LQTTrack:Multi-Object Tracking by Focusing on Low-Quality Targets Association

Multi-object tracking(MOT)has seen rapid improvements in recent years.However,frequent occlusion remains a significant challenge in MOT,as it can cause targets to become smaller or disappear entirely,resulting in lowquality targets,leading to trajectory interruptions and reduced tracking performance.Different from some existing methods,which discarded the low-quality targets or ignored low-quality target attributes.LQTTrack,with a lowquality association strategy(LQA),is proposed to pay more attention to low-quality targets.In the association scheme of LQTTrack,firstly,multi-scale feature fusion of FPN(MSFF-FPN)is utilized to enrich the feature information and assist in subsequent data association.Secondly,the normalized Wasserstein distance(NWD)is integrated to replace the original Inter over Union(IoU),thus overcoming the limitations of the traditional IoUbased methods that are sensitive to low-quality targets with small sizes and enhancing the robustness of low-quality target tracking.Moreover,the third association stage is proposed to improve the matching between the current frame’s low-quality targets and previously interrupted trajectories from earlier frames to reduce the problem of track fragmentation or error tracking,thereby increasing the association success rate and improving overall multi-object tracking performance.Extensive experimental results demonstrate the competitive performance of LQTTrack on benchmark datasets(MOT17,MOT20,and DanceTrack).

A Fusion Localization Method Based on Target Measurement Error Feature Complementarity and Its Application

In the multi-radar networking system,aiming at the problem of locating long-distance targets synergistically with difficulty and low accuracy,a dual-station joint positioning method based on the target measurement error feature complementarity is proposed.For dual-station joint positioning,by constructing the target positioning error distribution model and using the complementarity of spatial measurement errors of the same long-distance target,the area with high probability of target existence can be obtained.Then,based on the target distance information,the midpoint of the intersection between the target positioning sphere and the positioning tangent plane can be solved to acquire the target's optimal positioning result.The simulation demonstrates that this method greatly improves the positioning accuracy of target in azimuth direction.Compared with the traditional the dynamic weighted fusion(DWF)algorithm and the filter-based dynamic weighted fusion(FBDWF)algorithm,it not only effectively eliminates the influence of systematic error in the azimuth direction,but also has low computational complexity.Furthermore,for the application scenarios of multi-radar collaborative positioning and multi-sensor data compression filtering in centralized information fusion,it is recommended that using radar with higher ranging accuracy and the lengths of baseline between radars are 20–100 km.

Point association analysis of vessel target detection with SAR, HFSWR and AIS

A space-borne synthetic aperture radar （SAR）, a high frequency surface wave radar （HFSWR）, and a ship automatic identification system （AIS） are the main remote sensors for vessel monitoring in a wide range. These three sensors have their own advantages and weaknesses, and they can complement each other in some situations. So it would improve the capability of vessel target detection to use multiple sensors including SAR, HFSWR, and A/S to identify non-cooperative vessel targets from the fusion results. During the fusion process of multiple sensors＇ detection results, point association is one of the key steps, and it can affect the accuracy of the data fusion and the efficiency of a non-cooperative target＇s recognition. This study investigated the point association analyses of vessel target detection under different conditions： space- borne SAR paired with AIS, as well as HFSWR, paired with AIS, and the characteristics of the SAR and the HFSWR and their capability of vessel target detection. Then a point association method of multiple sensors was proposed. Finally, the thresholds selection of key parameters in the points association （including range threshold, radial velocity threshold, and azimuth threshold） were investigated, and their influences on final association results were analyzed.

SEQUENTIAL ALGORITHM FOR MULTISENSOR PROBABILISTIC DATA ASSOCIATION

Based upon a multisensor sequential processing filter, the target states in a3D Cartesian system are projected into the measurement space of each sensor to extend thejoint probabilistic data association (JPDA) algorithm into the multisensor tracking systemsconsisting of heterogeneous sensors for the data association.

Track Association for Dynamic Target Tracking System Based on AP Algorithm

Track association of multi-target has been recognized as one of the key technologies in distributed multiple-sensor data fusion system,and its accuracy directly impacts on the performance of the whole tracking system.A multi-sensor data association is proposed based on aftinity propagation(AP)algorithm.The proposed method needs an initial similarity,a distance between any two points,as a parameter,therefore,the similarity matrix is calculated by track position,velocity and azimuth of track data.The approach can automatically obtain the optimal classification of uncertain target based on clustering validity index.Furthermore,the same kind of data are fused based on the variance of measured data and the fusion result can be taken as a new measured data of the target.Finally,the measured data are classified to a certain target based on the nearest neighbor ideas and its characteristics,then filtering and target tracking are conducted.The experimental results show that the proposed method can effectively achieve multi-sensor and multi-target track association.

Rough Sets Probabilistic Data Association Algorithm and its Application in Multi-target Tracking

A rough set probabilistic data association(RS-PDA)algorithm is proposed for reducing the complexity and time consumption of data association and enhancing the accuracy of tracking results in multi-target tracking application.In this new algorithm,the measurements lying in the intersection of two or more validation regions are allocated to the corresponding targets through rough set theory,and the multi-target tracking problem is transformed into a single target tracking after the classification of measurements lying in the intersection region.Several typical multi-target tracking applications are given.The simulation results show that the algorithm can not only reduce the complexity and time consumption but also enhance the accuracy and stability of the tracking results.

Study of p-^(11)B Reaction Associated With Clean Fusion Fuel

Ｓｔｕｄｙｏｆｐ┐１１ＢＲｅａｃｔｉｏｎＡｓｏｃｉａｔｅｄＷｉｔｈＣｌｅａｎＦｕｓｉｏｎＦｕｅｌ①ＹｕａｎＪｉａｎ，ＬｉＨｕｉａｎｄＳｕｎＺｕｘｕｎＣｈｉｎａＩｎｓｔｉｔｕｔｅｏｆＡｔｏｍｉｃＥｎｅｒｇｙ，Ｐ．Ｏ．Ｂｏｘ２７５（１０），Ｂｅｉｊｉｎｇ...

利用伯努利滤波的多目标机动雷达误差配准方法

传统的组网雷达多目标误差配准方法通常假设数据关联关系已知,但在平台机动的情况下,系统同时存在雷达测量偏差和平台姿态角偏差,且雷达观测过程中会受到杂波干扰,导致数据关联尤为困难。针对这一问题,该文提出一种基于伯努利滤波的多目标机动雷达误差配准方法。首先建立系统偏差的量测与状态方程,然后将系统偏差建模成伯努利随机有限集,利用公共坐标系下的原始量测可实现系统偏差在伯努利滤波框架下的递推估计,有效避免了数据关联问题。同时,为了充分利用多目标量测信息,提出一种修正的贪婪量测划分方法,在每个滤波时刻挑选出系统偏差对应的最优量测子集,利用量测子集中的多量测信息实现系统偏差的集中式融合估计,提高了系统偏差的估计精度和收敛速度。仿真实验表明,所提方法能够在数据关联未知的多目标多杂波场景下对雷达测量偏差和平台姿态角偏差进行有效估计,在平台姿态角变化率较低时,所提方法具有较强的适应性。

多源视频数据驱动的舰面保障作业三维重建

针对现有公开的航母舰面保障作业视频数据碎片化、相机角度多变、像素质量低下等问题,提出一种多源视频数据驱动的舰面保障作业三维重建方法.该方法首先对视频中对象轨迹方向进行预测,并采用联合外观和方向相似度的跟踪方法实现视频数据的多目标轨迹跟踪;然后基于跟踪结果,依次采用相似度关联、特征轨迹分割、轨迹聚类提取多源数据中的保障作业对象并进行关联与融合;最后对融合得到的轨迹进行逼真三维重建.在MOT20数据集和舰面保障作业公开视频数据上,实验和用户调研数据的t检验结果表明,采用所提方法有99.5%的把握得到可信的舰面保障作业三维重建效果.

面向目标跟踪的多传感器数据融合研究综述

随着现代目标跟踪场景的日益复杂化,单一传感器探测系统已无法满足目标跟踪的实际需求,多传感器数据融合技术成为复杂场景下目标跟踪问题的重要解决方案。本文综述了多传感器数据融合中的数据关联和估计融合这两项关键过程的基本原理与研究现状,整理了神经网络、强化学习等人工智能技术在多传感器数据融合领域应用的相关研究,并对多传感器数据融合方法在目标跟踪领域的发展进行了展望。

基于动态估计反馈的灰色理论航迹关联算法

针对传统基于灰色理论航迹关联算法未充分利用航迹历史信息,在密集目标以及时变噪声协方差等场景下关联正确率下降问题,在传统灰色分析法基础上增加了一种动态估计反馈机制。改进后的算法引入了Sage-Husa估计器来实时估计传感器的噪声协方差作为评估输出数据质量的依据,并使用Critic赋权法将实时噪声协方差转换为各时刻序贯权重,从而保证最大程度上的利用航迹历史信息。仿真证明,在密集目标与时变噪声协方差等特殊关联场景下,所提算法明显优于传统灰色分析法以及模糊法、加权法等经典算法,充分证明了所提算法的性能优越性以及鲁棒性。

基于AIMM-PF的多机动目标协同跟踪

针对常规线性卡尔曼滤波越来越不能满足多机动目标跟踪精度需求的问题,提出一种基于自适应多模型粒子滤波的协同跟踪方法.首先,主车和协同车分别执行自适应交互式多模型粒子滤波(adaptive interactive multi model particle filter,AIMM-PF)算法,获得环境中目标车辆的运动状态;其次,协同车通过车车通信将跟踪到的目标状态发送给主车;最后,利用基于匈牙利算法和快速协方差交叉算法的数据关联和数据融合技术实现多机动目标的协同跟踪.搭建了V2V通信、雷达和定位仿真系统,选定两辆智能车作为主车和协同车,感知并跟踪200 m范围内的7辆目标车,进行了仿真试验.结果表明,与传统的单车跟踪相比,协同跟踪扩大了感知范围,且在不影响跟踪效率的情况下使跟踪误差降低了31.1%.

基于时频融合特征的肺动脉高压心音分类模型

先心病相关肺动脉高压是一种严重的心血管疾病,致死率高,对其进行早期筛查与识别对于治愈尤为重要。目前临床是通过右心导管术确诊,此为有创检查,不便于在大规模筛查中采用,研究一种无创便捷的识别方法迫在眉睫。文中建立了一种时频融合的心音分类模型。首先对心音信号进行预处理,然后使用融合滤波器组对信号进行转换并求取动态时频特征,最后将得到的融合特征参数输入表格式先验数据拟合网络(TabPFN)中进行分类识别。实验结果表明,该算法在正常、CHD-PAH和CHD中的平均准确率、精确率、灵敏度、特异度和F1分别为92.21%,92.15%,92.15%,96.11%,92.14%。对于先心病相关肺动脉高压的早期筛查与识别具有重要意义。

省级自然资源监测数据库建设研究

针对自然资源监测数据成果集成管理与动态更新需求,开展了省级自然资源监测数据库建设研究。在数据分层、数据模型、逻辑设计、物理实现等设计基础上,详细阐述了数据库建库与更新的技术流程。通过空间数据多时态管理技术、多源矢量数据关联融合、基于身份标识码的全生命周期管理、按需更新的增量更新模式等关键技术应用实施,解决数据标准化整合与数据库动态更新问题,从而为自然资源监测数据库纳入自然资源三维立体时空数据库体系提供支持。

线粒体融合蛋白2在缺血再灌注损伤中作用的研究进展

线粒体融合蛋白2(Mfn2)介导线粒体融合与分裂过程,参与调控线粒体动力学,是线粒体形态、结构和功能的动力素相关蛋白。缺血再灌注损伤(IRI)是外科手术中一种常见的并发症,也是创伤性休克、严重感染等致命疾病的主要死亡原因。近年来发现,Mfn2在IRI中具有修复作用,包括肠IRI、肺IRI、肾IRI和心脏IRI等。Mfn2通过多个途径参与不同IRI的修复,减缓疾病进展。深入研究Mfn2与IRI的关系及相关机制,可为IRI的防治提供新的靶点和思路。

基于个性化随机游走的基因-表型关联分析

基因与表型间的关联分析对揭示生物的内在遗传关联具有重要意义.随机游走算法可以融合多组学数据,聚合一阶或高阶邻居的标签信息,对网络中不同节点间关联信息进行补全,提高关联预测的准确度,进而发现基因和表型间潜在的遗传关联.但现有随机游走算法通常平等地对待每个节点,忽略了不同节点的重要性,使非重要节点过度传播,降低了模型性能.为此,本文提出了一种基于多组学数据融合的个性化随机游走算法(individual Multiple Random Walks,iMRW),在由基因、miRNA及表型节点构建的多组学异质网络上,基于网络拓扑结构,设计个性化多元随机游走策略,为不同重要程度的节点分配不同的游走步长,并结合高斯相互作用属性核相似性与随机游走,对网络不同节点及节点间关联信息进行补全,最终实现多源基因-表型关联矩阵的融合,准确获取基因-表型关联预测矩阵.在不同实验设置下,与主流算法的对比实验结果均显示iMRW能够取得更优的预测性能.在玉米光合作用能力和淀粉含量表型的实验分析结果也进一步证实了iMRW在识别潜在的基因-表型关联的实用性与有效性.

子宫内膜样腺癌组织NuSAP1、MFN2表达与临床病理特征和预后的关系

目的探讨子宫内膜样腺癌(EA)组织核仁纺锤体相关蛋白1(NuSAP1)、线粒体融合蛋白2(MFN2)表达与临床病理特征和预后的关系。方法选取2019年3月至2020年10月南阳市第一人民医院收治的121例EA患者作为研究对象,应用免疫组织化学染色法检测患者癌组织及其对应癌旁组织NuSAP1、MFN2阳性表达率,分析癌组织NuSAP1、MFN2阳性表达率与患者临床病理特征的关系。出院后随访3年,完成随访112例,应用Kaplan-Meier生存曲线分析NuSAP1、MFN2阳性表达组和阴性表达组的预后差异,应用多因素COX回归分析EA患者预后的影响因素。结果EA癌组织NuSAP1阳性表达率为72.37%,明显高于癌旁组织的35.54%,MFN2阳性表达率为38.02%,明显低于癌旁组织的80.17%,差异均有统计学意义(P<0.05);国际妇产科联盟(FIGO)临床分期Ⅱ~Ⅲ期、中低分化、有淋巴结转移、深肌层浸润癌组织中NuSAP1阳性表达率分别为88.68%、83.54%、96.15%、90.63%,明显高于FIGO临床分期Ⅰ期的60.29%、高分化的57.14%、无淋巴结转移的55.79%、浅肌层浸润癌组织的66.29%,差异均有统计学意义(P<0.05);FIGO临床分期Ⅱ~Ⅲ期、中低分化、有淋巴结转移、深肌层浸润癌组织中MFN2阳性表达率分别为18.87%、30.38%、7.69%、15.63%,明显低于FIGO临床分期Ⅰ期的52.94%、高分化的52.38%、无淋巴结转移的46.32%、浅肌层浸润癌组织的46.07%,差异均有统计学意义(P<0.05);Kaplan-Meier法分析结果显示,NuSAP1阴性表达组和MFN2阳性表达组患者的3年生存率分别为90.00%和87.80%,明显高于NuSAP1阳性表达组的69.51%和MFN2阴性表达组的67.61%,差异均有统计学意义(P<0.05)。结论EA患者组织中NuSAP1阳性表达率升高,MFN2阳性表达率降低,NuSAP1、MFN2表达水平与患者肿瘤分化、FIGO临床分期、淋巴结转移及深肌层浸润有关,且是患者死亡的危险因素,提示NuSAP1、MFN2可能参与了EA患者的疾病进展。

基于图注意力网络预测人类微生物与药物关联

目的采用图注意力网络(graph attention network,GAT)预测人类微生物与药物之间的潜在关联。方法选取三个常用的微生物-药物关联(microbe-drug associations,MDA)数据集(MDAD、aBiofilm和Drug Virus),基于数据集中丰富的生物信息构建一个异构网络,并提出一种基于GAT框架预测MDA的模型——GATMDA模型,用于预测微生物与药物间的关联。结果与现有的8种预测方法相比,GATMDA通过三种交叉验证方法在三个数据集上具有较好的预测效果。在5折交叉验证的性能评估中,在三个数据集上的受试者工作特征曲线下的面积(area under the curve,AUC)分别为0.9886、0.9941和0.9836,精确率-召回率曲线下的面积(area under the precision-recall curve,AUPR)分别为0.9667、0.9869和0.8795。通过病例研究进一步验证了GATMDA在预测MDA方面的有效性。结论基于GAT,GATMDA模型可以通过构建的异构网络对微生物-药物进行有效的关联预测。

基于数据中台的数据融合集成关键技术研究

当前开发新的业务系统往往需要从其他系统引接数据,系统之间的直接数据交互增加了系统开发的复杂性。为解决上述问题,研究基于数据中台的数据融合集成关键技术,为数据中台建设提供参考。设计数据中台技术架构,实现数据引接、数据管理和数据服务。覆盖数据生命周期,设计数据中台业务流程。瞄准数据赋能目标,提升数据融合集成核心能力,采用混合架构完成多源异构数据存储,利用数据层次模型实现数据映射,应用知识图谱进行数据关联。提出的数据中台技术架构已在实际项目中运用,达到预期效果。

碱熔分离富集-XRF溶液法测定伴生放射性矿中钍含量

为提高伴生放射性矿中钍含量检测效率,建立碱熔分离富集-X射线荧光光谱(XRF)法测定伴生放射性矿中常量钍的方法。试样经Na_(2) O_(2)分解,在过氧化氢存在条件下,用5%乙醇(TEA)-0.1 mol/L二醚二胺四乙酸(EGTA)浸提剂提取熔块,使钍与大部分基体元素U、Pb、Sr、Ti、Fe、P分离。沉淀经4 mol/L热盐酸溶液溶解收集于50 mL容量瓶,准确加入5 mL Sr标准溶液(0.10 mol/L),Th与内标元素Sr共存于溶液体系,4 mol/L盐酸溶液定容至刻度,摇匀。XRF分析中选择LiF200晶体同时测定钍的Lα分析谱线及锶的Kα分析谱线强度(kcps)。以两者强度比I_(Th)/I_(Sr)为横坐标,Th质量(μg)为纵坐标绘制工作曲线。钍的质量浓度在0~200μg/mL具有良好的线性关系,线性相关系数为0.9999,钍的方法检出限为0.36μg/mL。测定结果的相对标准偏差为4.4%~7.4%(n=3),对3种国家标准物质进行测定,测定值与标准值的相对误差小于4.8%,测定值与认定值基本相符。