联合实体边界检测的命名实体识别方法

针对传统命名实体识别方法无法有效利用实体边界信息的问题,提出一种联合实体边界检测的命名实体识别方法,即将实体边界检测作为辅助任务,增强模型对实体边界的判断能力,进而提升模型对实体的识别效果。首先,利用Bert预训练语言模型对原始文本进行特征嵌入获取词向量,并引入自注意力机制增强词对上下文信息的利用;其次,在命名实体识别任务的基础上,添加实体边界检测辅助任务,增强模型对实体边界的识别能力;再次,对比联合实体边界检测的命名实体识别方法与基线方法的有效性,并对测试结果进行消融实验;最后,进行样例分析,分析损失权重β对实体边界检测的影响。实验结果表明,在英文社交媒体数据集Twitter-2015上,联合实体边界检测的命名实体识别方法相较于基线模型取得了更高的精准率、召回率和F1值,其中F1值达到了73.57%;并且,边界检测辅助任务提升了基线方法的检测效果。所提方法能有效利用实体边界信息,从而获得更好的实体识别效果,促进了人机交互系统的发展,对自然语言处理下游任务有重要意义。     

基于轻量化深度学习网络的数字仪表识别

数字式仪表常用于变电站、工厂等生产环境,是一种直观的设备监测仪器。然而当前数字式仪表的读取方式还依赖于人工巡检,手动记录等,这些传统的巡检方式来监测设备的运行状态大大降低了巡检效率。为了实现传统行业的数字化转型,本文提出基于轻量化深度学习的数字仪表识别方法,通过改进的YOLOv5的目标检测框架,针对数字仪表目标区域在整张图片大小不一致的情况,提出对于感兴趣区域（ROI）的迭代目标检测方法,首次检测将感兴趣区域进行检测并切割统一到相同的尺度,随后迭代检测网络针对感兴趣区域内的字符进行检测并分类,以达到精确读数的目的。为提升多尺度检测性能,本文采用Res2Net模块主干网络中的的残差模块。采用GIoU取代通用的IoU作为位置损失函数加速模型训练效果的收敛。实验表明,改进后的框架实现了99.62%的准确率和99.55%的召回率,相比基线网络分别提升了12.72%和5.85%。通过将框架在边缘计算平台上的终端部署,在实际生产中取代了人工巡检,实现了商业化运行。     

数据驱动下交通异常行为的双层识别模型

为实现对车辆异常行为的准确识别,提高车辆行驶的安全性,提出了一种基于支持向量机(support vector machine, SVM)、监督学习与长短期记忆(long short term memory, LSTM)深度学习的交通异常驾驶行为双层识别模型。首先,对车辆轨迹数据筛除和滤波,构建异常行为数据集;其次,从异常行为轨迹特征中提取出特定异常行为的特征标签,并人为标定在训练集中;再次,构建SVM模型对训练集进行粗识别,基于SVM的二分法原理,从测试集中筛选出异常行为;最后,通过LSTM时间序列模型构建具体种类的异常行为模型,并通过深度学习的方法,从异常行为数据中细分为蛇形驾驶、急速变向、侧滑、大半径转弯、快速U型转弯、急刹车等具体的异常驾驶行为。本次实验选用下一代仿真(next generation simulation, NGSIM)数据中US-101高速公路和peachtree城市道路的数据集的轨迹数据验证SVM和LSTM双层识别模型的性能,包括均方根误差、识别准确率等。结果表明,构建的双层识别模型在第一层有98%的识别准确率,第二层有超过80%的识别准确率,可以较为准确地识...     

含有陡峭势阱和凹凸非线性项的Kirchhoff型问题的多重正解*

在这篇文章中, 作者研究涉及凹凸非线性项的Kirchhoff型问题-(a + b ∫R3|▽u|²dx) Δu + λV (x)u = μf(x)|u|^q?2u + |u|^p?2u, x ∈ R³,u ∈ H¹(R³),其中a,b > 0 是常数, λ, μ > 0 是参数, 1 < q < 2, 4 < p < 6 且 V 是一个非负连续位势. 在f(x) 和 V 的合适条件下,此问题正解的存在性和集中性能够通过Nehari 流形和Ekeland 变分原理得到.     

Shear-wave splitting and anisotropy in crust of Dabieshan orogenic belt

The shear-wave splitting was analyzed on the deep seismic sounding profile crossing the Dabieshan area. The direction difference between Sg and SmS fast wave polarization indicates that the modem regional tectonic stress field in the Dabieshan area exists only in uppcr crust, while the crack alignment caused by the Triassic collision between the North China and the Yangtze blocks had been preserved in mid-lower curst.     

木兰溪两岸浅井水对紫露草四分体微核率的影响

采用紫露草四分体微核技术,对莆田木兰溪下游河溪水和饮用的浅井水进行致突变性监测。结果表明,１０份河溪水能明显诱发紫露草四分体微核率的增高;与对照组相比１０份浅井水也均不同程度提高了四分体微核效应,８份已达到显著或极显著水平（Ｐ〈０．０５或Ｐ〈０．０１）,且与诱发的紫竹梅四分体微核效应显著相关,ｒ＝０．９７７７。说明浅井水内存在致植物细胞染色体畸变物质,提示当地居民上消化道肿瘤高发与饮水污染有一定相     

甲醇燃料车尾气净化催化剂的研究Ⅰ.甲醇深度氧化用钯催化剂上吸附物种的原位红外研究

利用原位红外技术研究了ＣＨ３ＯＨ、ＣＯ、Ｏ２等在Ｐｄ／γＡｌ２Ｏ３上的吸附情况及ＣＨ３ＯＨ和Ｏ２共吸附时表面物种的变化，结果表明：表面纯净的Ｐｄ／γＡｌ２Ｏ３催化剂对甲醇具有很好的解离吸附性能，在无氧条件下，ＣＨ３ＯＨ可在其上脱氢产生ＨＣＨＯ及ＣＯ。Ｐｄ对Ｏ２的吸附力强，氧均以解离方式吸附于Ｐｄ／γＡｌ２Ｏ３上。当吸附的［ＣＨ３Ｏ］与［Ｏ］在Ｐｄ／γＡｌ２Ｏ３上相互作用时，可产生ＨＣＨＯ、ＣＯ、甲二氧基、甲酸根等中间物种     

K对Mn—Co—O的结构及氧化活性的影响

利用XRD,TPD,TFD-MS及催化氧化反应等实验技术,研究了K对Mn-Co-O的结构及氧化活性的影响。XRD结果表明,Mn-Co-O系已形成反尖晶石型的MnCo_2O_4结构;K-Mn-Co-O系中除MnCo_2O_4型结构外,还有新相KMnO_2存在。根据O_2的TPD-MS及吡啶的TPD结果,Mn-Co-O中添加适量的K能提高供氧活性和增加供氧数目,但酸中心数目减少,强度降低。K对Mn-Co—O催化剂氧化活性的影响随反应物分子结构的不同而异,对乙酸乙脂、苯甲酰氯等极性有机物的完全氧化反应,K能提高氧化活性,但对苯、正己烷等非极性有机物的完全氧化反应,K使其活性降低。     

Detection of single nucleotide polymorphisms using electrospray ionization mass spectrometry: validation of a one-well assay and quantitative pooling studies

Single nucleotide polymorphisms (SNPs) are currently being mapped and databased at a remarkable pace, providing a viable means for understanding disease susceptibility, differential drug response and human evolution. Consequently, there is an increasing demand for SNP genotyping technologies that are simple, rapid, cost effective and readily amenable to automation for high-throughput analyses. In this study, we improved the Survivor Assay, a SNP detection method based on electrospray ionization mass spectrometry (ESI-MS), with several developments. One improvement is the development of a one-well assay, requiring no off-line purification of the polymerase chain reaction product, achieved by simple addition of reagent solution into a single well. Another is the on-line separation of magnesium and dideoxynucleotides using an in-house made monolithic metal chelating column, eliminating any off-line sample preparation prior to mass spectrometric analysis. Here the Survivor Assay is extended from a proof-of-principle concept to a validated method by genotyping six SNPs from five different regions of human genomic DNA in 55 individual samples with 100% accuracy. This improved Survivor Assay eliminates the tedious and time-consuming steps of sample preparation, minimizes sample handing and offers a high-throughput analysis of SNPs by ESI-MS. The current combined preparation and analysis time is 2 min per sample. The simplicity of this method has potential for full automation and parallel chromatography and, thus, reduced analysis time. In addition, we have adapted the Survivor Assay for quantitative SNP analysis in pooled DNA samples. The capabilities and sensitivity of this approach were evaluated. We demonstrate that an allele occurring at a frequency of 2% can consistently be quantitated.