最新刊期
卷 34 ， 期 11 ， 2026
- 摘要：A retrieval method for pressure-elevation parameters based on satellite polarization imaging data is developed in this study， aiming to retrieve cloud-top pressure and reconstruct surface elevation. Based on observations from the DPC instrument onboard the GF-5（02） satellite， the sensitivity of oxygen A-band absorption to atmospheric path length is used to establish a nonlinear mapping relationship between pressure and observed signals， thereby enabling cloud-top pressure retrieval. Furthermore， by incorporating the physical relationship between atmospheric pressure and altitude， surface elevation reconstruction is achieved. Key factors affecting the retrieval process are systematically analyzed， including surface elevation errors， aerosol optical thickness， and atmospheric air mass. The results show that， in the multi-angle stability validation over the Qilian Mountains， the mean coefficient of variation of the pressure retrieval is approximately 3.98%， with overall fluctuations limited to within 8%， indicating good stability of the proposed method. In the DEM-based validation over the Asir Mountains in western Saudi Arabia， the retrieved pressure shows a correlation coefficient of 0.91 and an RMSE of approximately 28.6 hPa compared with DEM-derived pressure. The reconstructed elevation further shows a correlation coefficient of approximately 0.93 and an RMSE of about 0.23 km compared with DEM， effectively reflecting regional topographic variations. Sensitivity analysis indicates that aerosol optical thickness in the range of 0.4-2.0 has a limited impact on the retrieval error， generally within about 1%， whereas elevation perturbations exert a more significant influence. Under ±10% elevation perturbations， the relative pressure error ranges from approximately 1% to 8%. Overall， this study verifies the feasibility of retrieving cloud-top pressure and reconstructing surface elevation based on satellite polarization imaging data， and provides a basis for future three-dimensional reconstruction schemes under integrated observations with high spatial， spectral， and polarization resolution.关键词：multi-angle polarimetric instrument;cloud-top pressure;oxygen A absorption band;differential absorption spectroscopy159|103|0更新时间：2026-06-24
- 摘要：In full-Stokes polarization imaging based on liquid crystal modulation and rotating waveplate methods， the influence of phase retardation errors arising from the neglect of oblique incidence in polarization elements has not been adequately considered， leading to reduced measurement accuracy. Building upon conventional polarization measurement approaches， error correction models for waveplate and liquid crystal phase retardation were established as functions of the beam incidence angle. The quantitative relationships between waveplate retardation and both incidence angle and wavelength， as well as between liquid crystal retardation and incidence angle， driving voltage， and wavelength， were systematically characterized. A 1∶1 secondary imaging telecentric relay optical system was designed to enable flexible switching among standard commercial lenses. Experimental results demonstrate that， with the incorporation of incidence-angle correction， the maximum relative errors in the degree of polarization， linear polarization degree， circular polarization degree， and ellipticity angle were reduced by approximately 7.72%， 8.31%， 10.50%， and 12.93%， respectively. These findings provide a theoretical foundation and technical support for high-precision polarization imaging， as well as for the optimized design and engineering application of polarization optical systems under oblique incidence conditions.关键词：polarization imaging;liquid crystal modulation;rotating waveplate method;full-stokes;error correction92|104|0更新时间：2026-06-24
- 摘要：To address the surface normal ambiguity in Shape from infrared polarization and the lack of physical constraints in purely data-driven methods， a reconstruction approach integrating infrared polarization physical priors with deep learning is proposed to improve the accuracy and stability of surface normal estimation and achieve high-precision 3D reconstruction. First， polarization information， including infrared intensity， degree of linear polarization （DoLP）， and angle of polarization （AoP）， is extracted from raw polarization images. Based on this， the zenith angle of the surface normal is recovered according to the relationship between the zenith angle and DoLP derived from a mixed polarization radiation model. Meanwhile， two candidate azimuth angles are computed using the correspondence between AoP and surface geometry， thereby constructing candidate surface normals. In parallel， a ThermalUNet is designed to generate reference surface normals from polarization information. The reference normals are further employed to constrain and refine the candidate normals derived from the physical model， resulting in a consistent and stable normal field. Experimental results on the public ThermoPol16 long-wave infrared polarization dataset demonstrate that the proposed method achieves a mean angular error of 8.36°， with 81.97% of pixels having angular errors less than 11.25°， outperforming existing methods. Furthermore， validation on a self-collected mid-wave infrared polarization dataset shows that the Dataset-level mean angular error is 7.3° for spherical targets with different temperatures and materials， indicating that the proposed method can stably and accurately recover surface geometry. In summary， the proposed method effectively resolves the normal ambiguity in polarization-based 3D reconstruction and exhibits strong accuracy and robustness under varying materials， temperatures， and infrared spectral bands.关键词：infrared polarization imaging;shape from polarization;computational imaging;surface normal vector104|113|0更新时间：2026-06-24
- 摘要：A method for retrieving leaf surface roughness based on a multi-angular polarized reflectance model is proposed. By incorporating a non-polarized component into the Litvinov polarized reflectance model， an improved model was developed to establish a quantitative relationship between the leaf reflectance factor and surface roughness. Nineteen leaf samples from five plant species with distinct surface structures were used， and multi-angular leaf measurements in the principal plane were conducted to systematically evaluate the model performance in reflectance simulation and roughness inversion. The results showed that the improved model accurately reproduced the leaf reflectance factor across the 350-2 500 nm spectral range， with high overall accuracy （R²=0.99） and strong stability under varying viewing zenith angles （relative root mean square error <5%）. For roughness retrieval， robust performance was achieved across most wavelengths （R² generally >0.5）， with the highest accuracy observed at 450 and 550 nm （R²>0.7）. Differences in leaf surface structure were effectively distinguished， and relative variations in roughness were reliably captured. Sensitivity analysis indicated that leaf surface roughness and illumination-viewing geometry jointly controlled the magnitude and angular distribution of the polarized reflectance factor. Specifically， roughness was found to govern the magnitude and angular distribution of the leaf reflectance factor in the principal plane. This study demonstrates the feasibility of retrieving leaf surface roughness using a polarized reflectance-based model. The proposed approach overcomes the limitations of conventional intensity-based methods in characterizing surface structural parameters and provides a new pathway for vegetation structure retrieval， with potential extension to canopy-scale applications.关键词：polarized reflectance model;leaf surface roughness;multi-angular observation;leaf parameter inversion63|91|0更新时间：2026-06-24
- 摘要：To address contrast degradation and texture loss caused by severe color casts and highly turbid underwater environments， a multi-scale deep learning network for image enhancement based on spatial-frequency domain fusion is proposed. By integrating the Retinex physical mechanism with multi-scale deep features， the proposed method effectively suppresses scattering noise interference in turbid water. In addition， a collaborative framework combining hybrid convolution with spatial-frequency domain attention is constructed， in which a dual-path attention structure is embedded within the encoder-decoder stages to simultaneously capture dual-domain features， thereby improving the modeling capability under complex degradation conditions. Furthermore， an adaptive selective-kernel fusion mechanism is employed to refine deep semantic features， enabling accurate texture reconstruction and color correction in complex scenes. Experimental results demonstrate that， on turbid images with a turbidity of 54.11 NTU， including color-neutral， green-cast， and blue-cast samples， the proposed method achieves average PSNR improvements of 0.464， 0.473， and 1.944 dB， respectively， over state-of-the-art methods， while reducing the LPIPS metric by an average of 0.94%. These results indicate that the proposed method delivers superior performance in turbid underwater image enhancement and provides an effective solution for underwater robotic vision.关键词：computer vision;underwater image enhancement;turbid water;deep learning52|22|0更新时间：2026-06-24
Polarization remote sensing
- 摘要：To address the challenges of small target size， low brightness， strong noise interference in event-camera imaging， and limited centroid extraction accuracy for dynamic star targets， a dynamic star centroid extraction method based on asymmetric-neighborhood density clustering and robust event-shape fitting using iterative reweighted least squares is proposed. irst， the event stream of dynamic star targets is segmented into event frames using a fixed time window. Within each frame， a directional anisotropic density-based clustering approach is employed to aggregate events via density connectivity， effectively separating true star-event clusters from noise. Subsequently， the extracted star-event clusters are projected onto a frame-centered plane along the principal component analysis direction， where an elliptical envelope is constructed in the two-dimensional projection domain. An improved iterative reweighted least squares fitting method is then applied to suppress outliers and estimate subpixel centroid positions. Experimental results demonstrate that， at a target motion speed of 20 （°）/s and an output frequency of 60 Hz， the root-mean-square error of centroid localization reaches 0.107 8 pixels. The proposed method enables accurate subpixel localization and rapid tracking of high-dynamic targets in event cameras， indicating strong potential for application in star sensor systems.关键词：star point;centroid extraction;event camera;high-dynamic target tracking;robust least squares41|21|0更新时间：2026-06-24
- 摘要：To address consistency errors among multiple channels in conventional point-source polarized light sensors and to improve the accuracy and reliability of angular measurements， a precision scanning polarized light detection system is proposed. Based on the distribution characteristics of atmospheric polarized light and the principles of polarization detection， a precision turntable is employed to drive the rotation of a polarizer. The positioning accuracy of the turntable directly determines the accuracy of polarization angle estimation. Encoder raw data are processed to obtain high-resolution， high-precision angular information， which is used as the phase reference for polarization calculation.Polarized light intensity data are acquired at discrete angular positions and fitted to a theoretical polarization response model using a linear least-squares method， enabling accurate determination of the polarization angle. Real-time temporal and geographic coordinate information provided by the carrier positioning system are used to calculate the theoretical solar azimuth angle in the geographic coordinate system， while the observed solar azimuth angle is simultaneously derived in the carrier coordinate system. By aligning these datasets along a unified time axis and performing point-by-point comparison and statistical analysis with time offset as a variable， the system’s angular measurement accuracy and stability are quantified.Experimental results obtained in indoor and clear， open outdoor environments demonstrate that the proposed system achieves angular measurement accuracies of ±0.01° and ±0.02°， respectively， thereby confirming its high precision and effectiveness in polarization angle measurement.关键词：biomimetic navigation;polarized light detector;scanning type;single channel53|31|0更新时间：2026-06-24
Modern Applied Optics
- 摘要：Microstructure arrays are critical functional elements in advanced optical systems， and their mass production relies heavily on compression molding. Molds are typically fabricated from difficult-to-machine materials such as mold steel， making the efficient and high-precision fabrication of surface microstructures a key technical challenge. To address this issue， a hybrid method integrating high-frequency elliptical vibration cutting （HEVC） with fast tool servo （FTS） technology is proposed to enhance both machining accuracy and efficiency.A transfer matrix-based analytical model for longitudinal–bending coupled vibration was established to guide the structural design of the high-frequency elliptical vibration device. In conjunction with finite element simulations for optimization， a lightweight ultrasonic elliptical vibration device with a resonant frequency of 110 kHz was developed. The device is capable of delivering stable vibration amplitudes of 2.5 μm and 1 μm in the bending and longitudinal directions， respectively. A hybrid machining platform was subsequently constructed. The influence of load mass on the output amplitude of the FTS system was systematically investigated， identifying an optimal working load range （≤100 g） and confirming the feasibility of dynamic coupling between the developed vibration device and the FTS system.Machining experiments on S136 mold steel were conducted to fabricate microstructure arrays. The results demonstrate that the produced microstructures exhibit uniform morphology， with depth and width errors of 11.62% and 3.61%， respectively， and negligible tool wear. These findings validate the feasibility and stability of the proposed hybrid machining strategy and provide an effective technical approach for the efficient and ultra-precision fabrication of complex microstructures on difficult-to-machine materials.关键词：ultra-precision machining;mold steel;microstructure arrays;ultrasonic elliptical vibration cutting;fast tool servo46|21|0更新时间：2026-06-24
- 摘要：Ceramic materials are widely used in engineering friction pairs because of their high hardness， excellent wear resistance， and chemical stability. However， their intrinsic brittleness makes them susceptible to severe stress concentration under dry friction or boundary lubrication， leading to aggravated wear and premature failure. Surface micro-texturing has been recognized as an effective strategy for regulating interfacial contact states and improving tribological performance. In this study， femtosecond laser processing was employed to fabricate friction-reducing and wear-resistant micro-textures on zirconia （ZrO₂） ceramic surfaces. Finite element simulations， response surface methodology （RSM） optimization， and lubricated friction and wear tests were integrated to systematically elucidate the effects of micro-groove geometric parameters on stress distribution and wear behavior at the ceramic friction interface. The finite element results show that scanning interval， groove depth， and groove orientation significantly influence the maximum equivalent stress distribution on the ceramic surface. Smaller scanning intervals and greater groove depths tend to induce stress concentration， whereas an appropriate groove orientation facilitates the redistribution of contact stress. The RSM analysis indicates that laser energy is the dominant factor governing groove depth and width， while scanning speed exerts an inhibitory effect and the number of scanning passes promotes groove expansion through cumulative ablation. Friction and wear tests demonstrate that， under the optimized parameters， the average friction coefficient of ZrO₂ ceramics is reduced from 0.57 to 0.358 4， corresponding to a reduction of 37.12%， with a wear volume of 0.267×10⁶ μm³. These results demonstrate that femtosecond laser-fabricated micro-groove structures can effectively enhance the friction-reducing and wear-resistant performance of ceramic materials， providing a feasible approach for improving the service reliability of ceramic friction pairs.关键词：ceramic materials;femtosecond laser;micro-groove textures;friction and wear;finite element simulation51|24|0更新时间：2026-06-24
Micro\/Nano Technology and Fine Mechanics
- 摘要：Floating object detection in complex water environments is challenged by false positives， missed detections， and limited multi-scale perception. A lightweight and efficient method based on YOLOv8n is proposed. A Multi-scale Adaptive Context-Enhanced Layer Aggregation Network （MAC-ELAN）， integrating reparameterized convolution （RepConv）， is introduced into the backbone and the deeper layers of the neck. Through cross-layer feature fusion and contextual enhancement， feature discrimination in complex backgrounds is strengthened， while structural reparameterization balances detection accuracy and computational efficiency. In addition， P2 and P6 detection heads are incorporated to improve sensitivity to extremely small and large-scale targets. In the shallow layers of the neck， a lightweight feature fusion module， C2f-FA， is designed by reconstructing the C2f structure with Partial Convolution （PConv） and the EMA attention mechanism. The Wise-IoUv3 loss is further adopted to enhance localization accuracy. Experimental results on a self-constructed dataset demonstrate that， compared with the baseline， the proposed model improves mAP@50 and mAP@50：95 by 4.2% and 3.9%， respectively， with only 1.13M additional parameters. Cross-domain evaluations on the Flow-Img and IWHR-Floater_V1 datasets further verify the effectiveness of the proposed approach in complex water-surface scenarios， achieving a superior balance between detection accuracy and model lightweightness while maintaining low computational complexity.关键词：floater detection;complex water environment;reparameterized convolution;feature fusion;lightweight network55|19|0更新时间：2026-06-24
- 摘要：To address the challenges of small-scale defects， strong textural interference， and diverse defect morphologies in textile inspection， a lightweight and high-accuracy model， YOLOv8s-WRHU， is proposed based on the YOLOv8s framework. The model is designed to enhance robustness and localization accuracy for defect detection in complex-textured fabrics. Systematic improvements are introduced in three key aspects： feature fusion， detail enhancement， and localization optimization. First， to alleviate information loss and fixed weight assignment in multi-scale feature fusion， a learnable weighted feature concatenation module is designed， and a weighted feature pyramid network built upon this module is introduced to adaptively fuse features from different levels. Second， Haar wavelet downsampling is incorporated into the early stage of the backbone network to replace conventional downsampling methods， thereby reducing the loss of high-frequency details. Third， a dynamically weighted unified IoU loss function is employed to improve bounding box regression accuracy. In addition， optimizations of the multi-scale fusion strategy in the neck and the shallow downsampling modules effectively reduce the parameter count and computational complexity of the model. Experimental results on the public Roboflow fabric defect dataset show that， compared with YOLOv8s， the proposed model improves precision， mAP， and recall by 2.9%， 3.4%， and 1.1%， respectively， while reducing the number of parameters by approximately 24.37% and computational cost by 4.4 G. High detection accuracy is maintained with significantly improved inference efficiency. The proposed model achieves a favorable balance between detection accuracy and model complexity， demonstrating strong potential for practical industrial applications.关键词：defect detection;lightweight YOLOv8s;weighted feature pyramid network;Haar wavelet downsampling;unified IoU loss function84|19|0更新时间：2026-06-24
- 摘要：Fault detection in freight trains is essential for ensuring railway transportation safety and improving operational efficiency. However， the high computational cost of vision foundation models remains a major barrier to their direct deployment on resource-constrained trackside equipment. To address this challenge， a lightweight fault detection framework integrating feature knowledge transfer， attention enhancement， and logic-aware distillation is proposed to efficiently transfer prior knowledge from foundation models to deployable networks. Specifically， a fused student network is first constructed using a multi-source pretraining strategy that combines the semantic representation strengths of FastSAM with the robust detection capability of YOLOE. Subsequently， a lightweight feature enhancement module is embedded into the backbone network to improve feature extraction and representation under complex visual conditions. Finally， a logic-aware multi-component distillation strategy is designed to compress the rich knowledge of the teacher model into the student network， thereby significantly improving detection accuracy while maintaining low computational cost. Experimental results on a self-constructed freight train fault dataset demonstrate that the proposed method achieves competitive detection accuracy with substantial reductions in both parameters and floating-point operations， indicating strong potential for real-time deployment on edge devices.关键词：computer vision;fault detection;knowledge distillation;knowledge transfer;real-time detection36|21|0更新时间：2026-06-24
Information Sciences
- 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