最新刊期
- 摘要：MethodUsing a commercially available transparent phone case as the substrate， a silver@copper （Ag@Cu） micro- and nano-composite antibacterial coating was successfully constructed on its surface via LIBT.ResultExperimental results demonstrate a positive correlation between laser power and the elemental content of the micro-structured coating： the mass ratio of Ag and Cu increases with higher laser power. The process enables efficient fabrication under ambient atmospheric conditions， offering simplicity and cost-effectiveness. The micro-structured coating exhibits significant inhibitory effects against both Gram-positive （Staphylococcus aureus） and Gram-negative （Escherichia coli） bacterial strains. When the laser power reaches ≥16W， the antibacterial rate exceeds 99.99%， with the increased mass ratio of copper directly enhancing antibacterial efficacy.ConclusionThis work overcomes the limitations of traditional technologies， enabling the high-precision construction of micro- and nano-structures on transparent substrates. The process parameters are precisely controllable， making it suitable for large-scale production and providing a universal solution for the functionalization of other transparent organic material surfaces. This study not only develops a practical antibacterial phone case fabrication technology but also offers innovative ideas and methodological references for the design and manufacture of novel antibacterial surfaces.ObjetiveIn response to the problem of bacterial contamination on the surface of mobile phone casings and the lack of antibacterial performance， this study innovatively proposes the use of Laser-Induced Backward Transfer （LIBT） technology to fabricate highly efficient antibacterial mobile phone protective cases.关键词：Laser-induced backward transfer;Mobile;Protective case;Ag@Cu micro- and nano- structure;Antibacterial195|15|0更新时间：2025-11-25
- 摘要：To meet the requirements of augmented reality （AR） for optical waveguides in terms of large field of view （FOV）， thin and lightweight design， and high immersion， an optical system for a mass-producible， thin-lightweight transmissive head-mounted display （HMD） based on an arrayed geometric thin-film waveguide structure was designed and implemented. This system consists of an eyepiece system， an arrayed waveguide system， and an illumination system. The arrayed waveguide structure adopts an array of semi-transparent and semi-reflective films with customized angle-selective film coatings as the optical combiner， enabling pupil expansion. Through the "dual waveplate" polarization mixing design， the display uniformity is improved from 59.66% to 61.12% （for single polarized light） to 76.61% （in actual tests）. By means of systematic optical simulation and design， and relying on a mature domestic optical manufacturing chain （including precision cutting， grinding and polishing， vacuum deposition of complex film systems， and precision bonding processes）， while adding an additional waveplate lamination process， the key technical difficulties restricting the mass production of arrayed waveguides were successfully overcome. A prototype was fabricated and mass production was realized. This design addresses the pain point of AR head-mounted displays where "performance and mass production are difficult to balance" and meets the needs of both consumer and industrial applications.关键词：waveguide;Near-eye display;optical design247|31|0更新时间：2025-11-25
- 摘要：Spectral imaging enables the acquisition of spectral information at every pixel of a scene， providing essential support for high-precision target recognition and classification. Conventional scanning-based spectral imaging systems sequentially scan along spatial or spectral dimensions to obtain complete spectral images， which inherently limits their acquisition speed and signal-to-noise ratio. In contrast， computational spectral imaging， grounded in compressed sensing theory， introduces optical encoding elements to project high-dimensional spectral image data into low-dimensional measurements， which are subsequently decoded into spectral images using advanced reconstruction algorithms. Computational spectral imaging technology demonstrates remarkable advantages in structural compactness， acquisition speed， and manufacturing cost. In recent years， it has achieved rapid progress in both theoretical methodologies and system implementations， leading to a surge of high-quality research outcomes. Meanwhile， its consumer-level applications have been increasingly extended to platforms such as smartphones， unmanned aerial vehicles， and remote sensing satellites， serving diverse fields including color imaging， environmental monitoring， and medical diagnostics. This paper systematically reviews the theoretical foundations and methodological advances of computational spectral imaging. Specifically， it examines representative optical encoding strategies—including amplitude encoding， wavelength encoding， wavefront encoding， and multi-aperture encoding—as well as mainstream reconstruction approaches ranging from iterative algorithms with prior constraints to end-to-end deep learning models. Finally， current trends and key challenges are discussed. With strong alignment to strategic emerging industries such as intelligent manufacturing， artificial intelligence， the low-altitude economy， and smart agriculture， computational spectral imaging is expected to play an increasingly significant role across a wide spectrum of applications.关键词：computational imaging;spectral imaging;Compressed sensing;deep learning462|40|0更新时间：2025-11-25
- 摘要：In order to realize the detection and control of 3D printed pieces and improve their printing accuracy， the research of 3D reconstruction of 3D parts and position estimation is completed. The system is based on the peripheral scanning visual detection principle of binocular structured light， adopts binocular structured light illumination， and takes the peripheral scanning imaging mode of dual-color camera to realize image acquisition and visual calibration through the color and infrared scene at different positions， binocular vision and scattered structured light depth information， etc. It completes the image processing and analysis， such as image fusion， point cloud coloring， multi-frame point cloud alignment fusion， segmentation， etc.， so as to realize the reconstruction of object field Point cloud reconstruction. The camera position estimation scheme based on EPNP and ICP algorithms is adopted， and the EPNP algorithm completes the coarse alignment of the reconstructed object scene point cloud and single-view point cloud， while the ICP algorithm completes the fine alignment of the reconstructed object scene point cloud and single-view point cloud to obtain the position estimation of the target. The accuracy of 3D printed pieces’ 3D reconstruction is evaluated by calculating the chamfer distance between the scene point cloud and the standard point cloud， and the average accuracy is 0.675mm； the accuracy of position estimation is evaluated by the reprojection method， and the average accuracy is 1.669mm.Through the systematic research， a better evaluation method is provided for the printing inspection of 3D pieces， and a better reference is provided for the subsequent inspection and control of the accuracy of 3D pieces.关键词：Machine Visualization;3D reconstruction;position estimation;image processing;EPNP;ICP1163|851|0更新时间：2025-05-30
- 摘要：In response to the application requirements of high-resolution Reflection High-Energy Electron Diffraction （RHEED） in the fields of microelectronics manufacturing and surface analysis， and considering that a long working distance， micro-beam spot size， and small beam half-angle electron gun are key components for achieving high-resolution RHEED detection， a micro-beam spot RHEED quasi-parallel beam electron gun was developed. The characteristics of the electron gun's electron optical system were analyzed， and a low-aberration focusing magnetic lens was designed utilizing electron optical simulation software. An experimental platform was set up to measuring the beam spot diameter， beam current， and beam half-angle performance of the developed electron gun， as well as to conduct diffraction imaging tests on highly oriented pyrolytic graphite （HOPG） samples. The experimental results show that at a working distance of 500 mm， the beam spot diameter of the RHEED quasi-parallel beam electron gun is 47.1 μm （at an acceleration voltage of 30 kV）， The emission current and the beam half-angle are 144.96 μA and 0.289 mrad respectively （at an acceleration voltage of 15 kV）. Clear diffraction spots which intensity corresponding to the crystal structure factors were obtained on the HOPG sample.关键词：electron optics;Reflection high energy electron diffraction （RHEED）;Electron gun;Quasi-parallel electron beam1049|601|0更新时间：2025-05-30
- 摘要：In the traditional optical system design process， optical designers focus more on optimizing the performance of the optical system without considering the as-built performance of these systems. To reduce the tolerance requirements for optical systems， this paper proposes a desensitization design method for freeform reflecting telescopes based on nodal aberration theory. This method provides mathematical expressions to describe aberrations generated by freeform terms on general decentered and tilted optical surfaces. To obtain the aberration coefficients of off-axis freeform surfaces， transformed pupil vectors are introduced into the aberration expansion， and only paraxial chief rays and marginal rays are traced to derive the aberration expressions. Then， an as-built performance evaluation model for the optical system is constructed based on this analysis framework. Using this method， two kinds of off-axis two-mirror and off-axis three-mirror systems containing freeform surfaces were designed， and the assembly sensitivity of the system was analyzed by 2000 Monte Carlo ray tracing simulations. The results indicate that after about 10 minutes of optimization， the average wavefront aberration of the off-axis two-mirror telescope decreased by 26%， and the average wavefront aberration of the off-axis three-mirror telescope decreased by 14%， effectively verifying the effectiveness of the desensitization design method proposed in this paper.关键词：Nodal Aberration;Zernike Freeform Surface;Off-Axis Reflective System;Desensitization Design990|577|0更新时间：2025-05-30
- 摘要：The analysis of the seed respiration process is of critical importance for accurately assessing seed viability. Therefore， the development of a rapid， stable， and highly sensitive seed respiration detection device is essential. In this study， a CO2 gas sensor was designed and constructed based on STM32 microcontroller and Cavity Ring-Down Spectroscopy （CRDS） technology， and it was subsequently applied to seed respiration detection. The sensor employs an STM32-based control and data acquisition scheme， comprising a threshold detection and shut-off module formed by a high-speed comparator and RS flip-flop， a high-speed analog signal acquisition module， an STM32 master control unit， and an upper computer data processing system. The threshold detection module achieves a cutoff delay of 0.45 μs， while the maximum sampling rate of the analog signal acquisition module reaches 31.25 Msps. Experimental verification confirmed that the sensor can detect CO2 concentrations as low as 1.5 ppm， demonstrating good sensitivity. The sensor successfully captured the respiration process of rice seeds， producing a curve of CO2 concentration changes over time for 10 g of rice seeds. Within 2.5 hours， the CO2 concentration increased by approximately 730 ppm. This study highlights the potential and application value of CRDS technology in seed respiration detection.关键词：Cavity ring-down spectroscopy;STM32;carbon dioxide;seed respiration692|484|0更新时间：2025-05-30
- 摘要：MethodFirstly， polishing experiments were conducted on polymethyl methacrylate （PMMA） using MCF slurry. Changes in the MCF morphologies before and after polishing were observed， and the surface roughness， material removal rate， temperature， and normal force were measured to analyze the performance variations over time. Additionally， supplementary abrasive particle and α-cellulose solutions were added every 10 minutes to evaluate their effectiveness in restoring the performance of MCF and extending its service life. Finally， long-term polishing experiments were conducted to assess the service life of the MCF slurry under the influence of the supplementary solutions.ResultThe results indicate that after 60 minutes of continuous polishing， the MCF slurry morphology changed from uniform magnetic clusters to fragmented clusters. The surface roughness reduction rate decreased from 97.06% to 65.97%， showing significant performance degradation. By adding abrasive particle and α-cellulose solutions， the normal force was stabilized at 6.4 N and 7.3 N， respectively， with the surface roughness reduction rate maintained above 85%. Further investigations demonstrated that adding 0.1 mL of supplementary solution every 10 minutes extended the service life of the MCF slurry from 60 minutes to 180 minutes， while maintaining the surface roughness below 0.05 μm and the material removal rate above 1.80 × 10⁸ μm³/min.ConclusionDuring the polishing process， the MCF slurry undergoes the loss of water， abrasive particles， and α-cellulose， leading to a decline in polishing performance. By quantitatively replenishing these key components， the polishing performance of the MCF slurry can be effectively restored， ensuring improved polishing stability and significantly extending its service life.ObjetiveMagnetic Compound Fluid （MCF） polishing is an efficient precision polishing technology. However， the MCF slurry experiences performance degradation during prolonged use. This study investigates the effects of component loss on the polishing performance of MCF slurry and explores the feasibility of restoring its performance through the addition of supplementary solutions to extend its service life.关键词：magnetic compound fluid;polishing;surface roughness;material removal rate;Life1066|475|0更新时间：2025-05-30
- 摘要：The distributed optical fiber vibration sensing system usually uses the relative intensity ratio method to identify and locate the vibration event， which ignores the influence of noise level on the system dynamic range， thus restricting the improvement of the event location accuracy and dynamic range. In this paper， the relationship between the noise level and the dynamic range of the system is studied under the condition that the event location accuracy of the system is ensured， and then virtual noise level that corresponds to the average voltage of the near-zero amplitudes in the optical time domain reflectometry signal is added， and then a performance optimization algorithm based on virtual noise level is proposed. In the experiment， a phase-sensitive optical time domain reflectometry system is built， and laser pulse with pulse width of 50ns is used to conduct vibration event location testing along a sensing fiber with length of about 15km， and based on optimization algorithm with the virtual noise level， the noise level is removed from the original optical time domain reflecting data， and virtual noise levels with amplitude of 25%， 50%， 75%， 100%， 125%， 150%， 175% and 200% of the mean value of the near-zero amplitude of the signal are respectively added to finally determine the virtual noise level that ensures the system dynamic range and the signal-to-noise ratio of the vibration signal are overall optimal， whose amplitude is about 75% of the mean value of the near-zero amplitude of the signal， and the dynamic range of the system is improved by 11.26dB， compared with the traditional relative intensity ratio method， so the adoption of virtual noise level improves the performance of the system by reducing the sensitivity of the near-zero amplitude.关键词：Distributed optic fiber sensing;Optical time domain reflectometry;vibration monitoring;Dynamic range;Virtual noise1207|592|0更新时间：2025-05-30
- 摘要：To address the limited interference resistance and large size of conventional monolithic triaxial gyroscopes， a novel quad-mass high-frequency triaxial MEMS gyroscope is presented. Compared with traditional gyroscopes， this device has a relatively higher resonance frequency （~30kHz）， the size of the sensitive structure is only 3mm×2.56mm. This paper describes the sensitive structural form and working principle of the designed gyroscope. Moreover， the width of the beams of the sensitive structure is optimized through a multi-objective genetic algorithm to make the resonance frequency of the working mode higher than 30kHz， and the frequency difference between the drive mode and the sense mode less than 200Hz. It is fabricated by using a three-wafer bonding technology and characterized by a swept-frequency testing system. The experimental results show that the resonance frequencies and quality factors of its drive mode， X-axis， Y-axis， and Z-axis sense mode are 29692Hz and 1274， 31290Hz and 354， 29881Hz and 305， 30721Hz and 393， respectively. The results verify the correctness of the design and research methods， and provide a feasible solution for the development of high-frequency triaxial MEMS gyroscopes.关键词：MEMS gyroscope;monolithic triaxial;high-frequency;multi-objective optimization1127|629|0更新时间：2025-05-30
- 摘要：To enable the retrieval of nearshore lake water turbidity using spaceborne LiDAR data， this study processes ICESat-2 data to extract photon distribution characteristics over lake surfaces. Leveraging the variation in photon distribution patterns under different turbidity conditions， turbidity levels are inferred accordingly. Lake Erie， one of the North American Great Lakes， is selected as the study area. An adaptive-parameter pruned quadtree algorithm is employed to denoise the ATL03 photon data from ICESat-2， isolating valid water surface photon returns.Key photon features：penetration depth， photon density， and attenuation rate are extracted based on the processed data and matched with in situ turbidity measurements. A turbidity retrieval model is then developed using machine learning regression algorithms. Experimental results demonstrate that the Random Forest algorithm yields the best performance， achieving a coefficient of determination （R²） of 0.91， a mean absolute error （MAE） of 1.66 NTU， and a root mean square error （RMSE） of 2.17 NTU， indicating high retrieval accuracy within the 0-50 NTU turbidity range.To further assess the method’s applicability under different turbidity conditions， the dataset is divided into low-to-moderate turbidity （0-30 NTU） and high turbidity （>30 NTU） subsets. Results show that retrieval accuracy is slightly higher for the low-to-moderate turbidity group. This study provides a novel technical approach for remote sensing-based monitoring of lake water turbidity.关键词：ICESat-2;Nearshore Lake Surface Water;Turbidity Inversion;lidar;random forest1139|553|0更新时间：2025-05-30
- 摘要：The asymmetrical shape of the off-axis aspheric surface and the nonlinear change of its surface curvature bring challenges to the high-precision measurement of surface shape. In order to solve the problem that the existing off-axis aspheric surface shape measurement methods are highly dependent on precise initial alignment of the test piece and poorly adaptive to curvature variations， we propose a spatially-constrained differential confocal adaptive measurement method to realize the adaptive and high-precision measurement of off-axis aspheric surface without initial pose dependence. Firstly， according to the translation-rotation scanning measurement principle and the performance of differential confocal technology in anti-surface inclination accurate fixed focusing， we develope a spatial constraint model incorporating both distance and tilt angle parameters between designed measurement points and actual test locations. This model enables optimized spiral scanning path planning with curvature-adaptive adjustment capability for off-axis aspheric surfaces. Experimental verification demonstrated surface form accuracy with root mean square （RMS） errors below 10 nanometers and 3σ values under 5 nanometers compared to Zygo interferometer references， meeting the requirements of high-precision measurement of off-axis aspheric surfaces.关键词：Spatial constraints;Differential confocal;Off-axis aspherical;adaptive measurement;path planning623|280|0更新时间：2025-05-30
- 摘要：Inverse perspective mapping （IPM） of pavement images is a prerequisite for image-based vehicle distance perception and pavement damage measurement. The traditional static IPM methods have the problem that the transformation parameters cannot be dynamically adjusted， and the existing dynamic IMP methods are highly dependent on the information such as road lane lines and textures，which often lead to suboptimal correction of perspective distortion. To solve these problems， this study proposes a dynamic IPM method for pavement images based on depth camera semantic segmentation and 3D plane fitting. First， a semantic segmentation model is used to extract pavement regions from RGB images， and 3D plane fitting is performed on the corresponding point cloud data within the pavement regions， eliminating the interference of non-pavement point clouds on pavement fitting. On this basis， using pavement information and the spatial positional relationship between the camera and the pavement， the relative pose of the camera with respect to the pavement is calculated through a camera pose estimation method. Finally， based on the imaging relationship of the pavement under different camera poses， a constructed pavement image IPM model is used to correct perspective distortion from the original image to any reference point. Simulation experiments show that the perspective distortion correction error of the proposed method is stable at 10-2mm when the camera pose has common variations， which is better than the current advanced IPM methods， demonstrating that the proposed method effectively improves the quality of pavement image IPM. Real-world experiments further validate the effectiveness of the proposed method.关键词：machine vision;pavement image;dynamic inverse perspective mapping;semantic segmentation;pose estimation1016|292|0更新时间：2025-05-30
- 摘要：Industrial Computed Tomography （ICT） technology provides a non-contact solution for precision measurement of small module plastic gears， effectively addressing multiple challenges in their metrological processes. This study provides a data processing method for plastic small modulus gear point clouds obtained from ICT measurements. In data preprocessing， the gear measurement model obtained by ICT scanning are covert to the gear measurement point cloud. Meanwhile， a design point cloud is generated based on the gear CAD model. Subsequently， the point cloud registration process is performed between the measured point cloud and the designed point cloud， including four steps： pre-registration， coarse registration of hole point cloud， fine registration of hole point cloud， and registration of gear tooth point cloud. Then， a measurement point cloud segmentation method guided by design point clouds are adopted. The article first uses the DBSCAN algorithm to segment the designed tooth surface point cloud. Then the designed tooth surface point clouds are used to guide the segmentation of the measured tooth surface point cloud. Pitch deviation evaluation points are extracted through this procedure for comprehensive pitch deviation assessment. Experimental results demonstrate that compared with gear measuring center measurements， the proposed method achieves the difference with maximum absolute value of single pitch deviation of 2.8μm for left and right tooth flanks， and the difference with maximum absolute value of total cumulative pitch deviation of -6.6μm. The developed ICT-based data processing methodology for gear pitch deviation measurement establishes a methodological foundation for precision evaluation of complex CT-measured gears， while providing valuable references for ICT-based measurement of other intricate precision components.关键词：Industrial CT;gears;point cloud registration;point cloud segmentation;Pitch deviation924|357|0更新时间：2025-05-30
- 摘要：To address the issues of large number of parameters and false detection and missing detection of garbage detection model in complex environments， this paper proposes a lightweight garbage detection model based on improved YOLOv8n. Firstly， a lightweight network structure MobileNet V3_ECA was proposed as the backbone network of YOLOv8n， which improved the ability of the model to express garbage feature regions and reduced the number of model parameters. Secondly， the Context Anchor Attention （CAA） mechanism is introduced into the backbone network to enhance the model's ability to extract garbage features. Then， the Omni-Dimensional Dynamic Convolution （ODConv） was used to replace the standard convolution in the neck network， and the local feature mapping was refined to realize the fusion ability of the local fine-grained features of garbage. Finally， the Wise Intersection Over Union （WIoU v3） boundary loss function is used to improve the regression performance of the network bounding box. Compared with the original YOLOv8n， the improved model is improved by 1.1% in mAP@0.5， the detection speed is increased by 11.7%， and the parameter Params， model size and floating-point operation FLOPs are reduced by 70.8%， 66.1% and 70.7%， respectively. Experimental results demonstrate that the improved model can effectively improve the detection accuracy and significantly reduce the complexity of the model， which has important engineering significance for the deployment and application of the model to the edge detection equipment.关键词：garbage detection;lightweight;MobileNet V3_ECA;YOLOv8n;deep learning1011|267|0更新时间：2025-05-30
- 摘要：A fusion method was proposed to solve the problem that the single-energy X-ray cannot detect the complete decoration and disease information of the corroded ancient bronze mirror due to the uneven thickness of the mirror edge and the mirror center area. The method combined intuitionistic fuzzy set entropy measure and salient feature detection to fuse ancient bronze mirror X-ray images. Firstly， the effective guided filtering was introduced to enhance the contrast of the decorative structure of high-energy X-ray images. Secondly， a three-scale decomposition model was designed by using joint bilateral filtering and structure-texture decomposition strategy. The model extracted the energy layer， residual layer and detail layer information of different energy X-ray images. Then， the energy layer obtained the fused energy image through the
- 摘要：Haze in natural environments is usually non-homogeneous and irregular， which has a large impact on computer vision tasks. Therefore， this paper proposes Enhanced-edge-feature Dual-branch Fusion Dehazing Network （EDFDNet）. In order to retain the realism of the image and at the same time effectively improve the visibility after dehazing in the case of severe blurring， the transmission graph fine branch is constructed， which is the premier branch of the network， and the U-shaped network hierarchical codec structure that fuses the discrete wavelet transform is used to extract the multi-scale fine feature information， and the mathematical method for the determination of the enhanced edge information is defined； the feature extraction branch tandemly connects the ResNet residual block and the Transformer combined with dual attention for parallel feature extraction module， which fuses the extracted local and global features， improves the network's ability to understand and process non-uniform haze images， and further restores the visibility of the images， and joins the above two branches into the backbone framework of Generative Adversarial Network （GAN）， and defines a mathematical method to strengthen the determination of edge information. GAN） backbone framework to form the defogging network EDFDNet.The results of the experiments show that the average PSNR and SSIM of this method on the outdoor synthetic dataset are improved by 1.2567 and 0.0308， respectively， compared with the optimal results of the current mainstream methods.Meanwhile， in the test on the real-world dataset， the PIQE， RI， and VI reach the optimal indexes of 21.471， 0.9711 and 0.9003.EDFDNet achieves good results in both realism enhancement and visibility restoration， and is suitable for dehazing real-world non-uniform haze images.关键词：real-world haze;non-homogeneous dehazing;feature fusion;edge feature enhancement;generative adversarial network911|484|0更新时间：2025-01-08
- 摘要：In order to achieve cell-level operations such as cell capture， cutting， separation and injection， a flexible parallel piezoelectric positioning stage for biocellular engineering is designed， modeled， simulated and tested in this paper. The positioning stage consists of a moving platform， a base， a three-stage amplification mechanism and three piezoelectric actuators. The displacements generated by the piezoelectric actuators are amplified by the three-stage amplification mechanism， and the precise movement of the positioning stage is realized through feedback control， so as to achieve the target positioning effect. In the design process， the pseudo-rigid-body method combined with the flexible hinge stiffness calculation model is adopted to analyze the kinematic statics of the mechanism. The Lagrange equation is used to establish the dynamics model of the designed flexible parallel piezoelectric positioning stage using the lumped mass method. After determining the structural parameters， finite element analysis is carried out to verify the derived theoretical model， and the simulation results show that the error between the theoretical and simulation models is less than 10%， and the mechanism is able to achieve a large stroke as well as a higher frequency of motion. In addition， a prototype system for the flexible parallel piezoelectric positioning stage is also built and experimentally tested to evaluate its open and closed loop performance. The experimental results show that the designed positioning stage has a working stroke of 125μm×126μm， the natural frequencies in the X- and Y-direction are 128.9 Hz and 132.8 Hz， and the corresponding motion resolution are both better than 400 nm， respectively.关键词：Cell micromanipulation;piezoelectric actuators;flexible parallel mechanism;micro-positioning stage923|826|0更新时间：2025-01-08
- 摘要：Aiming at the problems of unclear texture details and poor visual perception due to neglecting illumination in infrared and visible image fusion under low-light conditions， a low-light enhancement and semantic injection multi-scale infrared and visible image fusion method is proposed. Firstly， a network suitable for low-light enhancement is designed to realize the enhancement of visible image in nighttime scenes by repeated iterations of residual models. Then， a feature extractor based on Nest architecture is used as the encoder and decoder of the network， in which the deep features can capture the complex structure and semantic information of the images， a semantic prior learning module is designed to further extract the semantic information of the deep infrared and visible images through cross-attention， and a semantic injection unit is adopted to inject the enhancement features into each scale step by step. Thirdly， a gradient enhancement branch is designed， where the mainstream features are firstly passed through the hybrid attention， and then the Sobel operator stream and Laplacian operator stream are divided from the mainstream as a way to enhance the gradient of the fused image. Finally， the features at each scale are reconstructed by dense connections between the same layers and jump connections between different layers in the decoder. Experimental results show that this method improves the visual information fidelity， mutual information， disparity correlation coefficient， and spatial frequency， on average， by 23.1%， 16.3%， 18%， and 39.8%， respectively， in comparison with the nine methods， which effectively enhances the quality of fused images in low-light environments， and helps to improve the performance of the advanced visual tasks.关键词：infrared and visible image fusion;multiscale fusion networks;low-light enhancement;cross-attention;semantic injection953|533|0更新时间：2025-01-08
- 摘要：Due to the attenuation and scattering of light in an underwater environment， the images directly captured by imaging equipment suffer from significant quality degradation. Although learning-based underwater image enhancement methods improve the original image imaging quality to a certain extent， most of the existing methods use artificially synthesized or model-generated paired datasets for training. Meanwhile， there is a large domain difference between artificial or model-generated images and real underwater images in distribution， which leads to problems of excessive enhancement and no obvious removal of color shift in the enhancement results. Focus on these problems， an underwater image enhancement model that integrates domain transfer and attention mechanism has been proposed in this paper. First， an image generation network with domain transfer is designed， and combined with the physical imaging model and the water type classifier. In this way， the feature description mapping between images in different domains and scenarios could be learned， hence reducing the difference between the generated images and the real images. Furthermore， a multi-scale hybrid attention encoder-decoder network is designed. With the help of efficient feature connections and different attention fused structures， the image local details recovery ability of the model could be improved. Finally， a global domain association consistency loss function is proposed to better train the network model parameters and improve the quality of image enhancement by constructing content and structure consistent associations of the generated images at each stage of the domain transfer. The proposed model achieved accuracies of 3.1401， 0.6021 and 3.0768， 0.6124 for the UIQM and UCIQE metrics on the underwater real datasets UIEB and EUVP， respectively. The experiments show that the proposed model could effectively improve the color recovery ability of underwater images， and more details could be recovered.关键词：underwater image enhancement;domain transfer;generative adversarial networks;Attention enhancement979|457|0更新时间：2025-01-08
- Address：No.3888 Dong Nanhu Road, Changchun, Jilin, China Postal code：130033
- Tel：0431-86176855 Email：gxjmgc@ciomp.ac.cn
- Technical support is provided by Beijing Founder electronics co., LTD 吉ICP备11002662号-17
京公网安备11010802024621 - It is recommended to read the content of this site in Chrome&IE9+. Please switch to extreme mode in browser 360.
- Cookies We use cookies to help provide and enhance our service and tailor content. By continuing, you agree to the use of cookies.
0