The length of interaction between light and material is limited in the fiber optics salinity sensors based on optical passive scheme,and a separate design of fluid channel is usually required,which increases the detection complexity.Therefore, an optical fiber laser salinity sensor based on thin-walled microtube and ring cavity is proposed.The thin-walled microtube cavity is used as the salinity sensitive unit and optical filter, and is connected with the erbium-doped fiber to form a fiber ring cavity.The whispering gallery mode is excited in the thin-walled microtube cavity through the coupling of tapered optical fibers.The central wavelength is linearly corresponding to the salinity,which realizes the measurement of salinity of solution in the thin-walled microtube cavity.Compared with the transmission spectrum of thin-walled microtube without gain,laser can provide sensing signal with higher signal-to-noise ratio and narrower linewidth.Experimental results show that the sensitivity and the minimum detection limit of the proposed fiber laser salinity sensor are 36.5 pm/‰ and 0.485 5‰,respectively, during the salinity range of 0‰—45‰,the linearity is 0.999 24, and the salinity measurement error is less than ±0.751 3‰,and does not require additional fluid channel, which is expected to become a useful tool for ocean salinity detection.

In order to take into account of higher temperature sensing sensitivity and larger measurement range,we proposed atemperature sensor based on the vernier-effect of cascaded Fabry-Perot interferometer (FPI) and Mach-Zehnder interferometer (MZI),and carried out by experiment.The FPI used in the experiment is composed of two sections of single-mode fiber and a quartz wave plate coated at both ends.Its structure is stable and not affected by the vibration of the incubator,so it is used as a temperature sensing element.MZI filter is made up of two 3 dB couplers.By controlling the length of the two arms,the free spectral range (FSR) is close to the FSR of FPI， so that the temperature sensing sensitivity can be amplified in a cascaded manner based on the vernier-effect.The experimental results show that the temperature sensitivity of the cascade interferometer is 72.4 pm/℃ under the temperature change of 20 ℃—70 ℃.Compared with a single FPI (8.72 pm/℃),this structure magnifies the sensitivity of temperature sensing by 8.3 times and has a larger measurement range.The experimental results are consistent with the theory.

The current feature point matching method for simultaneous localization and mapping (SLAM) is generally affected by the change of perspective,which makes the matching of feature points difficult,which in turn deteriorates the accuracy of feature point matching, and ultimately influences the construction of three-dimensional (3D) point cloud maps and the estimation accuracy of camera motion pose.For this reason,this paper presents an attention based on feature point matching network for SLAM.The innovation of this article is that compared with the existing SLAM and we replaces the feature point matching method of the visual odometer module in SLAM with an attention based on feature point matching network for feature point matching.And we make a new combination of feature point extraction and matching with the traditional feature point extraction method to form a new visual odometer and a new SLAM.Firstly,we encode the extracted feature points and descriptor vectors and we learn through the graph attention neural network to obtain matching descriptors.Then we create a score matrix based on the matching descriptors and use the optimal transmission algorithm to solve the optimal score matrix.In the end we calculate the optimal matching point pair and complete camera positioning,mapping,and loop detection based on the optimal matching point pairs.The experimental results show that when the viewing angle is unstable,an attention based on feature point matching network for SLAM can significantly improve the accuracy of the camera′s trajectory and the estimation accuracy of camera motion pose.

The protein content of milk will affect the quality of milk.The feasibility of predicting the protein content of milk is studied by using the spectral feature information of hyperspectral image.In this paper,a prediction modeling method (CARS-SPA-BP) based on competitive adaptive reweighted sampling (CARS) and successive projections algorithm (SPA) combined with multilayer feedforward neural network (back propagation,BP) is proposed.In the experiment,250 groups of hyperspectral data of five kinds of milk were collected by the visible/near infrared hyperspectral imaging system.Through the experimental comparison,the standardized method was used to preprocess the obtained absorption spectrum,and then the CARS combined with SPA was used to select the characteristic wavelength,18 characteristic wavelengths are obtained.Through experiments,the determination coefficients R2c and R2ｐ of training set and test set of CARS-SPA-BP model reach 0.971 and 0.968 respectively,and the root mean square error of calibration (RMSEC) and root mean square error of prediction (RMSEP) reach 0.033 and 0.034,respectively.It is found that the prediction results of multilayer back propagation (BP) neural network model based on CARS and SPA are not significantly lower than that of full wavelength model,Therefore,the CARS combined with SPA for wavelength screening and BP neural network can basically complete the prediction of milk protein content.In order to verify the prediction ability of CARS-SPA-BP model,the traditional partial least squares regression (PLSR) is used to model under the same data environment.The experimental results show that CARS-SPA-BP has significantly improved R2p and RMSEP compared with PLSR.The results show that CARS-SPA-BP can make full use of the spectral characteristics of milk to achieve high-precision detection of milk protein content.

Aiming at the problems of uneven illumination of underwater image caused by supplementary illumination of deep sea and night waters,image noise,low contrast and color deviation caused by suspended particles in water,an underwater image enhancement method with nonuniform illumination is proposed.Firstly,the noise of underwater image is removed by Gaussian filter.Secondly,the maximum inter class variance (OTSU) method is used to segment the light and dark region mask of the image, divide the brightness map into light and dark regions, and homomorphic filtering is performed on the dark regions to correct the shadows caused by uneven illumination. Then,the new brightness map is obtained by fusing the light and dark areas with weighted average method,and the color image is synthesized again.Finally,the underwater image is processed with contrast limited adaptive histogram equalization (CLAHE),and gray world to enhance contrast and color distortion correction,get enhanced underwater image.The experimental consequences express that the proposed method can valid ameliorate the uneven illumination problem,remove the underwater image noise and enhance the image contrast,which is conducive to the subsequent target detection,tracking and other tasks.

In order to fully extract the complementary information between source images and improve the shortcomings of traditional image fusion algorithms in brightness maintenance,energy preservation and edge information preservation,a medical image fusion algorithm based on pulse coupled neural network (PCNN) image segmentation is proposed in this paper.The algorithm combines non-subsampled shearlet transform (NSST) and PCNN.Firstly,the source image with large standard deviation is selected as the segmented image and the source image with small standard deviation is used as the reference image.The source image is decomposed by NSST to obtain the low-frequency subband coefficients and high-frequency subband coefficients of the source image; In the low-frequency fusion,the parameter adaptive PCNN is used to segment the low-frequency subband of the segmented image,and the fused low-frequency subband coefficients are obtained according to the segmentation results; In high-frequency fusion,the product of regional energy and Laplace energy is used as the judgment function to obtain the fusion high-frequency subband coefficient; The fused image is obtained by inverse NSST transform.Finally,using the algorithm proposed in this paper,the fusion simulation of three groups of computerized tomography/magnetic resonance imaging (CT/MRI) images such as brain atrophy, acute stroke and hypertensive encephalopathy is carried out,and the simulation results are compared with the fusion images of five proposed algorithms in the international journal after 2018.The results show that the image obtained by using the fusion algorithm proposed in this paper effectively enhances the information complementarity between different modes, maintains the same brightness between the fused image and the source image,and retains the edge information of the low brightness part of the source image,which is more in line with the human visual characteristics and has higher objective evaluation indexes.

Aiming at the problems of low efficiency and accuracy in detecting internal defects of automobile wheels under traditional methods,and the accuracy is not up to industry standards,this paper proposes a method for segmentation of image defects in X-ray images of wheels based on improved U-Net neural network,AW-Net.This method cascades two U-shaped networks to extract image features in a three-level jump connection mode; at the same time,the attention mechanism is integrated in the jump connection process to solve the problem that the change of small targets is easy to be missed,and passes Experiments verify that a combination of multiple activation functions is used to achieve more accurate semantic segmentation of X-ray images of the hub, increase the fitting ability of the network, and improve the robustness of the network.The experimental results show that the improved algorithm has a false detection rate of 2.73%, a leakage rate of 0 and a recognition rate of more than 93% for the internal defects of automotive wheels in the data set constructed in this paper, and its segmentation accuracy is higher than that of traditional image segmentation networks, such as fully convolutional network (FCN) and U-Net, and the edge segmentation of this method is flatter and meets the needs of nondestructive detection of internal defects of modern wheels.

Wheels are one of the key components of the running part of the train.Defects in its tread will directly affect the safety of train operation.In order to accurately identify different types of wheel tread defects during inspection,a texture feature extraction method based on gray-gradient co-occurrence matrix is proposed.After analyzing the gray and gradient features of the tread image,image texture feature vector is extracted according to the gray-gradient co-occurrence matrix.Then combined with the K-means clustering optimization algorithm to cluster the characteristics of tread defects,thereby classifying the types of tread defects,and displaying the classification results with visual data.The experimental results show that the accuracy of classifying and identifying different types of wheel tread defects is over 96% by using the above-mentioned algorithm.

This article in view of the different scale changes lead to the detection rate of weld defect effect is not ideal,is proposed based on a faster region-based convolutional neural network (Faster R-CNN) of weld defect detection algorithm of improved algorithm using convolution expansion characteristics under the different expansion rate,combination of convolution kernels under different receptive field more comprehensive to extract the feature information of different scales,to improve the target detection accuracy at the same time,the deep separable convolution is used to compress the model to improve the detection speed.The experiment shows that the improved network can improve the detection speed while ensuring the operation speed and the detection accuracy can reach 72%.

For the traditional human non-destructive testing and recognition methods,there are problems of insufficient accuracy and reliability as well as few kinds of defects to detect.To solve them,this paper proposes an aerospace composite material defect detection algorithm incorporating frequency domain features.The algorithm can be divided into three main steps.Firstly,the input information of the frequency domain of the image is added to the feature extraction backbone network which is used to improve the feature extraction effect of defect images.Secondly,a module of informational concentration is proposed in order to improve the visualization capability and detective accuracy of defects,and on the basis of mask region-based convolutional neural network (Mask R-CNN),the segmentation mask loss function is improved.Finally,combined with the cascaded neural network structure of cascade region-based convolutional neural network (Cascade R-CNN),a new instance segmentation network is formed.In addition,the proposed instance segmentation network was experimentally verified in the aerospace composite material defect X-ray image data set,and the average accuracy of the model detection reached 95.3%,which achieved better results than other instance segmentation algorithms,such as Mask R-CNN and cascade mask region-based convolutional neural network (Cascade Mask R-CNN).The research result has been applied to the intelligent detection of several common aerospace composite material defects in actual industrial production.

In order to improve the spectral characteristics of laser-induced breakdown spectroscopy (LIBS),a LIBS detection system under the constraint of planar mirrors was built.The planar mirrors were placed on both sides of the sample.Planar mirror spacing of 7 mm,9 mm,11 mm and 13 mm were selected to carry out the experiment.It was obtained that the plasma radiation intensity decreases with the increase of the plane mirror spacing.The effects of different plane mirror spacing on the plasma properties of Fe,Al and Pb elements in soil samples were studied.The experimental results show that: compared with no plane mirror constraint,when the plane mirror spacing is 7 mm,the signal-to-noise ratio of the three elements Fe I,Al I,and Pb I in the sample is increased by 29.9%,39.4%,and 31.0%,respectively.The plasma temperature was increased by 484.54 K,and the plasma electron density was increased by 2.41×1015 cm－3.The detection limit was reduced from 86.9 mg/kg to 51.2 mg/kg and the relative standard deviation (RSD) was reduced from 7.8% to 4.6%.It can be seen that the use of planar mirrors is a simple and effective way to improve the spectral characteristics of laser.

The accumulation of micro-damage such as matrix crack and fiber fracture in carbon fiber reinforced polymer (CFRP) under cyclic loading will seriously affect the mechanical properties of CFRP.The size of micro-damage is small and the locations of micro-damage are scattered,which are difficult to be accurately identified by traditional nondestructive testing methods.Laser ultrasonic detection technology has the advantages of non-contact,fast detection speed,wide measurement range,et al.Especially combined with the advantages of laser long distance excitation and large angle incidence,it has a great potential in the damage detection of large size and curved structure materials.Based on the thermoelastic effect of laser,the generation process and propagation characteristics of ultrasonic wave in CFRP are systematically studied on the basis of analyzing the distribution of temperature,stress and displacement field after laser is applied to CFRP.Through the analysis and comparison of ultrasonic echo signals with defects in different places in CFRP,the corresponding relationship between defect position and echo signal characteristics is obtained,so as to realize the reverse performance of the key information of defect position in CFRP from the echo signal characteristics.

In order to improve the amplification performance of the fiber Raman amplifier,a photonic crystal fiber with high Raman gain coefficient and larger negative dispersion was designed.The full vector finite element analysis method is used to numerically analyze the photonic crystal fiber with a regular octagonal cladding,explore the effects of changes in air pore structure and core doped germanium concentration on effective mode field area and Raman gain coefficient.Finally,a photonic crystal fiber with small mode field area,high Raman gain coefficient and large negative dispersion is obtained.The research results show that the germanium-doped photonic crystal fiber structure with a cladding air hole diameter of 1 μm and a hole spacing of 1.2 μm at a pump wavelength of 1 450 nm and a signal wavelength of 1 550 nm can obtain a high Raman gain coefficient of 19.97 W－1·km－1.At the same time,a large negative dispersion of －327.6 ps/(nm·km) can be obtained at 1 550 nm.The comprehensive characteristics of this fiber are of great significance to the improvement of the amplification performance of the Raman amplifier.

Due to the bit correlation of high-order modulation,polarization code can not effectively polarize the channel in optical communication system.In this paper,a polarization interleaver for polarization multiplexing optical communication system is proposed.The proposed interleaver uses a unified encoder to encode the information bits of the system,and then interleaves the coded sequence,which breaks the bit correlation of high-order modulation,reduces the impact on the channel polarization of the polarization code,and improves the performance of the communication system.At the same time,the 2D and 4D modulation and coding schemes are compared and analyzed,and the scaling factor is added in the decoding symbol metric calculation to improve the bit error rate (BER) performance of the system.Simulation results show that compared with the traditional bit interleaved coded modulation (BICM),the BER performance of the polarization interleaver is improved by about one order of magnitude when the channel is additive white Gaussian noise (AWGN),and the BER performance of the system is improved by about 0.5 dB when the scaling factor is added.In DP-16QAM coherent optical communication system,the performance of 4D polarization interleaving scheme is improved by about 1.3 dB compared with the traditional BICM scheme when the BER is 10－4.Moreover,the tolerance of 4D modulation scheme to polarization mode dispersion (PMD) is better than that of 2D modulation scheme under different PMD.

Aiming at the problems of dispersion regulation and structural parameter optimization in dual-ring resonant dispersion compensation microstructure fiber, a dispersion characteristic prediction method of orbital angular momentum mode based on neural network is proposed in this paper.By designing a multi-layer neural network model and adjusting the number of hidden layers,neurons and super parameters of the neural network,accurate dispersion prediction results are obtained.This proposed method establishes the relationship between fiber structures and dispersion characteristics.Compared with the traditional optical simulation method,this method can calculate the dispersion characteristics for different fiber structures more quickly and efficiently,which could provide a new approach for the design and optimization of fiber structures.