The measurement of pressure and temperature in oil well is an important part of intelligent completion system.In order to realize single point and double parameter measurement of pressure and temperature in downhole environment,and the maximum pressure can reach 30 MPa and the temperature can reach 120 ℃,a pressure and temperature sensor with π phase shift grating in center-shaped side hole fiber (SHF) is proposed in this paper.Firstly,the theoretical derivation of simultaneous measurement of pressure and temperature with side hole π phase shift fiber grating is carried out to prove the theoretical feasibility of simultaneous measurement of pressure and temperature.Secondly,based on the theory,the SHF is simulated and optimized,and the fracture limit is introduced.The optimal size of the SHF is obtained as 27 μm from the center of the hole and 21.5 μm from the radius of the hole.The results show that the SHF can realize the measurement of single point and double parameters,and the introduction of π phase shift grating improves the measurement resolution.The larger the hole radius of the SHF,the closer the hole to the core is,the higher the pressure sensitivity,but considering the fiber fracture,the larger the radius is not the better.This research has guiding significance for the design of edge hole grating pressure and temperature sensor in a specific pressure measurement range.

To narrow the single-channel bandwidth of the comb filter and increase the utilization rate of limited frequency band resources,a surface plasmon polaritons (SPPs) interference comb filter was proposed in this paper.A significant increase in pass band transmittance ,narrowing of the single-channel bandwidth and channel numbers was achieved by this filter.To explore the physical mechanism and optimize the filter structure,the effect of the structural parameters of the comb filter on the transmission spectrum was numerically simulated by finite difference time domain (FDTD) method.Experiment result shows that the single-channel ultra-narrow bandwidth of the comb transmission spectra results from dual interference of SPPs and graphene microcavity.As a result,the comb transmission spectra are mainly concentrated in the near-infrared bands (1—1.24 μm).By optimizing the structure parameters,the single-channel bandwidth can reach 4.2 nm,and the number of channels is 18.Additionally,the adjacent channel spacing,single-channel bandwidth and channel numbers can be selected by adjusting the width of the silica top layer.Particularly the comb channel is excited by its material coupling resonance in the filter,which means that it is simple and more suitable for on-chip integration.

In order to solve the problems of single prior information and incomplete salient object detection in the traditional algorithm,a new salient object detection algorithm based on background prior and center prior is proposed.Firstly,the edge nodes of the image are used as background seeds to manifold ranking,and the rough foreground area is obtained; Then,Harris corner detection and clustering are used to detect the prior significance of the center and capture the significant information of the center; Finally,the final saliency map is obtained by fusing the center saliency on the preliminary saliency map.In this paper,the comprehensive index,precision recall curve,F-measure value and mean absolute error (MAE) value of average absolute error are evaluated experimentally.The experimental results on the open data sets MSRA-10K and ECSSD show that compared with 10 mainstream algorithms,the algorithm in this paper has good performance in different evaluation indicators,and can accurately highlight significant targets andimprove the effect of background suppression.

In order to solve the problems of low accuracy,high computational cost and failure to make full use of space spectrum information of traditional hyperspectral image classification methods,a hyperspectral image classification method based on multi-dimensional parallel convolution neural network (3D-2D-1D PCNN) is proposed in this paper.Firstly,the algorithm uses different dimensions of convolutional neural network (CNN) to extract the spatial spectral features,spatial features and spectral features of hyperspectral image information.Then,the same parallel convolution layer is used to fuse the combined spatial spectral features,spatial features and spectral features.Finally,hyperspectral image information is accurately classified by linear classifier.The proposed method can not only extract the deeper spatial and spectral feature information in hyperspectral images,but also fuse the features of different dimensions of spectral images to reduce the computational cost.Comparative experiments are carried out on Indian Pines,Pavia Center and Pavia University data sets.The results show that the proposed algorithm obtains the optimal results,and the classification accuracy reaches 99.210%,99.755% and 99.770% respectively.

Due to the absorption and scattering of light in the water environment,the collected underwater image has the defects of color distortion,blurred details,uneven illumination and poor contrast.Therefore,an underwater image enhancement method based on brightness correction and multi-space transformation is proposed to resolve this defect.Firstly,color balance algorithm is used to correct the color of degraded images.Secondly,the color corrected image is converted from red-green-blue (RGB) space to hue-saturation-value (HSV) space,the V channel is processed by the constructed two-dimensional function,and then transferred back to RGB space.Finally,the weighted fusion of color balance image and brightness correction image are carried out,and the fusion image is converted from RGB space to LAB space.The L channel is processed by the contrast-limited adaptive histogram equalization algorithm,and then transferred back to RGB space to obtain the final enhanced image.In order to verify the effectiveness of the algorithm in this paper,subjective visual effects and three objective indicators are used for verification.The results show that this algorithm can effectively correct image brightness and improve image clarity.

Aming at the difficulty of image matching caused by different imaging modes,time phases and resolutions of heterogeneous remote sensing images,a remote sensing image matching algorithm based on cycle generative adversarial strategy is proposed.A cross-data domain image feature migration cycle generative adversarial network (GAN) was constructed,a SmoothL1 loss function was designed to optimize the network,the accuracy of remote sensing image feature extraction was improved,and based on the result of image feature migration,triple margin ranking loss function (TMRL) was established to reduce remote sensing image mismatched points,to achieve accurate matching of heterogeneous remote sensing images.The test results show that the method in this paper improves the average accuracy of heterogeneous remote sensing image matching by 33.51%,and has a better remote sensing image matching effect than the cross modality matching net (CMM-Net) method.In addition,this method not require the annotation information of the target domain image,and the matching time is shortened by 0.073 s,which can quickly and accurately achieve heterogeneous remote image matching.

This paper proposes the Hartmann ray tracing method to measure the gas-phase parameters of the gas-liquid two-phase flow,and conducts theoretical and experimental research.First,the Hartmann template is used to obtain the array beam,and the propagation process of the light in the gas-liquid two-phase flow is simulated and traced to study the relationship between the emitted light and the bubble size and position in the two-phase flow.Second,the BP neural network model is established to achieve accurate inversion of bubble parameters.The simulation results show that the relative error of bubble positioning is within 7%,and the relative error of bubble size is within ±4%.On this basis,an experimental system is built to simulate the process of a single gas phase passing through the pipeline with transparent particles of known size settling in a vertical pipeline.Experimental research is carried out.The results show that the relative error of bubble size can be controlled within ±6%.

To solve the problems of real-time feature extraction on weld surface images,a feature evaluation algorithm based on incremental block principal component analysis (IBlockPCA) is proposed.First,the weld surface images are segmented into sub-image blocks,and then the blocks are reconstructed.Next,the incremental feature extraction is performed on the local block images by using the proposed IBlockPCA,and the KNN is used to classify and recognize the evaluated principal components.Finally,the performances are compared on the weld dataset.The experimental results show that the IBlockPCA is superior to other principal component analysis (PCA) algorithms in the convergence rate,classification rate and complexity.The classification rate is 97.5%,and the average processing speed can reach 50 frames per second.It can meet the real-time processing requirements of weld surface images.

Polarization imaging has attracted extensive attention in recent years due to its ability to detect object polarization information and effectively suppress noise signals of background.The previous research on sinusoidal amplitude grating imaging is basically carried out under the conditions of fully coherent,incoherent or partially coherent illumination.While this paper is based on the unified theory of polarization coherence,and it takes the polarization state of light into account under the condition of partially coherent illumination,and designs a partially coherent light′s polarization imaging system illuminated by a sinusoidal amplitude grating,according to which the Stokes frequency domain calculation model is obtained,and a method for calculating the frequency domain relationship of the Stokes object image is provided.Also,the frequency response of the polarization imaging system is described by the transmission cross coefficient,the nonlinearity of the polarization imaging system is showed,direct component,the fundamental frequency and the second harmonic in the grating image intensity are analyzed,and a set of normalized Stokes parameters is given,under which the partial coherent light irradiates the sinusoidal amplitude grating on the image plane intensity map,and this paper uses MATLAB to draw several sets of Stokes image intensity maps under the ratio of the illumination coherence interval of the object to the diffraction pattern size of the imaging system are plotted,which extends the application of the sinusoidal trajectory analysis in the performance evaluation of the polarization imaging system.

To solve the problems of low flexibility of fiber grating preparation,and not conducive to the realization of distributed wavelength division multiplexing in measurement,an eccentric fiber Bragg grating (EFBG) sensor based on point-by-point method was proposed.Using the technology of point-by-point method with femtosecond laser (FSL),the grating writing position is vertical deviation from the center by 3 μm,the length of the grating is 5 mm,the center wavelength of the grating is 1 633 nm.Different from traditional fiber Bragg grating (FBG),the eccentric structure of the grating can excite a wider range of cladding mode resonance.By analyzing the wavelength drift of the cladding mode resonance peak,the applied strain or temperature can be characterized.The experimental results show that in the strain measurement range of 0—500 με,the sensitivity of cladding mode is 0.98 pm/με,and in the temperature measurement range of 30—80 ℃,the sensitivity of cladding mode is 10.89 pm/℃.Besides,the difference between cladding mode sensitivity and core mode sensitivity is small,so the sensor measurement of strain or temperature can be realized.

To solve the problem of poor identification effect of current sensor pattern noise (SPN) extraction model in compressed video source detection,an improved SPN extraction model based on multi-scale transform domain adaptive Wiener filtering and a weighted maximum likelihood estimation is proposed.Firstly,the video coding and decoding is interfered,the video frame is extracted in front of the loop filter module of the coder,and then the video frame is input into the dual-density and dual-tree complex wavelet transform adaptive Wiener filtering model to extract the noise residual.Finally,the SPN is estimated from the noise residual by weighted maximum likelihood estimation.Test and comparison are completed on the public video source database VISION.The experimental results show that the proposed improved SPN extraction model performs better than the traditional SPN extraction algorithm on ROC curve and Kappa statistical coefficient.

Aiming at the problem of poor robustness of image watermarking algorithms when the attack is strong,a robust method based on graph-based transform (GBT),discrete wavelet transform (DWT) and singular value decomposition (SVD) is proposed.First,we perform non-overlapping block processing on the carrier image,select the sub-blocks with higher pixel variance values and perform DWT on each sub-block to obtain its low-frequency coefficient matrix,and then perform GBT and SVD on the low frequency coefficient matrix in turn to obtain the singular value matrix,and finally embed the watermark information to the largest singular value matrix.The experimental results show that when Pirate image structure similarity (SSIM) reaches 0.97 or more,the algorithm in this paper can effectively resist attacks such as noise,filtering,JPEG compression,cropping,and exchange of ranks,and the normalization coefficient (NC) value is all above 0.9.

To solve the problem of poor segmentation effect caused by small blood vessels in retinal images,loss of detailed features, gradient descent and explosion,a U-Net retinal vascular image segmentation model with residual block,cyclic convolution module and spatial channel extrusion excitation module is proposed.First,the training set is expanded by using a series of random enhancements,and then residual blocks are introduced into the U-Net model to avoid the segmentation accuracy reaching saturation and then rapidly degrading as the network depth increases.The bottom of the U-Net is replaced with a circular convolution module,the low-level features of the image are extracted,and features are continuously accumulated,the semantic information between contexts is enhanced,and a more effective segmentation model is obtained.Finally,the concurrent spatial and channel squeeze and channel excitation module is embedded between the convolutional layers.The excitation module finds the channel with stronger characteristic signal, and emphasizes this channel, compresses irrelevant channels,and reduces the interference of irrelevant characteristic information.Through the verification results on the DRIVE data set,the accuracy of the model proposed in this paper is 98.42%,the sensitivity reaches 82.36%,and the specific value reaches 98.86%.Compared with other network segmentation methods,the segmentation method proposed in this article has better segmentation effect.