Lidar-based 3D object detection achieves superior performance. However, the unevenly distributed point clouds on foreground objects can weaken their geometric representation. Moreover, far-away objects typically have very few points, which further impairs detection performance. In this paper, we present a novel framework PUDet (Point Cloud Upsampling 3D Detector), which integrates generative models into discriminative detectors. We leverage a point cloud upsampling network with prior knowledge to enhance geometric details of foreground objects, aiding the detector in achieving more accurate prediction. PUDet incorporates two key modules: LDEM (Local Distribution Enhancement Module) for nearby objects, which optimizes point distribution while minimizing computational costs, and DDAM (Distant Density Augmentation Module) for distant objects, which increases point density to better delineate object contours. To validate the optimization of geometric contours, we conducted experiments comparing uniform loss before and after enhancement for both nearby and distant objects, demonstrating the efficacy of LDEM and DDAM. We also display the attention maps on object point clouds, explaining the observed accuracy gains. Experimental results on the KITTI testing set show that our framework improves the baseline CT3D by 1.84 mAP, confirming the effectiveness of PUDet. Code will be available at https://github.com/bellamyhsu/PUDet/tree/main.
The temperature drift problem of micro electromagnetic force weighing sensors mainly stems from the thermal expansion and contraction of the mechanical structure, the temperature drift of circuit components, and the change in the magnetic induction intensity of the permanent magnet. For a sensor with a range of 200 g and a graduation value of 0.1 mg, the study analyzed the temperature drift characteristics of the mechanical lever transmission ratio, the voltage reference of the driving circuit, the sampling resistor, and the permanent magnet through the mathematical modeling. The main influencing factors and the installation position of the temperature-compensated sensor were determined. Through a linear heating test (record the indication drift every 10 degrees C), a temperature drift compensation function was established using quadratic fitting, and a dynamic compensation method based on the zero reference point, the half-scale reference point, and proportional following the maximum scale reference point in the interval was proposed. This method enables the influence of the mechanical and circuit parts to simultaneously affect the starting reference point of the sensor in the software, ensuring that the length of the sensor's measurement range interval is precisely symmetrical with the real-time temperature. In addition, the concept of dynamic compensation sensitivity was introduced, and the matching relationship between the compensation amount and the graduation value was adjusted in real time to improve the compensation accuracy. Experiments show that this method achieves dynamic temperature compensation within the range of 5 degrees C- 35 degrees C, with the absolute value of the compensation error being less than 0.5 mg, significantly improving the sensor's adaptability in environments with large temperature fluctuations.
In the long-term use process, the performance of battery continues to decline. The capacity degeneration prediction of battery is crucial, which can effectively avoid some security risks. At present, most of the methods using fuzzy systems to predict battery capacity degeneration are based on historical capacity data, and these methods do not consider the variable charging and discharging conditions. Therefore, an improved adaptive neural fuzzy inference system (ANFIS) applied to random operating conditions is proposed in this paper, which is more suitable for practical application. Features are extracted from raw data and the initial system structure is determined by the correlation coefficients of the features. The fuzzy cluster center is obtained and optimized by fuzzy c-means (FCM) and adaptive particle filter. What's more, the activation mechanism is applied to reduce the number of fuzzy rules. The randomized battery dataset of National Aeronautics and Space Administration (NASA), the temperature-varying dataset and the current-varying dataset of the Center for Advanced Life Cycle Engineering (CALCE) laboratory are used to demonstrate the effectiveness of the proposed system, with the RMSE of 3.73 %, 4.09 % and 3.48 % respectively. Compared with the existing methods, the proposed system has higher accuracy and interpretability relatively.
To address the issue of low defect detection accuracy in IC devices due to insufficient contrast under either visible light or infrared conditions alone,this paper introduces a multi-spectral fusion ap-proach.Initially,to overcome scale inconsistency and contrast inversion challenges during IC device image registration,we enhance the ORB(Oriented FAST and Rotated BRIEF)algorithm with a Laplacian pyra-mid and feature descriptor recombination strategy.Following image registration,we propose the NSST_VP image fusion method,which processes the infrared and visible images'low and high frequency subbands through Non-Subsample Shearlet Transform(NSST).For fusion,the low frequency subband us-es a visual significance map(VSM)weighted rule,and the high frequency subband employs a PA-Pulse Coupled Neural Network(PA-PCNN)decision rule,with the final image produced by reversing the NSST.The fused image is then analyzed using the YOLOv8s model.Experimental findings reveal an 87.8%average accuracy with the improved ORB registration,marking a 62%enhancement over the stan-dard ORB.The NSST_VP fusion algorithm significantly boosts both subjective and objective metrics,achieving an mAP of 83.15%-surpassing single light mode detections by 22.97%and 28.31%,and outper-forming Dual-Tree Complex Wavelet,Non-Subsampled Contourlet,and Curvelet Transform fusion meth-ods by 13.14%,15.01%,and 20.35%,respectively.
Hyperspectral image has rich spectral information and can accurately reflect the real condition of ground objects, so it is widely used. However, there are huge challenges in hyperspectral image, such as few training samples, the phenomena of “the foreign matter same spectrum” and “the same thingdifferent spectrum” can not be distinguished. The hyperspectral image classification method based on convolutional neural network can solve the above problems and achieve good classification performance. Most classification methods based on convolutional neural network only extract the deep semantic information at the end of the network, ignoring the shallow details. At the same time, Methods of this type increase the receptive field by designing deep network structure, which is prone to gradient vanishing. To solve the above two problems, a multiscale dilated dense network is proposed for hyperspectral image classification. Firstly, hyperspectral data cubes with different neighborhood sizes are selected as network inputs to construct a multi-channel network structure with different scale inputs. Secondly, a multiscale feature mapping module with fusion of dilated convolution is designed. Different numbers of dilated convolution are used to increase the model's receptive field without increasing network parameters, and spatial-spectral features of different scales are fused. Skip connections are added between the designed modules to fully combine the shallow and deep features of the network, without skip connections in the modules, reduce the computation of the overall network. Finally, the resulting features are input into the fully connected layer and softmax to complete the classification. In addition, in order to prevent the network from overfitting, we add dropout regularization method after the fully connected layer to improve the generalization ability of the model. Experiments are conducted on three publicly available datasets, Indian Pines, University of Pavia and Salinas, with accuracy rates of 99.28%, 99.48% and 99.25%, respectively. The experimental results show that compared with the classification methods based on convolutional neural network, the proposed method can effectively extract multiscale features of hyperspectral image and enhance the expression ability of spatial-spectral fusion features, and the classification performance is better.
To address the problem that the orthopedic force of the traditional bone external fixator cannot be accurately controlled and has poor safety,the article establishes a knee bone external fixation robot based on the RCM configuration and proposes an orthopedic force tracking control method with indirect adaptive impedance.Due to the existence of mechanical friction and external uncertainty,to address the problem of poor robustness of impedance control,the force error signal is adopted as the driving force of the target impedance,a new impedance model is formulated to adapt to the changes in the orthopedic environment.According to the changes in the contact force,the adaptive law is designed to regulate the impedance parameters in real time,which compensates for the uncertainty of the environmental dynamics on the line.Therefore,the system's force tracking error is zero.Comparative simulations and prototype experiments of the proposed control algorithm for tibial orthopedic force control are implemented.The results show that the impedance control simulation results of the weighted average of the performance indicators are 2.12,8.58 and 13.2,and the maximum error of the force tracking experiment is 20 N.Compared with impedance control,the weighted average of the performance indicators of adaptive impedance control simulation is only 0.36,0.18 and 0.61,and the force tracking fluctuation error is controlled to be within±3N,which has better robust and adaptive ability.
In aviation, aerospace, and other fields, nanomechanical resonators could offer excellent sensing performance. Among these, graphene resonators, as a new sensitive unit, are expected to offer very high mass and force sensitivity due to their extremely thin thickness. However, at present, the quality factor of graphene resonators at room temperature is generally low, which limits the performance improvement and further application of graphene resonators. Enhancing the quality factor of graphene resonators has emerged as a pressing research concern. In a previous study, we have proposed a new mechanism to reduce the energy dissipation of graphene resonators by utilizing phononic crystal soft-supported structures. We verified its feasibility through theoretical analysis and simulations. This article focuses on the fabrication of a phononic crystal soft-supported graphene resonator. In order to address the issues of easy fracture, deformation, and low success rate in the fabrication of phononic crystal soft-supported graphene resonators, we have studied key processes for graphene suspension release and focused ion beam etching. Through parameter optimization, finally, we have obtained phononic crystal soft-supported graphene resonators with varying cycles and pore sizes. Finally, we designed an optical excitation and detection platform based on Fabry–Pérot interference principle and explored the impact of laser power and spot size on phononic crystal soft-supported graphene resonators.
Significance With the continuous development of three-dimensional(3D)acquisition equipment such as 3D Lidar,3D point cloud data has recently become more accurate and easier to obtain.As the most important representation of 3D data,3D point cloud is widely used in visual tasks in the fields of autonomous driving,robotics,remote sensing,cultural relic restoration,augmented reality,and virtual reality,among others.Owing to the large amount of original point cloud data and the fact that the acquisition process is easily mixed with noise and outliers,the direct use of the original point cloud data is not effective.Therefore,it is critical to study the processing methods for 3D point clouds. A point cloud is a collection of spatial sampling points of the target surface properties in the same coordinate system.The sampling points contain geometric information such as the 3D coordinates and size,as well as characteristic information such as the object color and texture features.Traditional 3D point cloud processing methods are based on geometric analyses.Point cloud data are processed by estimating the geometric information such as the normal vector,curvature,and density of the point cloud,and by combining traditional feature descriptors.Although its accuracy is high,it is not suitable for complex point cloud scenes such as large rotations,and the calculation function is extremely cumbersome.Classical machine-learning methods can process 3D data and learn effective feature information;however,machine learning is highly dependent on accurate manual identification features.Massive 3D data not only increase the number of manual labels but also make labeling significantly more difficult than two-dimensional(2D)images.Deep learning methods can train and calculate large-scale data,autonomously learn latent-space features and advanced laws in the input information,and are suitable for processing massive amounts of point cloud data.Although deep learning methods require a considerable time for training the samples to learn the parameter information,the test time is significantly shorter than that of machine learning methods,and the prediction results are more accurate.Considering the irregular,sparse,and uneven internal structure of 3D point clouds,the efficient implementation of 3D point cloud processing based on deep learning has recently become the focus of researchers.Therefore,this study reviews the research progress of deep learning-based 3D point cloud processing methods over the past six years and presents the future research trends,aiming to provide inspiration and ideas for researchers in point cloud processing. Progress In this study,we focus on deep learning-based 3D point cloud processing tasks and provide a development route for the most commonly used deep learning methods for four point cloud processing tasks over the past six years(Fig.1).The 3D point cloud mainly includes the following four types of processing tasks:1)denoising and filtering,2)compression,3)super-resolution,and 4)restoration,completion,and reconstruction. Deep learning methods for point cloud denoising and filtering tasks can be classified into the following five types:CNN-based,upsampling-based,filter-based,gradient-based,and others.PointProNet and GeoGCN learn feature differences based on convolutional networks to remove noise;however,point cloud information is lost during the preprocessing stage.DUP-Net,PUGeo-Net,and PU-GACNet are classic upsampling-based denoising methods that denoise by modifying the feature extractor and feature expander while ignoring certain local features.NPD and PointCleanNet combine filtering ideas with deep learning and can simultaneously achieve noise removal and point cloud geometric feature retention.The Score-based method constructs a gradient field according to the distribution characteristics of the noise point cloud,and the robustness is enhanced;however,relatively few studies have been conducted.NoiseTrans draws on the idea of a Transformer to achieve the effective extraction and retention of fine features in point clouds.Table 2 presents a comparison of the advantages and disadvantages of the common methods. Deep learning methods for point cloud compression tasks are generalized.According to lossless and lossy compression,they are divided into two categories and analyzed(Tables 3 and 4).The point cloud lossless compression methods,OctSqueeze and VoxelDNN,improve the accuracy of point cloud probability prediction;however,part of the point cloud information is lost.PCGCv2,TransPCC,and SparsePCGC are the typical point cloud lossy compression methods.The point cloud feature is learned through a network structure,which prevents the loss of detailed information and improves the quality of the reconstructed point cloud. Subsequently,deep-learning methods for point cloud super-resolution tasks are outlined.Classification and comparative analyses are performed for the following four methods:convolutional neural network(CNN),graph convolutional neural network(GCN),generative adversarial network(GAN),and other structures(Table 5).PU-Net and PU-GCN extract rich detailed features based on CNN and GCN,respectively;however,numerous calculations are required.PU-GAN exploits the dynamic adversarial optimization details of the generator and discriminator.MPU and PU-Transformers combine the idea of a 2D super-resolution algorithm with the PointNet structure,which is a new idea worth trying. The deep learning methods for point cloud restoration,completion,and reconstruction tasks include three aspects,which are image-based,sampling-based,and completion-based,for which a comparative analysis is performed(Table 6).PCDNet reduces the number of computations by extracting 2D image features and deformations.Sampling-based methods use networks to generate dense and complete point clouds.PCN,TopNet,and SA-Net can fill in missing structures with input point clouds;however,completion-based methods are susceptible to incomplete point clouds. Recently,KITTI,PCN,nuScenes,and other public point cloud datasets and performance indicators,such as CD,P2M,and RMSE,have significantly promoted the in-depth research of point cloud processing tasks(Tables 7 and 8). Conclusions and Prospects 3D point cloud processing methods based on deep learning have gradually become an important research direction in the field of computer vision.Although several positive achievements have been made,there is significant room for further development.The following aspects should be considered when conducting in-depth research:the combination of multiple processing tasks,point cloud data feature processing,low-cost network models and hardware devices,and adaptable datasets.
Self-paced treadmill is the key human-robot interactive equipment for virtual reality. The current study focuses on the jumping interaction control technology for self-paced treadmill to enrich the application scenarios. For the purpose to analyze the stability of human jump landing, a novel variable stiffness spring-mass loaded inverted pendulum model is proposed, which takes into account the combined effects of lower limb bones and joint muscles. Experimental results show that the proposed model can realize the modeling of the mass center motion trajectory and the analysis of the jumping stable domain, the accuracy of stability recognition is 93.0%. Based on the proposed model and the stability analysis, the jumping interaction control strategies for the self-paced treadmill are proposed to improve human stability during jumping landing. The simulation and experimental results show that the proposed method can improve the stability of human jump landing significantly. Meanwhile, the proposed method reduces the torque of lower limb joints effectively. The peak torque of the knee joint reduces from 230 N/m to 210.7 N/m, and the peak torque of the ankle joint reduces from 143.6 N/m to 131 N/m, which is expected to lower the risk of injury.
Rolling bearing is one of the most important components of rotating machinery systems to ensure safe operation. It is important to carry out studies on rolling bearing feature recognition for theoretical and practical application. The commonly used deep learning rolling bearing feature recognition methods require supervised labeled data or unsupervised fault data to participate in the training, and labels of data and fault data are not easily accessible to meet the rolling bearing feature recognition requirements. This article proposes an edge computing method for differential evolution of generative adversarial networks rolling bearing feature recognition, namely the EC-DE method. The training process uses only healthy data to train the generative adversarial networks and learn the distribution pattern of healthy data. The edge node compares the distribution difference between the input samples and the generative samples of generative adversarial networks for identification and exits early according to the health confidence level to improve the system′s real-time performance. The cloud node uses a differential evolution algorithm to search the generator latent space of the generative adversarial networks to obtain the latent variables corresponding to the input samples, which improves the recognition accuracy. The proposed method achieves 99.8% accuracy on CWRU rolling bearing public data set and is insensitive to hyper-parameters, and the inference stage takes less time, which is valuable for a practical production application.
为解决传统耳鸣治疗声难以兼顾治疗效果和音频悦耳程度等问题,本文提出了一种兼顾音频悦耳程度和耳鸣治疗效果的创新音频处理算法.首先遵循耳蜗感音的对数分布特性,基于固定上下限截止频率的数字滤波器组对原始自然声的音频进行分频滤波;然后对原始自然声的音频沿着频率对数轴均衡,使得声频在全频率范围内实现能量均衡的同时,兼顾治疗效果.结果显示,经过对数能量均衡处理后的音频,不但整体符合粉红噪声的分布,且在 4 001～8 000 Hz和 8 001～10 000 Hz高频段内的能量得到显著提高,处理后的音频能量分别超过原始音频能量 15 倍和 100 倍.本文提出的创新算法不但满足耳鸣患耳对音频感音的悦耳习惯,而且为高频耳鸣无疗效的国际难题提供了治疗方案.
Variational mode decomposition（VMD）, a very active branch in the field of adaptive signal decomposition, has become a hot research direction in the field of signal processing. VMD shows good performance in processing non-stationary and nonlinear signals. Aiming at VMD model and its parameter selection, many extended models and parameter optimization methods have been studied. This article reviews the research progress of VMD in recent decade, summarizes and analyzes the relevant literature. Firstly, it analyzes the principle advantages of VMD and its application potential in various fields. Secondly, according to the matching ability of the model to different signal types, the different characteristics and applicable scenarios of extended models of VMD are summarized by classification. Then, the research progress of parameter optimization methods for VMD and its extended models are summarized to discuss and analyze the characteristics of different model parameter optimization methods and the latest research trends. Finally, 6 prospects are put forward for the future development of VMD, pointing out the direction for subsequent research.
汽车装配件的缺陷检测是汽车制造流程中的重要环节,不仅可以提升产品质量,降低退货率,避免成本浪费,还可以为驾驶人员提供安全保障.最早的缺陷检测依靠专家经验,准确度低,人力成本大,而无损检测技术依靠介质,且效率不高.引入机器视觉不仅可以平衡检测精度和效率的问题,还能提高检测系统的鲁棒性,是最有发展潜力的缺陷检测技术之一.本文首先给出了视觉缺陷检测的定义和主要流程,简述了视觉缺陷检测系统中的图像采集硬件,然后从常用的缺陷分割方法、特征提取方法、卷积神经网络 3 个方面综述了近年来汽车装配件缺陷检测的研究进展,并对比分析了相关方法的优缺点.接着把汽车的装配件大致分为轮毂轮胎、车身漆面、零件、发动机等 4 类,总结了缺陷类型及其缺陷检测算法的研究现状.随后介绍了与汽车工业相关的 10 个数据集和缺陷检测性能评价指标.最后指出针对汽车装配件的缺陷检测目前面临着诸多方面的技术挑战,并对进一步的工作进行了展望.
故障诊断是工业系统健康监测的重要内容,现有的数据驱动故障诊断方法多是利用类别平衡的数据集进行建模的.但在实际应用中,工业系统往往产生大量类别不平衡的样本,给数据驱动故障诊断带来挑战.这一问题受到了学术界和工业界的广泛关注,围绕该方面的研究取得了丰硕的成果.但是,目前针对不平衡分布的数据驱动故障诊断的研究进展综述相对较少,因此无法明确现实的挑战以及未来的研究方向.本文针对不平衡分布的约束问题,从数据驱动诊断方法和诊断应用场景这两个方面综述了国内外的研究进展,并提出了面临的挑战及未来的展望,为故障诊断的研究与应用提供参考.
针对像素尺寸差异、相位误差等导致相机与投影仪同名点像素坐标匹配错误、有效像素点缺失问题,提出一种基于虚拟视野的改进包裹相位-坐标映射方法.首先,对不同重叠视场下的条纹信息进行分析,确定较优投影模式;其次,设计两组高低频率的横纵条纹,提取低频相位极值计算虚拟投影视野、高频周期对真实和虚拟视野逐级编号,实现视野小范围匹配;最后,改进包裹相位坐标映射方法和相位差阈值判别准则,逐编号求解投影像素坐标,获得像素间的精准映射关系.实验结果表明,相位中相同像素区间的包裹相位均方根误差,相较于连续相位降低了 78.6%.在平面和复杂表面实验中,有效像素数量相较于传统匹配增长了 9.21 倍和 9.43 倍,像素误匹配坐标比例由传统相位匹配的 80.55%、59.4%降低至 14.26%、12.56%,为自适应条纹测量技术中像素同名点匹配提供了可行的解决方案.
一般情况下电子产品在失效前性能已经发生退化,而传统的寿命预测方法没有利用退化信息.以智能脱扣器电源模块为研究对象,分析MOSFET开关周期和电路薄弱环节储能电容退化之间的关联关系,提出将MOSFET开关周期作为电源模块性能退化的特征量,建立以MOSFET开关周期为特征参量的智能脱扣器电源模块性能退化模型;将电源模块划分为健康、注意和危险 3 种健康状态,确定健康状态转移图和转移时间的计算方法,建立其健康状态评估模型;对电源模块进行温度应力下的加速退化实验,验证性能退化模型和健康状态评估模型,并预测电源模块在 40℃工作环境下由健康状态转移至注意状态的平均转移时间为 3 906 天,转移至危险状态的平均转移时间为 9 296 天.
针对高空气象探测用温度传感器存在响应时间长、测温精度差以及辐射误差大等现状,研制了一种封装体积小、响应速度快、抗辐照能力强的温度传感器.采用固相法制备了 Mn-Ni-Cu-Fe-O 基片状热敏电阻,尺寸分别为 0.4 mm×0.4 mm×0.25 mm与0.6 mm×0.6 mm×0.25 mm.在片状热敏电阻敏感元表面依次沉积绝缘膜和铝金属反射薄膜.结果表明,研制的热敏电阻在-80℃～60℃范围内,其阻值范围分别为1 000 kΩ～4.2 kΩ与375 kΩ～1.6 kΩ;热耗散系数为1.034 mW/℃,响应时间为 0.63 s.利用太阳光模拟器对热敏电阻进行太阳辐射测温误差研究,在 100 W/m2 的辐照强度下测温误差最小为 0.22℃.该热敏电阻有望应用在高空气象探测的探空仪中.
伴随当前航空、航天、大型科学仪器智能制造场景下多环节、多过程、多装备制造工艺与制造信息不断融合优化,正推动长度测量需求从传统局部、单参数、离线、分时测量快速转变为全局、多目标的在线、实时、同步测量,现有激光测距方法受到原理限制,在精度、实时性与并行能力等方面面临瓶颈.本文提出了一种基于飞行时间法的双飞秒激光多目标绝对距离测量系统,设计特定的光纤分束方案,使飞秒激光光源同时到达多个目标,并构造时域可区分特征匹配多组脉冲,携带距离信息的脉冲经强度互相关模块后进入FPGA数据处理单元,实现多路绝对距离的实时解算,并完成系统的集成化设计.实验表明系统在2 kHz的更新速度下阿伦偏差评定的测量精度优于 10-5 m,与干涉仪对比的测距偏差小于 4 μm,在此基础上,搭建平行光路结构实现多自由度的同时测量,测量偏差优于 5″.
针对现有雷达辐射源信号(RES)特征信息易受噪声影响、分选时效性低等问题,提出一种基于模糊函数多维结构度量特征的信号数据流在线分选方法.首先,应用图像全局相似性思想,以积分加速后的非局部均值平滑方法对信号模糊函数进行去噪处理.其次,从处理后模糊函数的主、侧两个角度提取多维度结构分布特征,形成特征向量.最后,优化了一种半监督学习分选模型并在线作用于不断输入的信号特征向量流,得到实时分选结果.实验结果表明,在先验数据量较少的情况下,所提方法在8～18 dB的信噪比环境中均可保持99%及以上的分选成功率,即使处于2 dB环境下,准确率仍可达91.8%.同时,提取单个信号特征平均耗时仅需 0.29 s.结果验证了所提方法的有效性和实时性,具有一定工程价值.
针对供热管网泄漏检测、定位困难的问题,本文充分考虑了管网负压波传播的多径效应,提出了一种基于负压波传播最短路径规划的供热管网泄漏定位方法.本方法首先搜索出各潜在漏点到各压力变送器的最短路径,计算负压波在管网中传播的最短时延,构成时延标准库;然后将现场测得的负压波的实际到达时延与时延标准库进行比对,确定漏点位置.在区域面积为 13 km×5 km的现场供热管网对本方法开展了泄漏定位实验验证;当有效压力监测点数为 5 时,本方法可实现对 1 000 m半径比对范围内泄漏的零偏差定位.本方法采用了精确的非模糊型的原理,且避免了在现场大型管网所有支管上安装压力变送器,具有重要的应用价值.
