Enhanced Temporal Correlation for Universal Lesion Detection

Universal lesion detection(ULD)methods for computed tomography(CT)images play a vital role in the modern clinical medicine and intelligent automation.It is well known that single 2D CT slices lack spatial-temporal characteristics and contextual information compared to 3D CT blocks.However,3D CT blocks necessitate significantly higher hardware resources during the learning phase.Therefore,efficiently exploiting temporal correlation and spatial-temporal features of 2D CT slices is crucial for ULD tasks.In this paper,we propose a ULD network with the enhanced temporal correlation for this purpose,named TCE-Net.The designed TCE module is applied to enrich the discriminate feature representation of multiple sequential CT slices.Besides,we employ multi-scale feature maps to facilitate the localization and detection of lesions in various sizes.Extensive experiments are conducted on the DeepLesion benchmark demonstrate that thismethod achieves 66.84%and 78.18%for FS@0.5 and FS@1.0,respectively,outperforming compared state-of-the-art methods.

Spatiotemporal Prediction of Urban Traffics Based on Deep GNN

Traffic prediction already plays a significant role in applications like traffic planning and urban management,but it is still difficult to capture the highly non-linear and complicated spatiotemporal correlations of traffic data.As well as to fulfil both long-termand short-termprediction objectives,a better representation of the temporal dependency and global spatial correlation of traffic data is needed.In order to do this,the Spatiotemporal Graph Neural Network(S-GNN)is proposed in this research as amethod for traffic prediction.The S-GNN simultaneously accepts various traffic data as inputs and investigates the non-linear correlations between the variables.In terms of modelling,the road network is initially represented as a spatiotemporal directed graph,with the features of the samples at the time step being captured by a convolution module.In order to assign varying attention weights to various adjacent area nodes of the target node,the adjacent areas information of nodes in the road network is then aggregated using a graph network.The data is output using a fully connected layer at the end.The findings show that S-GNN can improve short-and long-term traffic prediction accuracy to a greater extent;in comparison to the control model,the RMSE of S-GNN is reduced by about 0.571 to 9.288 and the MAE(Mean Absolute Error)by about 0.314 to 7.678.The experimental results on two real datasets,Pe MSD7(M)and PEMS-BAY,also support this claim.

Quantitative assessment of the spatio-temporal correlations of seismic events induced by longwall coal mining

Rock failure process as a natural response to mining activities is associated with seismic events, which can pose a potential hazard to mine operators, equipment and infrastructures. Mining-induced seismicity has been found to be internally correlated in both time and space domains as a result of rock fracturing during progressive mining activities. Understanding the spatio-temporal(ST) correlation of mininginduced seismic events is an essential step to use seismic data for further analysis, such as rockburst prediction and caving assessment. However, there are no established methods to perform this critical task. Input parameters used for the prediction of seismic hazards, such as the time window of past data and effective prediction distance, are determined based on site-specific experience without statistical or physical reasons to support. Therefore, the accuracy of current seismic prediction methods is largely constrained, which can only be addressed by quantitively assessing the ST correlations of mininginduced seismicity. In this research, the ST correlation of seismic event energy collected from a study mine is quantitatively analysed using various statistical methods, including autocorrelation function(ACF), semivariogram and Moran’s I analysis. In addition, based on the integrated ST correlation assessment, seismic events are further classified into seven clusters, so as to assess the correlations within individual clusters. The correlation of seismic events is found to be quantitatively assessable, and their correlations may vary throughout the mineral extraction process.

Temporal-spatial cross-correlation analysis of non-stationary near-surface wind speed time series

Temporal-spatial cross-correlation analysis of non-stationary wind speed time series plays a crucial role in wind field reconstruction as well as in wind pattern recognition.Firstly,the near-surface wind speed time series recorded at different locations are studied using the detrended fluctuation analysis(DFA),and the corresponding scaling exponents are larger than 1.This indicates that all these wind speed time series have non-stationary characteristics.Secondly,concerning this special feature( i.e.,non-stationarity)of wind signals,a cross-correlation analysis method,namely detrended cross-correlation analysis(DCCA) coefficient,is employed to evaluate the temporal-spatial cross-correlations between non-stationary time series of different anemometer pairs.Finally,experiments on ten wind speed data synchronously collected by the ten anemometers with equidistant arrangement illustrate that the method of DCCA cross-correlation coefficient can accurately analyze full-scale temporal-spatial cross-correlation between non-stationary time series and also can easily identify the seasonal component,while three traditional cross-correlation techniques(i.e.,Pearson coefficient,cross-correlation function,and DCCA method) cannot give us these information directly.

Effects of internal waves on signal temporal correlation length in the South China Sea

We simulated the temporal correlation of sound transmission using a two-dimensional advective frozen-ocean model with temperature data from a temperature sensor array on a propagation path in the South China Sea(SCS) Experiment 2009,and investigated the relationships of temporal correlation length,source-receiver range,and maximal sound speed fluctuation mainly caused by the solitary internal waves.We found that the temporal correlation length is-1.2-power dependent on source-receiver range and-0.9-power dependent on maximal sound speed fluctuation.The empirical relationship is deduced from one-day environmental measurements in a limited area,needing more works and verification in the future with more acoustic data.But the relationship is useful in many applications in the area of SCS Experiment 2009.

Temporal patterns of dengue epidemics:The case of recent outbreaks in Kaohsiung

Objective:To investigate whether major dengue outbreaks in the last two decades in Kaohsiung follow a precise temporal pattern.Methods:Government daily lab-confirmed dengue case data from three major dengue outbreaks occurring during the last two decades in Kaohsiung in2002,2014 and 2015,is utilized to compute the corresponding weekly cumulative percentage of total case numbers.We divide each of the three time series data into two periods to examine the corresponding weekly cumulative percentages of case numbers for each period.Pearson’s correlation coefficient was calculated to compare quantitatively the similarity between the temporal patterns of these three years.Results:Three cutoff points produce the most interesting comparisons and the most different outcomes.Pearson’s correlation coefficient indicates quantitative discrepancies in the similarity between temporal patterns of the three years when using different cutoff points.Conclusions:Temporal patterns in 2002 and 2014 are comparatively more similar in early stage.The 2015 outbreak started late in the year,but ended more like the outbreak in 2014,both with record-breaking number of cases.The retrospective analysis shows that the temporal dynamics of dengue outbreaks in Kaohsiung can strongly vary from one year to another,making it difficult to identify any common predictor.

Exploiting Correlations of Energy and Information:A New Paradigm of Energy Harvesting Communications

For deployment flexibility and device lifetime prolonging,energy harvesting communications have drawn much attention recently,which however,encounter energy domain randomness in addition to the channel state randomness and traffic load randomness.The three-dimensional randomness makes the resource allocation problem extremely difficult.To resolve this,we exploit the inherent correlations of energy arrival and information.The correlations include self correlations of energy profiles and mutual correlations between energy and information in both time and spatial domains.The correlations are explicitly explained followed by a state-of-art survey.Candidate mechanisms exploiting the correlations for the ease of resource allocation are introduced along with some recent progress.Finally,a case study is presented to illustrate the performance of the proposed algorithm.

The spatial-temporal evolution of coherent structures in log law region of turbulent boundary layer

The spatial-temporal evolution of coherent structures （CS） is significant for turbulence control and drag re- duction. Among the CS, low and high speed streak structures show typical burst phenomena. The analysis was based on a time series of three-dimensional and three-component （3D-3C） velocity fields of the flat plate turbulent boundary layer （TBL） measured by a Tomographic and Time-resolved PIV （Tomo TRPIV） system. Using multi-resolution wavelet transform and conditional sampling method, we extracted the intrinsic topologies and found that the streak structures appear in bar-like patterns. Furthermore, we seized locations and velocity information of transient CS, and then calculated the propagation velocity of CS based on spatial-temporal cross-correlation scanning. This laid a foundation for further studies on relevant dynamics properties.

Signal Analysis of the Climate: Correlation, Delay and Feedback

One of the ingredients of anthropogenic global warming is the existence of a large correlation between carbon dioxide concentrations in the atmosphere and the temperature. In this work we analyze the original time-series data that led to the new wave of climate research and test the two hypotheses that might explain this correlation, namely the (more commonly accepted and well-known) greenhouse effect (GHE) and the less-known Henry’s Law (HL). This is done by using the correlation and the temporal features of the data. Our conclusion is that of the two hypotheses the greenhouse effect is less likely, whereas the Henry’s Law hypothesis can easily explain all effects. First the proportionality constant in the correlation is correct for HL and is about two orders of magnitude wrong for GHE. Moreover, GHE cannot readily explain the concurring methane signals observed. On the temporal scale, we see that GHE has difficulty in the apparent negative time lag between cause and effect, whereas in HL this is of correct sign and magnitude, since it is outgasing of gases from oceans. Introducing feedback into the GHE model can overcome some of these problems, but it introduces highly instable and chaotic behavior in the system, something that is not observed. The HL model does not need feedback.

Two new approaches for image registration based onspatial-temporal relationship

How to improve the probability of registration and precision of localization is a hard problem, which is desiderated to solve. The two basic approaches (normalized cross-correlation and phase correlation) for image registration are analysed, two improved approaches based on spatial-temporal relationship are presented. This method adds the correlation matrix according to the displacements in x- cirection and y- directions, and the registration pose is searched in the added matrix. The method overcomes the shortcoming that the probability of registration decreasing with area increasing owing to geometric distortion, improves the probability and the robustness of registration.

Detecting remotely triggered temporal changes around the Parkfield section of the San Andreas fault

Detecting temporal changes in fault zone properties at seismogenic depth have been a long-sought goal in the seismological community for many decades. Recent studies based on waveform analysis of repeating earthquakes have found clear temporal changes in the shallow crust and around active fault zones associated with the occurrences of large nearby and teleseismic earthquakes. However, repeating earthquakes only occur in certain locations and their occurrence times cannot be controlled, which may result in inadequate sampling of the interested regions or time periods. Recent developments in passive imaging via auto- and cross-correlation of ambient seismic wavefields （e.g., seismic noise, earthquake coda waves） provide an ideal source for continuous monitoring of temporal changes around active fault zones. Here we conduct a systematic search of temporal changes along the Parkfield section of the San Andreas fault by cross-correlating relatively high-frequency （0.4-1.3 Hz） ambient noise signals recorded by 10 borehole stations in the High Resolution Seismic Network. After using stretch/compressed method to measure the delay time and the decorrelation-index between the daily noise cross-correlation functions （NCCFs）, we find clear temporal changes in the median seismic velocity and decorrelation-index associated with the 2004 M6.0 Parkfield earthquake. We also apply the same procedure to the seismic data around five regional/teleseismic events that have triggered non-volcanic tremor in the same region, but failed to find any clear temporal changes in the daily NCCFs. The fact that our current technique can detect temporal changes from the nearby but not regional and teleseismic events, suggests that temporal changes associated with distance sources are very subtle or localized so that they could not be detected within the resolution of the current technique （-0.2%）.

1960-2020年河南省极端降水时空演变特征

全球变暖加快了水文循环速度,导致极端降水事件频发,增强了城市排水负荷和洪涝风险,并影响了区域气候的时空特征。在河南省均匀选取26个雨量站对1960—2020年61年间的降水资料进行研究,采用趋势分析、Kriging插值、M-K突变分析、Morlet小波变换和Spearman相关系数等方法对河南省9个极端降水指数的时空演变特征进行分析。结果显示:河南省年降水量呈上升趋势,极端降水事件呈下降趋势,河南省整体气候逐渐变得湿润。河南省极端降水指数具有显著空间差异,东南部极端降水风险显著大于西北部。除降水强度、最大连续降水日数和1 d最大降水量外,其余指数与年降水量、汛期和6—9月的各月降水量具有良好的相关性,这对于评估年降水量及其年内分布特征、极端降水事件频率有较好的参考作用。结果表明:降水从短历时高强度向长历时低强度演变,要应对此类长历时极端降水的风险,防涝思路应从提高短时间大量排水能力转变到提高长时间低强度吸收降水能力。

石家庄暴雨时空分布特征及灾情评估

基于石家庄市2015—2021年暴雨洪涝灾情资料数据,以及17个国家站及268个区域自动气象站数据,采用气候统计诊断方法分析了石家庄暴雨时空分布的气候特征,并利用灰色关联分析及逐步回归方法,建立了石家庄市暴雨灾情评估及预评估模型.结果表明:①石家庄暴雨频次及强度随时间呈递增趋势,暴雨强度年际变化增大且极端性增强.②石家庄西北部暴雨频次多、强度大,西南部暴雨频次相对较少,但强度最大,其中平山、井陉为大暴雨、特大暴雨高发区,复杂的地理环境使该地区发生暴雨洪涝灾害的风险增加.③由灰色关联分析方法确定的暴雨灾情等级正确率83.33%,能够反映实际暴雨灾情等级,且有利于客观区分同一等级内暴雨灾情大小.④基于气象因子,利用逐步回归方法建立的暴雨灾情评估及预评估模型正确率可达68.75%.

时空相关的道路网络短时交通流预测模型

为有效解决复杂路网短时交通流预测问题中涉及的时空特征挖掘问题,提出一种基于改进长短时记忆神经网络(Improved Long Short-Term Memory, ILSTM)的交通流预测模型.首先,通过改进的遗传算法对长短时记忆神经网络(Long Short-Term Memory, LSTM)模型初始参数进行优化获得最优参数组合,解决LSTM初始参数设置对输出结果影响较大的问题.其次,针对复杂路网多路段交通流预测中遇到的空间特征提取问题,通过挖掘相关路段对目标路段交通流预测的影响程度,重新构建LSTM模型的损失函数,采用路网中相关路段对目标路段的影响系数,以损失函数输出值最小为终止条件,构建ILSTM模型.最后,选择加州公路局交通数据进行模型验证实验,采用遗传算法优化LSTM模型(Genetic Algorithm-LSTM, GA-LSTM)和单纯LSTM模型,以及皮尔森相关系数与LSTM组合模型(Pearson Correlation Coefficient-LSTM,PCC-LSTM),对工作日和周末数据的多次实验结果进行对比分析.实验结果表明:ILSTM模型能够充分考虑复杂路网交通流的时间和空间特征,预测平均误差约为1.16%,在收敛效率和预测精度方面均优于其他模型.

基于灰色关联模型的沿黄区域山东段生态安全综合评价及时空演化

以沿黄区域山东段为例,建立PSR模型构建生态安全评价指标体系,借助熵权法和灰色关联度模型对其2000—2018年展开生态安全综合评价,并进行时空演变分析。结果显示,整体上沿黄区域山东段生态安全水平呈现出波动中上升的趋势,由“不安全”向“比较安全”转变,子系统中生态响应和生态状态增加的幅度高于生态环境压力增大的幅度,生态环境有所好转;市级尺度下,山东段17地市生态安全水平有所提高,呈现出“下降-上升-下降”的倒“N”字形演变,生态安全类型以“一般安全”为主,高等级类型较少,城市生态安全问题比较严峻;城市生态安全内部空间差异明显,生态安全空间格局向东、向北转移,呈现出“东高西低”“北高南低”的高低错落分布,高等级生态安全区域从东部沿海向鲁北扩散,“临界安全”区域缩小至鲁中南和鲁西北。最后从生态环境治理与保护、优化调控三大空间、经济结构转型升级等方面提出优化路径,改善生态环境质量。

基于时空关联特征与B-LSTM模型的分布式光伏功率区间预测

提出一种基于时空关联特征与贝叶斯-长短期记忆神经网络(bayesian long short-term memory,B-LSTM)模型的分布式光伏功率区间预测方法。以长短期记忆神经网络(long short-term memory,LSTM)为基础构建近似贝叶斯神经网络,建立考虑时空关联特征的B-LSTM模型,利用其强大的记忆能力和特征提取不同特征尺度的模态分量,并进行分布式光伏功率区间预测。以某地区实际分布式光伏数据集进行算例分析,验证了所提方法的优越性。

时空多尺度关联特征融合的二维卷积网络细粒度动作识别模型

针对传统二维(2D)卷积网络提取时空特征尺度单一以及对细粒度动作数据集中帧与帧之间的远程时间关联信息利用不足的问题,本文提出时空多尺度关联特征融合的2D卷积网络细粒度动作识别模型。首先,为建模视频多尺度空间关联以加强对细粒度视频数据的空间表征能力,模型使用多尺度“特征压缩、特征激发”方式,使网络所提取空间特征更加丰富有效。然后,为充分利用细粒度视频数据时间维度上的运动信息,本文引入时间窗口自注意力机制,利用自注意力机制强大的远程依赖建模能力同时只在时间维度上进行自注意力操作,以较低计算成本建模远程时间依赖关系。最后,考虑到所提取时空特征对不同类型动作分类的贡献不均等,本文引入自适应特征融合模块,为特征动态赋予不同权重实现自适应特征融合。模型在2个细粒度动作识别数据集Diving48和Something-somethingV1上识别准确率分别达到86.0%和46.9%,分别使原始主干网络识别准确率提升3.8%和1.3%。实验结果表明,在只使用视频帧信息作为输入的情况下,本模型达到与现有基于Transformer和三维卷积神经网络(3D CNN)算法相当的识别准确率。

基于时空关联的车联网假位置筛选算法

目前,车联网位置隐私保护方法没有充分考虑到不同的时间段内假位置查询概率的变化,同时忽视了对于连续时间背景下同一位置单元每天的用户访问量变化趋势有较大差异的问题。针对上述缺陷,提出空间敏感度度量标准,在选择假位置时综合考虑位置语义相似度提出语义-空间敏感度期望的筛选指标,生成满足分时间段查询概率、匿名区域面积要求的假位置集合。实验结果证明了新算法的可行性及有效性,能够适应当下的车联网位置隐私保护需求。

冷能梯级利用的港口多能微网双层不确定性经济调度

为有效挖掘港口液化天然气(LNG)冷能利用的低碳灵活性潜力,充分发挥多时间尺度协同优化效应,提出一种考虑LNG冷能梯级利用的港口多能微网(MEMG)鲁棒-随机双层不确定性经济调度模型。首先,考虑LNG深冷-中冷-浅冷等各个温区的低碳灵活性潜力,建立低温碳捕集-冷能发电-直接冷却的冷能梯级利用模型,并以此为基础形成捕集-存储-利用协同的碳处理流程。其次,根据等概率逆变换生成考虑预测误差时序相关性的风电场景,并基于Wasserstein距离的0-1规划模型进行场景削减。再次,针对风电预测误差随时间尺度增加而增大的特性,构建多时间尺度优化的鲁棒-随机双层不确定性经济调度模型,上层通过分布鲁棒优化保证日前预调度决策鲁棒性,下层通过随机优化保证日内滚动调度决策经济性。最后,仿真结果表明,所提考虑冷能梯级利用的鲁棒-随机双层调度模型在解决日前长时间尺度预测精度低与日内短时间尺度易陷入局部最优矛盾的同时,可赋予港口MEMG更多经济性、低碳性及供电灵活性。