2026年 02期
基于改进AlignedReID++的行人重识别方法
Pedestrian Re-identification Method Based on Improved AlignedReID++
宋建辉;马赫遥;赵亚威; SONG Jianhui;MA Heyao;ZHAO Yawei;由于行人姿态变化的多样性、背景环境的复杂性等因素的干扰，会导致行人特征的表达能力不足，进而影响到行人重识别的准确性。本文基于深度学习的AlignedReID++模型进行改进，将DenseNet121网络与AlignedReID++模型的主干网络ResNet50融合，利用两者的优点，增强模型的特征提取能力；通过引入跨维交互注意力模块和正则化项，限制模型的复杂度，增强表达特征能力，提高泛化性能，进而提升重识别能力。将改进模型在Market1501、DukeMTMC数据集上进行验证，实验结果表明，相较于原始模型，平均精度均值在Market1501数据集上提升了5.6个百分点；在DukeMTMC数据集上提升了5.9个百分点，表明了改进后算法的有效性。
The interference factors such as the diversity of pedestrian pose variations and the complexity of background environments can lead to inadequate capability to express pedestrian features, thereby affecting the accuracy of pedestrian re-identification.This paper aims to improve the AlignedReID++ model based on deep learning by integrating the DenseNet121 network with the backbone ResNet50 network of AlignedReID++.Leveraging the strengths of both networks enhances the model's feature extraction capability.By introducing a cross-dimensional interactive attention module and regularization terms, the model's complexity is constrained, its feature representation capability is strengthened, and its generalization performance is improved, thereby boosting re-identification accuracy.Experimental validation on the Market1501 and DukeMTMC datasets demonstrates that, compared to the original model, the improved model achieves increases by 5.6 percentage points in MAP on the Market1501 dataset, and 5.9 percentage points in MAP on the DukeMTMC dataset.These results confirm the effectiveness of the proposed algorithm.
基于MCA与改进ProtoNet的滚动轴承小样本故障诊断方法
Few-Shot Fault Diagnosis Method for Rolling Bearings Based on MCA and Improved ProtoNet
崔琪;吴东升; CUI Qi;WU Dongsheng;针对工业环境中轴承故障数据稀缺、工况多变且常伴随噪声干扰导致特征提取困难、诊断准确率降低的问题，提出一种基于多维协作注意力(MCA)与改进原型网络(ProtoNet)的滚动轴承小样本故障诊断方法DMCA-ProtoNet。首先，采用离散小波变换(DWT)对轴承二维灰度图数据进行去噪和高频分解；其次，将MCA嵌入残差网络ResNet以提升模型的特征提取能力；最后，对ProtoNet中度量准则进行更新，采用马氏距离代替传统的欧氏距离。实验采用德国帕德博恩大学轴承数据集以及东南大学轴承数据集对模型进行验证，结果表明：在小样本变工况下，模型能够保持较高的诊断准确率，识别率可达到95.94%;在强噪声干扰下本模型具有较好的稳定性与准确率，验证了DMCA-ProtoNet模型在小样本变工况下依然具有良好的泛化能力以及抗噪能力。
To address the issues that bearing fault data is scarce in industrial environment, working conditions are changeable and often interfered by noise, which lead to difficulty in feature extraction and low diagnostic accuracy, a few-shot fault diagnosis method for rolling bearings based on multi-channel attention(MCA)and improved prototype network(ProtoNet)(DMCA-ProtoNet)is proposed.Firstly, the Discrete Wavelet Transform(DWT)is used to denoise the two-dimensional grayscale data of the bearing and decompose the high-frequency of it.Secondly, the multi-dimensional collaborative attention mechanism is embedded into ResNet to improve the feature extraction ability of the model.Finally, the metric criterion in ProtoNet is updated to replace the traditional Euclidean distance with the Marhalanobis distance.The results show that the model can maintain a high diagnostic accuracy under few-shot variable working conditions, and the recognition rate can reach 95.94%.The model has good stability and accuracy under strong noise interference, which verifies that the DMCA-ProtoNet model still has good generalization ability and anti-noise ability under few-shot variable conditions.
基于Fast Tube MPC算法的咖啡拉花机械臂轨迹控制方法研究
Research on Latte Art Robotic Arm Trajectory Control Method Based on Fast Tube MPC Algorithm
代勇;秦冠一;陈旭升;周智晨; DAI Yong;QIN Guanyi;CHEN Xusheng;ZHOU zhichen;针对咖啡拉花机械臂在轨迹控制过程中受到扰动(负载变化等)影响导致控制精度下降的问题，提出一种基于模型预测控制(MPC)的快速鲁棒控制策略(fast tube MPC)。该策略由名义fast MPC和辅助滑模控制(SMC)控制律组成。其中，名义fast MPC通过近似预测控制律实现快速在线凸优化，而辅助SMC控制律则构建了鲁棒补偿机制，确保实际系统与名义系统之间的状态误差能够迅速收敛至零邻域。仿真结果表明了所提出的fast tube MPC方案在咖啡拉花机械臂轨迹控制中的有效性。
To address the issue of decreased tracking accuracy of the latte art robotic arm due to unknown disturbances during the trajectory tracking control process, a fast robust control strategy based on model predictive control(MPC)is proposed, namely fast tube MPC.This strategy consists of a nominal fast MPC and an auxiliary sliding mode control(SMC).Among them, the nominal fast MPC achieves rapid online convex optimization through approximate predictive control laws, while the auxiliary SMC control law builds a robust compensation mechanism to ensure that the state tracking error between the actual system and the nominal system can rapidly converge to a zero neighborhood.Finally, the effectiveness of the proposed fast tube MPC scheme in the trajectory tracking simulation experiment of the latte art robotic arm is verified through simulation.
基于SMA的变刚度踝关节外骨骼机器人
Variable Stiffness Ankle Exoskeleton Robot Based on SMA
张佳琦;李翼晓;李敦超; ZHANG Jiaqi;LI Yixiao;LI Dunchao;为提升踝关节外骨骼康复效果及减小执行机构体积，设计一种基于形状记忆合金(shape memory alloys, SMA)的变刚度外骨骼。外骨骼通过SMA调整弹性组件的预压缩实现变刚度，构建驱动单元数学模型、本构模型和相变动力学模型，为踝关节外骨骼的变刚度调控提供可预测的力学参数基准。对踝关节外骨骼进行下蹲和行走实验，结果表明：在相同下蹲速度下，外骨骼弹簧预压缩量从10%提升至20%,可使偏置弹簧动态形变量减少约10 mm,踝关节外骨骼的刚度由0.05 N/m增至0.35 N/m;行走实验中，踝关节外骨骼刚度由0.05 N/m增至0.25 N/m;采用SMA提供刚度调整的驱动力具有能量密度高、可编程、响应速度快等特点，可缩小整体体积并有效辅助肢体平稳运动。
An ankle exoskeleton based on shape memory alloy(SMA)variable stiffness is designed to improve the rehabilitation effect of ankle and reduce the actuator volume.The exoskeleton achieves variable stiffness by adjusting the pre-compression of the elastic components of the SMA,and constructs the mathematical model of the drive unit, the intrinsic model and the phase transition dynamics model, which provide a predictable mechanical parameter benchmark for the variable stiffness regulation of the ankle exoskeleton.The results show that, at the same squatting speed, the pre-compression of the exoskeleton spring is increased from 10% to 20%,which can reduce the dynamic deformation of the bias spring by about 10 mm, and the stiffness of the ankle exoskeleton is increased from 0.05 N·m/deg to 0.35 N·m/deg.In the walking experiment, the stiffness of the ankle exoskeleton is increased from 0.05 N·m/deg to 0.25 N·m/deg.The stiffness-adjusted drive force using SMA has the features of high energy density, programmability, and fast response speed, which reduces the overall volume and effectively assists the limbs to move smoothly.
基于星闪SLE的改进EDF调度算法优化
Optimization of Improved EDF Scheduling Algorithm Based on SparkLink SLE
苗建军;徐雷霆;张东阳;宋仁捷; MIAO Jianjun;XU Leiting;ZHANG Dongyang;SONG Renjie;Unit 32668 of the People's Liberation Army of China;针对最早截止时间优先(EDF)算法在星闪低功耗接入技术(sparkLink low energy, SLE)中存在任务延迟与不稳定的问题，提出增强型最早截止时间优先(enhanced-EDF,EEDF)算法。首先基于SLE物理层特性，定义两种同步信号及动态数据帧结构，为数据的精确调度和高效承载提供基础；其次融合保证预留时隙(guaranteed time slot, GTS)与竞争接入时隙(contention access slot, CAS)机制，对任务进行分区优化，并设立任务可调度性判据，以优先保障关键周期性流量；然后结合量化误差分析，建立利用率和容量模型，用以评估系统负载和资源需求关系；最后通过Matlab进行仿真验证。实验结果表明，EEDF算法与EDF算法相比，将时隙利用率提升了19.3%,任务完成率提升了23.7%,平均响应时间减少了0.15 s,在高负载和动态环境中表现出更高的实时和稳定性，显著增强了SLE网络的任务调度能力。
To address the issues of task delay and instability of the earliest deadline first(EDF)algorithm in sparklink low energy(SLE),an enhanced-EDF(EEDF)algorithm is proposed.Firstly, based on the physical layer characteristics of SLE,two types of synchronization signals and a dynamic data frame structure are defined, providing a basis for accurate data scheduling and efficient data carrying.Secondly, the guaranteed time slot(GTS)and contention access slot(CAS)mechanisms are integrated to optimize task partitioning, and a task schedulability criterion is established to prioritize the guarantee of critical periodic traffic.Thirdly, combined with quantization error analysis, utilization and capacity models are established to evaluate the relationship between system load and resource requirements.Finally, MATLAB is used for simulation verification.The results show that compared with the EDF algorithm, the EEDF algorithm increases the time slot utilization by 19.3%,the task completion rate by 23.7%,and reduces the average response time by 0.15 s.It demonstrates higher real-time performance and stability in high-load and dynamic environments, enhancing the task scheduling ability of the SLE network.
基于DUHG-YOLO的教室学生行为检测
Classroom Student Behavior Detection Based on DUHG-YOLO
魏英姿;于聚壮;张航; WEI Yingzi;YU Juzhuang;ZHANG Hang;针对教室环境中学生行为检测存在人员密集、遮挡、模糊以及前后排目标尺度变化显著等问题，提出名为DUHG-YOLO的先进目标检测模型。模型以YOLOv11框架为基础，首先设计C3k2＿Dual模块，既在主干网络的C3k2中引入双重卷积模块(DualConv),以增强模型特征提取能力，减少计算冗余并提高检测精度。其次提出一种多尺度特征融合与注意力增强的网络框架ZSH,通过引入混合注意力机制(HybridAttention)和双线性插值(Bilinear)增强特征融合效果，提升特征表示能力。最后使用广义交并比损失函数(GIoU)优化非重叠目标的梯度更新，提高模型的检测精度。实验结果表明，相较YOLOv11n, DUHG-YOLO在StuDataset数据集上精确率、召回率、平均精度均值分别提升1.7%、2.6%、2.1%,可以有效应用于教室学生行为检测任务。
To address the challenges of dense crowds, occlusions, blurring, and significant scale variations of targets between front and back rows in student behavior detection within classroom environments, this paper proposes an advanced object detection model named DUHG-YOLO.Based on the YOLOv11 framework, the model first introduces the C3k2＿Dual module, which incorporates a dual convolution(DualConv) into the C3k2 block of the backbone network to enhance feature extraction capability, reduce computational redundancy, and improve detection accuracy.Then, a multi-scale feature fusion and attention-enhanced ZSH network framework is proposed, integrating a Hybrid Attention Mechanism(HybridAttention) and Bilinear Interpolation(Bilinear) to strengthen feature fusion and improve feature representation.Finally, the Generalized Intersection over Union(GIoU) loss function is employed to optimize gradient updates for non-overlapping targets, further enhancing detection accuracy.Experimental results demonstrate that, compared to YOLOv11n, DUHG-YOLO achieves improvements of 1.7% in precision, 2.6% in recall, and 2.1% in mean average precision(mAP) on the StuDataset, proving its effectiveness for classroom student behavior detection tasks.
基于SO-YOLO的遥感图像目标检测算法
Remote Sensing Images Target Detection Algorithm based on SO-YOLO
王佳;李芳; WANG Jia;LI Fang;针对遥感图像目标检测中存在图像分辨率低、小目标特征信息不足以及检测难度大等问题，提出一种基于改进YOLO11的遥感图像目标检测算法SO-YOLO。首先，设计一个新的卷积神经网络(CNN)构建块SDOD-Conv,由空间到深度转换层和全维动态卷积组成，代替主干网络中的跨步卷积和池化层，加强特征提取，在特征提取过程中避免细粒度信息损失；其次，在颈部网络中引入空间和通道重建卷积(SCConv),压缩特征之间的空间和通道冗余，减少冗余计算并促进代表性特征学习；最后，采用Inner-IoU损失函数作为回归损失，通过引入比例因子的辅助边界框计算IoU损失，获得更快、更准确的回归结果。在HRSC2016数据集和DOTA数据集上的实验结果表明，相较于YOLO11,改进后算法的平均精度均值分别提高了2.3%和2.0%,表明改进算法具有良好的检测性能。
To solve the problems of low image resolution and insufficient feature information of small targets and difficulty of target detection in remote sensing image, a remote sensing image detection method based on SO-YOLO is proposed on the basis of YOLO11 model.First, a new convolutional neural network(CNN)building block SDOD-Conv is designed, which consists of a space-to-depth layer and a full-dimensional dynamic convolution, replacing each stride convolution layer and pooling layer in the backbone network, to enhance the feature extraction and avoid the loss of fine-grained information during the feature extraction process.Second, Introducing Spatial and Channel Reconstruction Convolution(SCConv)into the neck network to compress spatial and channel redundancies in features, redundant computation is reduced and discriminative feature learning is enhanced.Finally, the Inner-IoU loss function is used as the regression loss, and the IoU loss is computed by introducing an auxiliary bounding box for the scale factor to obtain faster and more efficient regression results.The improved algorithm is verified on the HRSC2016 data set and DOTA data set, which are improved by 2.1% and 1.5% respectively, compared to the average accuracy of YOLO11,and demonstrates good detection performance.
基于多模态输入与改进Transformer的UWB非视距信号识别
UWB Non-Line-of-Sight Signal Recognition Based on Multi-Modal Input and Improved Transformer
徐佳杰;王晓青;周启航; XU Jiajie;WANG Xiaoqing;ZHOU Qihang;超宽带(UWB)技术在定位过程中易受非视距(NLOS)信号干扰，导致定位精度下降。因此，在使用UWB信号进行位置解算前必须对信号进行识别、分类。针对UWB的非视距信号识别问题，提出一种基于多模态输入与改进Transformer结合的方法。该方法通过融合三类具有互补性多模态特征提升对UWB信号的表征能力：一是保留原始信道脉冲响应(CIR)时间序列，捕捉其时域动态信息；二是提取一些额外统计特征，用于表征信道物理特性；三是将CIR信号转换为图像，挖掘潜在特征。此外，对Transformer架构进行优化，引入探针稀疏注意力机制添加相对位置编码，精确捕捉序列中的远程依赖关系。同时，采用自适应门控模块结合空间门控和通道注意力机制，有效提取并融合多模态特征，促进特征之间的互补性结合。经实验验证，该方法的识别准确率达到89.18%、精确率达到91.46%,在处理UWB非视距信号识别问题上相较于其他方法表现出显著优势。
Ultra wide band(UWB) technology is susceptible to non-line-of-sight(NLOS) signal interference during the positioning process, which can significantly degrade the accuracy of location estimation.Therefore, it is essential to classify and identify UWB signals before performing position calculations.To address the challenge of NLOS signal identification in UWB systems, this paper proposes a method based on multi-modal input combined with an improved Transformer architecture.This approach enhances the representation capability of UWB signals by integrating three complementary multi-modal features: first, preserving the original channel impulse response(CIR)time series to capture temporal dynamic information; second, extracting additional statistical features to characterize the physical properties of the channel; third, transforming the CIR signals into images to uncover potential spatial patterns.Furthermore, the Transformer architecture is optimized by introducing a probe-based sparse attention mechanism with relative positional encoding, enabling precise modeling of long-range dependencies within the sequence.Meanwhile, an adaptive gating module is employed, combining spatial gating and channel attention mechanisms to effectively extract and fuse multi-modal features, thereby promoting complementary integration across different feature modalities.Experimental results demonstrate that the proposed method achieves an accuracy of 89.18% and a precision rate of 91.46%,showing significant performance advantages over existing methods in addressing the problem of NLOS signal identification in UWB systems.
基于轻量化YOLOv11的道路目标检测
Road Target Detection Based on Lightweight YOLOv11
张熙函;刘军;齐向晶; ZHANG Xihan;LIU Jun;QI Xiangjing;在复杂道路环境下，自动驾驶系统在目标检测过程中仍然面临误检与漏检的挑战，难以兼顾精度和实时性。本文提出一种基于YOLOv11的轻量化改进道路目标检测方法。首先应用深度可分离卷积(SPD-Conv)替换部分传统卷积，以增强特征之间的关联性，提高卷积操作的表达能力。其次对C2PSA结构中的PSABlock模块进行优化，引入结合空间与通道的注意力机制SCSA,实现更全面的特征融合，有效捕捉多语义信息，降低关键信息遗漏的风险。最后将传统上采样模块替换为动态上采样(DySample)模块，既保持检测精度，又提高整体推理速度，实现模型的轻量化。实验结果表明，该方法在KITTI数据集上相比于原模型参数量降低18.9%,平均精度均值提升1.04%,推理速度提升16.9%,有效提升了检测性能与推理效率。
In complex road environments, automatic driving systems still face the challenges of misdetection and omission in the target detection process, which makes it difficult to balance accuracy and real-time performance.A lightweight improved road target detection method based on YOLOv11 is proposed.Firstly, a depth-separable convolution(SPD-Conv)is applied to replace part of the traditional convolution to enhance the correlation between features and improve the expressiveness of the convolution operation.Secondly, the PSABlock module in the C2PSA structure is optimized, and the SCSA attention mechanism combining space and channel is introduced to achieve more comprehensive feature fusion, effectively capture multi-semantic information, and reduce the risk of omitting key information.Finally, the traditional up-sampling module is replaced by the DySample module, which maintains the detection accuracy and improves the overall inference speed, realizing the lightweight of the model.The experimental results show that on the KITTI dataset, the method reduces the number of parameters of the original model by 18.9%,improves the average accuracy by 1.04%,and increases the inference speed by 16.9%,which effectively raises the detection performance and inference efficiency.
胆管支架材料的研究现状
Research Status of Biliary Stent Materials
邓乔元;董瑞旭;文峰; DENG Qiaoyuan;DONG Ruixu;WEN Feng;胆管支架植入是治疗胆石症以及胆管梗阻的重要手段之一，也是解决肝门部胆管癌症等恶性肿瘤引起的胆管堵塞问题的重要治疗措施。植入类支架的力学性能、生物安全性等需满足较高标准的要求，目前市场上的胆管支架主要是不可降解支架，分为金属支架与高分子支架，两类支架各有优劣，但都面临着因其不可降解而引起术后并发症或微生物聚集引发炎症等问题。因此，包括镁合金支架在内的可降解胆管支架与纳米银表面涂层等受到人们的关注，各种新型胆管支架及表面涂层陆续出现并逐渐投入临床使用。本文主要介绍了胆管支架的研究现状，分析比较了不同胆管支架材料的优劣，总结了部分新型胆管支架及表面涂层的研究进展，并对胆管支架及其表面处理的未来发展进行了展望。
Biliary stent implantation is one of the important methods to treat cholelithiasis, biliary stricture and biliary obstruction, and also an important measure to solve the obstruction of bile duct caused by malignant tumor including cancer of hilar bile duct.Implantable stents are required to adhere to high standards in terms of mechanical performance and biological safety.Currently, the majority of biliary stents commercially available are non-degradable.They can be broadly classified into metal stents and polymer stents, each presenting a distinct set of advantages and limitations.And both of them are facing the problems of postoperative complications caused by non-degradation of stents and inflammation caused by microbial aggregation.Therefore, biodegradable biliary stent including magnesium alloy stent and nano-silver coating have attracted much attention, various new types of biliary stents and surface coatings are emerging and being gradually introduced into clinical practice.In this paper, the research status of biliary stents is introduced, the advantages and disadvantages of different material biliary stents are analyzed and compared, and the research progress of some new types of biliary stent and their surface coating is summarized, the future development of biliary stents and their surface treatment is prospected.