最新刊期
卷 33 ， 期 16 ， 2025
- “在三维重构领域，研究者设计了激光雷达与相机融合系统，通过垂直扫描获取空间三维数据，结合深度相机提升重构质量，有效提升了三维重构的精度和完整性。”摘要：Single-layer LIDAR offers advantages such as simple structure， low cost， and low power consumption； however， its sensing capability is limited to planar scanning， rendering it incapable of independently achieving three-dimensional 3D reconstruction. Although visual sensors provide rich image information， they suffer from limitations in geometric measurement accuracy and robustness against environmental interference. To address the inability of single-layer LIDAR to directly acquire 3D spatial information and the low reconstruction accuracy of single-sensor systems， this paper presented a LIDAR–camera fusion-based 3D reconstruction system. The system employed a lead screw and roller mechanism to drive the single-line LIDAR in vertical motion， enabling spatial 3D data acquisition. Simultaneously， the RGB-D camera captured color and texture information to enhance reconstruction quality. The system architecture and scanning measurement principles were first established， followed by a detailed analysis of the 3D reconstruction pipeline. Intrinsic and extrinsic calibration between the LIDAR and the camera was performed to enable accurate point cloud registration and fusion， significantly improving the precision and completeness of the reconstruction. A simulated tunnel-like environment was constructed for validation. Using the calibrated system， point cloud coordinates were computed frame-by-frame for the reconstruction of 11 target objects. Experimental results show a registration error of 6.8 mm， with fused point cloud deviations not exceeding 5.5 mm within a 500 mm measurement range. Furthermore， a large-scale indoor floor reconstruction experiment demonstrates an overall reconstruction error within 12 mm over a 2 600 mm measurement range.关键词：lidar;depth camera;vertical motion;Joint calibration;3D reconstruction0|0|0更新时间：2025-09-28
- “在双目视觉领域，研究人员提出了基于多尺度注意力机制的TransUNet网络，有效提升了连铸坯定位与测量的效率和精度。”摘要：Aiming at the problems of low detection efficiency of traditional binocular-vision feature-detection algorithms， as well as the insufficient attention to globally important features and the excessive parameter count of most network models， a method of continuous-casting-billet localization and measurement based on TransUNet binocular vision with a multiscale attention mechanism was proposed. Firstly， left and right images of continuous-casting billets were collected with a calibrated parallel binocular camera to build a dataset. Subsequently， taking TransUNet as the backbone， an improved Transformer layer was introduced to extract global context information； a Global Spatial Group Attention （GSGA） module was appended after every decoder block to enhance focus on globally salient features through a grouped multiscale attention mechanism； and a Convolutional Block Attention Module （CBAM） was inserted after each encoder-decoder skip connection and bilinear interpolation to boost key-point recognition by combining spatial and channel attention. Finally， 3-D coordinate reconstruction and distance measurement were performed on the network’s key-point coordinates by leveraging binocular-vision principles. The experimental results show that compared with the Transformer model， the root-mean-square error and normalized error are reduced by 33.8% and 36.83%， the number of parameters and floating-point operations are reduced by 10.58% and 8.21%， and the single-batch inference time is shortened by 32.30%. In 3D ranging， the relative error of measurement reaches 0.137%， which is significantly better than the traditional feature detection algorithm and meets the binocular vision localization and measurement requirements.关键词：binocular vision;TransUnet;keypoints detection;attention mechanism2|0|0更新时间：2025-09-28
- “航空相机检焦镜面型变化问题研究取得进展，建立了理论模型并验证，为提高成像质量提供指导。”摘要：For the issue of focus failure caused by changes in the shape of the focusing mirror during the autocollimation focusing process of aerial cameras， a corresponding theoretical model was provided. By establishing the relationship between the difference in light intensity received by the sensor and the focusing curve， the impact of different focusing mirror shape accuracies on the distribution of imaging light intensity was calculated， and the effects of various aberrations on the focusing curve were analyzed. Combined with an actual camera model， simulations were conducted using Zemax software in non-sequential mode. By adjusting the temperature to control the shape of the focusing mirror， focusing experiments were carried out under different shapes， verifying the theoretical and simulation results. The results show that when the surface shape RMS of the focusing mirror reaches 0.048 3λ， the peak position of the focusing curve exceeds the half focal depth； when the surface shape RMS reaches 0.082 1λ， the focusing curve exhibits multiple peaks. This has guiding significance for the error allocation， design， and control of the optical autocollimation focusing mirror， as well as for improving the imaging quality of aerial cameras.关键词：aerial camera;autocollimation;focus inspection;curve fitting;surface figure error2|0|0更新时间：2025-09-28
- “在化工产品生产领域，专家建立了计数率与高压纹波的数学模型，采用自适应卡尔曼滤波算法抑制纹波，提高了密度检测精度。”摘要：In the production process of chemical products， when utilizing radiation from a radioactive source to measure the density of a solution by penetrating a pipeline， the count rate output from the photomultiplier tube is susceptible to degradation due to high-voltage ripple noise. This degradation leads to noticeable attenuation in the count rate， thereby reducing the density measurement accuracy. A series of methods were proposed to mitigate the impact of ripple noise on the count rate and compensate for the attenuation caused by the ripple. First， based on experimental data under various ripple conditions， a mathematical model was established to characterize the relationship between the count rate and high-voltage ripple. Analysis of this model revealed that reducing the ripple's amplitude and frequency effectively diminishes its influence on the count rate. Subsequently， an adaptive Kalman filtering algorithm was employed to suppress the ripple noise. By dynamically adjusting the process and measurement noise parameters， the algorithm enabled adaptive compensation of the count rate. Experimental validation was carried out using a 40% ethanol solution， and comparative analysis was performed on the count rate and density measurement results before and after compensation. The results demonstrate that， compared to the uncorrected state， the pulse count loss rate decreased to 3.54%–5.46% after compensation. Furthermore， the maximum deviation between the measured and actual solution density was reduced to 0.019 g/cm³， with overall density measurement accuracy improved by up to 1.51%.关键词：photomultiplier tube;ripple noise;count rate;kalman filtering;adaptive compensation2|0|0更新时间：2025-09-28
- “在三维场景扫描领域，专家提出了基于平面特征的多线激光雷达旋转平台外参标定方法，通过改进粒子群算法优化求解，验证了方法的准确性与有效性。”摘要：To achieve full-view point cloud data acquisition of three-dimensional （3D） scenes， this paper proposed a multi-beam LiDAR rotational scanning system. The quality of the 3D point cloud reconstructed by this system was highly dependent on the pose transformation relationship between the LiDAR and the rotating platform. However， existing calibration methods were mainly designed for single-line LiDAR applications and faced issues such as overly simplistic calibration models and severe parameter coupling. Therefore， this paper presented an external parameter calibration method for multi-beam LiDAR rotating platforms based on planar features. First， a mathematical model for calibrating the target parameters was constructed based on plane features by integrating the RAndom SAmple Consensus （RANSAC） algorithm. Then， an improved Particle Swarm Optimization （PSO） algorithm， which introduced a linearly decreasing inertia weight coefficient to replace the fixed inertia weight， was employed to optimize the constructed mathematical model. Finally， the accuracy and effectiveness of the proposed calibration strategy were validated through point cloud thickness analysis and 3D scene reconstruction evaluation in both simulation and real-world scenarios. Experimental results show that the point cloud thickness is reduced from 7.668 1 cm before calibration to 4.039 0 cm after calibration， indicating that the point cloud distribution reconstructed using the proposed calibration method is more uniform. The comparison of point cloud distribution in reconstructed scenes before and after calibration further verifies the effectiveness and reliability of the proposed method.关键词：multi-beam LiDAR;LiDAR calibration;external calibration;3D reconstruction;point cloud2|0|0更新时间：2025-09-28
Modern Applied Optics
- “在机器人领域，研究者提出了一种高精度手眼标定方法，有效降低深度误差对标定精度的影响，提高适用性。”摘要：Point cloud sensors usually have depth errors， and the existing hand-eye calibration methods rarely take this factor into account， which leads to the applicability of hand-eye matrices at different scales. A high-precision hand-eye calibration method based on alternating least squares optimization was proposed. First， by fitting the plane and voxels of the calibration plate to uniformly downsample， the point cloud data of the near， middle and far scales were preprocessed， and the density of the point cloud of the calibration plate at different scales was unified； Secondly， the GICP （Generalized-ICP）algorithm was used to obtain the registration matrix of the point cloud， and the initial value of the hand-eye matrix of each scale was solved， the hand-eye matrix of one scale was fixed alternately， and the least-squares problem of rotation translation error is constructed to optimize the registration parameters of the remaining scales. Finally， the Tsai-Lenz method was recalibrated based on the optimized registration matrix of each scale， and the hand-eye matrix to compensate for the depth error was obtained. The comparative experimental results show that the average error of the rotation matrix is 0.144 8°， the average error of the translation vector is 0.633 5 mm， and the root mean square error of the point cloud registration using the hand-eye matrix is 0.276 9 mm. Compared with the calibration ball method， the point cloud registration error of the near focus distance is reduced by 19% on average， and the point cloud registration error of the far focus distance is reduced by 16% on average. The method in this paper can effectively reduce the influence of depth error on calibration accuracy and improve applicability.关键词：depth error;hand-eye calibration;structured light;point cloud registration2|0|0更新时间：2025-09-28
- “在随钻测量领域，专家提出了基于磁惯性蜜獾算法的加速度计在线补偿方法，有效提升了井斜角测量精度，误差绝对值保持在1.43°以内。”摘要：To address the issue of insufficient measurement accuracy in downhole measurement-while-drilling （MWD） accelerometers， this paper proposed an online compensation method for MWD accelerometers based on the Magnetic-Inertial Honey Badger Algorithm （MIHBA）. Firstly， an error compensation model was established by analyzing the characteristics of MWD accelerometers， with error parameters formulated as a solution vector. Secondly， leveraging the constant magnitude of gravitational acceleration and its direction， a target function for accelerometer errors was designed. Error constraint conditions were established based on the relationship between MWD gyroscope， accelerometer， and magnetometer outputs. The MIHBA was developed from the Honey Badger Algorithm （HBA）， incorporating initial values derived from historical accelerometer data and processed outcomes. During the exploration phase， a trust-region mutation strategy was introduced to enhance population diversity， considering the recursive relationships among MEMS sensors. In the exploitation phase， reflection operations inspired by nonlinear simplex methods were applied to poorly adapted individuals， improving convergence precision. Vibration table tests and actual drilling experiments demonstrate that the MIHBA-compensated borehole inclination error accuracy improves by approximately 64.8%， with the error magnitude maintained within 1.43°. The algorithm effectively removes noise from magnetometer signals， significantly enhancing the overall accuracy of MWD accelerometer measurements.关键词：measurement while drilling;accelerometer;honey badger algorithm;error compensation4|0|0更新时间：2025-09-28
- “在远心成像系统高精度标定领域，专家提出了结合亚像素边缘检测与光照误差补偿的相机标定方法，有效提升了测量精度。”摘要：To address the high-precision calibration demands of telecentric imaging systems and the inadequacy of traditional pinhole models， this paper proposed a high-accuracy camera calibration method that integrated subpixel edge detection with illumination error compensation. First， an improved Gaussian integral curve fitting algorithm extracted subpixel edges of circular calibration targets， with center coordinates precisely determined via weighted iterative least squares. Second， a system magnification equation was derived based on the telecentric model and rotation matrix orthogonality to resolve extrinsic parameter sign ambiguities， accompanied by a disambiguation strategy. Subsequently， radial， tangential， and composite distortion models were constructed and evaluated for characterizing telecentric lens distortion. Distortion parameters were accurately estimated through Levenberg-Marquardt nonlinear optimization. An illumination error compensation model dynamically updated correction parameters within the calibration pipeline to eliminate edge offset errors. Experimental results show the calibration method achieves an optimal reprojection error of 0.059 pixels， with measurement errors within 1.6 μm for CMM probe ball diameter under 15-28 light intensity levels after compensation， validating its effectiveness.关键词：telecentric lens calibration;subpixel edge detection;weighted iterative circle fitting;compensation for light intensity errors2|0|0更新时间：2025-09-28
- “在精密装配领域，研究者研制了机械臂辅助系统，开发了小型化吸附装置，提出了视觉检测方法，显著提高了装配精度和效率，降低了成本。”摘要：Tiny optoelectronic components serve as the core elements in some high-performance sensors， where their assembly precision critically determines device performance. To address limitations in current assembly methods， such as complex configurations， high costs， and poor scalability， herein the research on the precision assembly method of tiny optoelectronic sensing components were performed. The primary contributions of this work are summarized as follows： First， a miniaturized parallel adsorption device and a robotic manipulator-assisted precision assembly system were developed. Second， considering the dual requirements of efficiency and accuracy， a two-step visual alignment method was established. This method combines "state freezing" with "marker-guided" referencing to determine the spatial relationship between coordinate systems. Finally， the assembly experiments involving miniature optoelectronic components were conducted using the developed system to validate the proposed method. Experimental results show that， by applying the robotic arm-assisted assembly method herein proposed， the assembly accuracy can be improved to the 10 μm level. This breaks through the limitation of the robotic arm's own repeat positioning accuracy， enabling the system to balance efficiency and precision while significantly reducing the complexity and cost of the equipment. We believe that it can be widely applied to various precision operations.关键词：precision assembly;robotic arm;visual measurement;micro-gripper2|0|0更新时间：2025-09-28
Micro/Nano Technology and Fine Mechanics
- “在三维点云配准领域，研究者提出了高鉴别强鲁棒的多视图几何分布特征描述符MGDS，为准确配准提供解决方案。”摘要：To address the problem of incomplete description of local surface geometric features by existing handcrafted descriptors， this paper proposed a high-discriminative and robust multi-view geometric distribution signature （MGDS）. First， a local reference frame （LRF） was constructed based on the keypoints and their neighboring points. The local surface was voxelized. The centroid distribution of the 3D voxels， the contour features of 2D sectors， the point density distribution of 2D grid， and the depth fluctuation of the surface were calculated to generate the geometric feature descriptor. Next， the local surface was rotated multiple times based on the LRF to generate new shape representations. The rotated surfaces were encoded using centroid， contour point， density， and z-value fluctuation information. By capturing these geometric feature descriptors from multiple viewpoints and concatenating them into a single feature vector， the final multi-view geometric distribution signature （MGDS） was obtained. Experiments are conducted on four datasets： RandomView， SpaceTime， Kinect， and B3R with different Gaussian noise and grid resolutions. The proposed MGDS descriptor is compared with ten existing descriptors. Compared to other descriptors， MGDS descriptors outperform existing local feature descriptors. Experimental results indicate that the proposed MGDS descriptor exhibits good descriptiveness and robustness， making it suitable for accurate registration of 3D point cloud.关键词：point cloud;local feature descriptor;multi-view features;multi-feature fusion;feature matching2|0|0更新时间：2025-09-28
- “在点云数据处理领域，专家提出了一种几何特征保持的层次化点云去噪算法，有效提高了去噪精度。”摘要：With the rapid development of 3D laser scanning technology， point cloud data had been widely applied in fields such as autonomous driving， 3D modeling， medical research， reverse engineering， and rural revitalization. However， due to the influence of instrument performance， surrounding environment， and the characteristics of the scanned target itself， the point cloud data obtained by scanning often contains a large amount of noise， which seriously affected the accuracy of subsequent point cloud processing. Therefore， it was necessary to denoise it. A hierarchical point cloud denoising with geometric feature preservation was proposed to address the issues of weak sensitivity to parameters， high computational complexity， and poor preservation of geometric features in traditional filtering algorithms. Firstly， the algorithm introduced point cloud density features into the radius filtering algorithm to improve initial parameter selection and achieve large-scale noise removal； Then， using KD tree （K-Dimensional Tree） to optimize the Density Based Spatial Clustering of Applications with Noise （DBSCAN） algorithm， and combining corner features to adaptively select the parameters of the DBSCAN algorithm， the point cloud data was divided into effective clusters， fuzzy clusters， and noise clusters to remove noise clusters； Finally， the distance consensus evaluation algorithm was used to determine the fuzzy clusters， and the distance between the fuzzy points and the point cloud fitting surface is calculated to determine whether they were noise points， in order to remove small-scale noise from the point cloud. The experiment used a public point cloud dataset and field collected rural point cloud data to validate the proposed algorithm. The results showed that compared with DBSCAN algorithm， improved forest denoising method， geometric feature preservation denoising method， improved density clustering denoising method， and multi feature grid denoising method， the proposed algorithm had better preservation of sharp geometric features， and the denoising accuracy was improved by about 43%， 27%， 29%， 21%， and 9%， respectively. This algorithm can effectively maintain geometric features while improving denoising accuracy， making it an effective point cloud denoising algorithm.关键词：Point cloud denoising;radius filtering;distance consensus assessment;density;corner point2|0|0更新时间：2025-09-28
- “在侧扫描声呐图像分类领域，专家提出了一种高质量伪图像生成方法，有效提升了分类任务的准确率，为水下环境识别研究开辟了新方向。”摘要：The existing side-scan sonar （SSS） image classification network faced a serious issue. It suffered from a shortage of high-quality SSS training samples. This led to poor performance in classifying specific targets outside the training data. A high-quality pseudo side-scan sonar image generation method was proposed based on feature weakening and image patches. Firstly， a feature weakening module was designed to suppress irrelevant optical content features， such as color and texture. By weakening specific domain features in optical images， it helped adjust pseudo-image target features. This made them resemble acoustic image features more closely. Secondly， an image patch-based style transfer pseudo-image generation method was proposed. This method achieved effective fusion of optical content and acoustic style features. It generated original pseudo-SSS images with optical content features. Finally， a fast guided filtering module was introduced. It enhanced and guided the original pseudo SSS images， which generated high-quality pseudo-SSS images closer to the real underwater acoustic imaging environment. Then， a large number of high-quality pseudo images were generated. These images expanded the training samples of the classification network. This improved the accuracy of SSS image classification tasks in real underwater environments. The results of comparative experiments are presented. The proposed method achieves the best average values across various stylization evaluation indicators. Compared with suboptimal methods， the proposed method achieved significant performance gains. The average values of style loss， LPIPS， content loss， and SSIM improved by 31.60%， 3.52%， 15.18%， and 17.55%， respectively. On the grayscale KLSG dataset， the classification task achieved a global accuracy of 86.35%. The average accuracy reached 78.54%.关键词：side-scan sonar image;feature weakening;style transfer;image generation;zero-shot image classification2|0|0更新时间：2025-09-28
- “在黑色素瘤图像分类领域，研究人员提出了一种改进的Vision Transformer模型，通过动态特征筛选和对比学习，显著提升了分类精度和诊断效率。”摘要：To address the issues of feature redundancy and insufficient generalization in Vision Transformer （ViT） models for melanoma image classification， we proposed an enhanced model that integrated dynamic feature selection and contrastive learning. This approach aimed to improve classification accuracy and clinical diagnostic efficiency. Specifically， a dynamic feature selection module was introduced， which adaptively enhanced key features while suppressing redundant information using a learnable weight matrix. Additionally， the InfoNCE contrastive loss function was incorporated into a multi-objective optimization framework alongside cross-entropy loss， thereby improving the discrimination of inter-class features. Furthermore， a feature importance guidance mechanism was embedded within the multi-head self-attention mechanism to achieve collaborative modeling of local details and global semantics. Experimental results on the ISIC2018 and ISIC2019 datasets demonstrate that the improved model achieves classification accuracies of 83.27% and 80.17%， respectively， surpassing the baseline ViT by 1.83% and 0.49%. Ablation studies confirm that the dynamic feature selection module reduces computational redundancy by 18.7%， while contrastive learning increases intra-class feature similarity by 23.6%. The proposed method significantly enhances the recognition capabilities of the ViT model for melanoma， offering superior classification accuracy and robustness compared to mainstream models. It provides a high-precision， low-redundancy automated solution for early skin cancer diagnosis， demonstrating significant clinical practical value.关键词：image classification;feature selection;InfoNCE loss;vision transformer model2|0|0更新时间：2025-09-28
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