and classification experiments.

The existing detection methods for Ponzi scheme smart contract are mostly based on the operation code features and account features,but using these methods to detect initially deployed contracts is ineffective.Therefore,a Ponzi scheme contract detection method based on deep residual network was proposed.Firstly,by analyzing the characteristics of smart contract,the single word embedding coding algorithm (SWEC) was proposed.Then the contract was recoded by this algorithm.Secondly,critical operation code and its weight were extracted and the weight module of critical operation code (CO) was designed to improve the deep residual network.Finally,the experiments were carried out on public data sets,the experimental results show that the Ponzi scheme contract detection based on deep residual network had 99.7% precision and 99.9% recall.Compared with the existing methods,Ponzi scheme contract was detected more accurately.

Aiming at the differences in tumor status presented by different modalities of MR brain tumor images and the limitations of feature extraction by convolutional neural networks (CNNs),a method of brain tumor image segmentation based on multimodal fusion is proposed.The segmentation model is based on the U-net network,which innovate a multimodal image fusion approach to enhance the feature extraction capability,while a channel cross transformer (CCT) module is introduced instead of the jump connection structure in the U-net to further the deep and shallow feature disparity and spatial dependency,fusing the multi-scale features effectively and enhancing the tumor segmentation capability.The results of multi-objective segmentation are verified on the BraTS dataset.Quantitative analysis and comparison of frontier network segmentation results shows that the proposed method has good segmentation performance.The Dice coefficients of three tumor regions are 80%,74% and 71% respectively.