基于领域词典与CRF双层标注的中文电子病历实体识别

医疗实体识别是电子病历文本信息抽取的基本任务。针对中文电子病历文本复合实体较多、实体长度较长、句子成分缺失严重、实体边界不清的语言特点以及标注语料难以获取的现状,提出了一种基于领域词典和条件随机场（CRF）的双层标注模型。该模型通过对外部资源的统计分析构建医疗领域词典,再结合条件随机场,进行了两次不同粒度的标注,将领域词典识别的准确性和机器学习的自动性融为一体,从中文电子病历文本中识别出疾病、症状、药品、操作四类医疗实体。该模型在测试数据中的宏精确率为96.7%、宏召回率为97.7%、宏F1值为97.2%。同时对比分析了采用注意力机制的深度神经网络的识别效果,因受到领域数据集大小的限制,在该测试数据集中后者表现不佳。实验结果表明了该双层标注模型对中文医疗实体识别的高效性。      

基于BiLSTM-Attention的钢板表面手写板号识别算法

国内钢铁企业生产厂的信息化物料跟踪大都依赖于钢板号。由于生产流程复杂,急需高准确率的板号在线识别技术。自然场景下机器喷号的识别技术较成熟,但复杂场景下的手写板号难以实现自动识别。针对复杂工作场景下钢板表面手写板号特点,提出一种以BiLSTM-Attention为主体结构的深度学习算法。首先结合复杂场景,对图像数据进行预处理,保证模型输入图片质量;然后利用残差神经网络(ResNet)提取图片特征、利用双向长短期记忆网络(BiLSTM)提取基于图像的序列特征;最后基于注意力机制捕获序列内的信息流,对每个字符的特征进行整合,形成文本特征向量以预测输出序列。经现场测试,实现钢板表面手写板号识别任务准确率达86.15%,结果表明算法可行有效,满足实际生产需求。     

基于空洞卷积神经网络与注意力机制GRU的滚动轴承故障诊断

针对传统故障诊断方法需要人工提取特征的不足,以及大数据下滚动轴承故障振动信号自适应特征提取与智能诊断问题,利用空洞卷积神经网络(DCNN)可以在不增加计算量的基础上兼顾不同尺度空间特征的能力、门控循环单元(GRU)善于从动态变化的序列数据中学习到时间上的关联性的能力,提出了一种将DCNN、注意力机制和GRU多路径融合的端到端故障诊断方法。首先利用DCNN从原始数据中自动提取时序信号特征,然后将注意力机制(Attention)的GRU通路和DCNN通路进行融合,最后将提取到的特征融合之后送入分类层进行分类。试验结果表明,所提方法的诊断准确率平均为98.75%,高于比较方法,更加适用于滚动轴承故障诊断。     

基于BiLSTM的公共安全事件触发词识别

提出基于双向长短期记忆网络（bidirectional long short-term memory,BiLSTM）和前向神经网络的融合模型完成公共安全事件的触发词识别任务.首先通过BiLSTM提取整段文本的高层语义特征,避免了以往机器学习方法需要人工提取特征的问题,其次采用特征拼接并在前向神经网络中识别并分类事件触发词.实验结果表明相较于基准模型,本文方法在中文突发事件语料库（Chinese emergency corpus,CEC）上取得了更为突出的性能,Micro-F1值为78.47%.此外本文讨论了不同拼接特征在触发词识别任务中的重要性,对文本分析中3类特征（词性、句法、实体）的重要程度进行了比较和分析,得出句法特征对于事件触发词识别任务助益最大的结论.      

基于深度学习的高效火车号识别

基于高性能的YOLOv3目标检测算法,提出一种分阶段高效火车号识别算法。整个识别过程分为两个阶段:第一阶段在低分辨率全局图像中检测出火车号区域位置;第二阶段在局部高分辨率图像中检测出组成火车号的字符,根据字符的空间位置关系搜索得到12位火车号,并利用每个字符的识别置信度及火车号编码规则进行校验得到最终火车号。另外,本文提出一种结合批一化因子和滤波器相关度的剪枝算法,通过对两个阶段检测模型的剪枝,在保证识别准确率不降（实验中略有提升）的条件下降低了存储空间占用率和计算复杂度。在现场采集的1072幅火车号图像上的实验结果表明,本文提出的火车号识别算法达到了96.92%的整车号识别正确率,平均识别时间仅为191 ms。      

电子病历在医疗纠纷中的常见问题及对策

电子病历应用于医院管理,不仅提高了现代医院工作效率,但也带来了医疗纠纷的隐患。针对电子病历在医疗纠纷中的常见问题,提出通过提高法律意识、加强复制管理、完善系统设计、严格质控制度等措施,消除因电子病历的问题可能引起的医疗纠纷隐患,从而保证医疗安全和医疗质量。     

电子病历中的临床路径分析与探讨

目的:探讨电子病历在临床路径管理中的应用.方法:随机抽取200份我院住院患者的资料,将其均分为两组,研究组给予电子病历管理,对照组给予传统电子病历管理,对两组患者的满意度进行随访,统计患者的满意率.结果:研究组患者满意率高达98%,对照组满意率仅为77%,经统计学处理分析X2=20.16,P<0.01,两组比较差异显著,具有统计学意义.结论:电子病历的实施和推广可有效提高临床路径管理的水平,既改善了医疗护理质量又降低了患者医疗费用,同时还增强了医院的竞争力.     

基于STCS-YOLO的带钢表面缺陷检测算法

带钢表面缺陷检测技术是高质量带钢产品生产的重要技术之一。针对以往带钢表面缺陷检测中存在漏检、定位不准、小尺度缺陷目标检测能力较差的问题，提出了一种改进YOLOv5的带钢表面缺陷检测算法(STCS-YOLO)。首先，在特征融合网络的输出部分采用Swin Transformer模块与原有C3模块相融合，增强对全局特征信息的交互与复用，显著提高了对小尺度缺陷目标的检测能力；其次，采用一种轻量级上采样算子CARAFE来替换传统上采样操作，以更好地恢复缺陷信息，提高对带钢表面缺陷的识别精度；最后，在特征提取网络中嵌入3-D权值注意力机制SimAM,以加强对前景特征信息的关注能力，提高对缺陷目标的强辨识能力。试验结果表明，所提算法在NEU-DET数据集上均值平均精度P_mA达到了79.7%,比原网络提高了3.9个百分点，并且在模型权重与计算复杂度几乎不变的情况下，单帧检测时间达到了10.9 ms,基本能够满足带钢表面缺陷准确、快速的检测需求。本研究提出的带钢表面缺陷检测算法为生产整洁、无瑕的高质量带钢产品奠定了技术基础。     

基于双特征融合的动态图像分析算法

为解决汽车碰撞实验过程进行测量和记录数据困难的问题,提出一种能从汽车碰撞动态图像中检测、识别和追踪标志目标的算法.该算法采用二值图像同或相关法分离出目标和背景;在找出感兴趣区域后提取相邻帧目标的坐标和纹理进行特征匹配;给出了对匹配量化值进行加权平均融合的策略,并由等错误率最小准则确定融合的最佳权系数;在融合量化值定义为相似度的基础上,通过决策阈值对相邻帧目标间的最大相似度组合进行识别;从而进一步提出了用同构映射原则来判断相邻帧目标的最佳配对.实验结果表明,该算法对相邻帧目标的配对准确率比传统单特征法提高5%,能更有效的对目标进行追踪.     

基于YOLOX-drone的反无人机系统抗遮挡目标检测算法

为解决现实场景下无人机目标被部分遮挡,导致不易检测问题,本文提出了基于YOLOX-S改进的反无人机系统目标检测算法YOLOX-drone.首先,建立无人机图像数据集;其次,搭建YOLOX-S目标检测网络,在此基础上引入坐标注意力机制,来增强无人机的目标图像显著度,突出有用特征抑制无用特征;然后,再去除特征融合层中自下而上的路径增强结构,减少网络复杂度,并设计了自适应特征融合网络结构,增强有用特征的表达能力,抑制干扰,提升检测精度.在DUT-AntiUAV数据集上的测试结果表明：YOLOX-drone与YOLOX-S、YOLOv5-S和YOLOX-tiny相比,平均准确率(IoU=0.5)提升了3.2%、4.7%和10.1%;在自建的无人机图像数据集上的测试结果表明：YOLOX-drone与原YOLOX-S目标检测模型相比,在无遮挡、一般遮挡、严重遮挡情况下,平均准确率(IoU=0.5)分别提高了2.4%、2.1%和6.4%,验证了改进的算法具有良好的抗遮挡检测能力.     

基于数控机床设备故障领域的命名实体识别

为了给数控机床故障的精准诊断提供保障，延长数控机床使用周期，以数控机床历史维修记录为研究对象，对数控机床设备故障领域的命名实体识别进行了研究。在分析历史维修记录中的故障描述特点后，提出了一种基于双向长短期记忆网络（Bidirectional long short-term memory, BLSTM）与具有回路的条件随机场（Conditional random field with loop, L-CRF）相结合的命名实体识别方法。首先，对输入语句进行分词和标注，使用Word2vec中的Skip-gram模型对标注语料进行预训练，将其生成的字向量通过词嵌入层转化为字向量序列；然后，将字向量序列输入BLSTM学习长期依赖信息；最后将句子表达输入L-CRF获取全局最优序列。实验结果表明，该方法明显优于其他命名实体识别方法，为数控机床设备的智能检修与实时诊断任务打下了坚实的基础。      

Mental health practitioners and HIPAA.

The privacy and security of patients' medical records continue to challenge mental health practitioners in an ever-increasing electronic environment. Although practitioners were to be in compliance of the 1996 Health Insurance Portability and Accountability Act (HIPAA) by April 20, 2006, many practitioners still struggle to understand the nuances of the regulations. This article will cover the areas of HIPAA that relate specifically to the mental health practitioner who treats both adults and children. The article begins with important definitions such as “covered entity,” “individually identified health information,” and “electronic transactions.” Establishing that many mental health practitioners likely meet the definition of covered entities, the article details The Privacy Rule and The Security Rule. Included are issues concerning consents, authorizations, and objections along with enforcement of HIPAA. Also changes in enforcement to HIPAA by the Health Information Technology for Economic and Clinical Health Act (HITECH), a part of the American Recovery and Reinvestment Act of 2009 are discussed. Examples of violations and resulting enforcement will help practitioners better understand the regulations and how best to comply with these regulations. (PsycINFO Database Record (c) 2011 APA, all rights reserved)     

Reading spaced and unspaced Chinese text: Evidence from eye movements.

Native Chinese readers' eye movements were monitored as they read text that did or did not demark word boundary information. In Experiment 1, sentences had 4 types of spacing: normal unspaced text, text with spaces between words, text with spaces between characters that yielded nonwords, and finally text with spaces between every character. The authors investigated whether the introduction of spaces into unspaced Chinese text facilitates reading and whether the word or, alternatively, the character is a unit of information that is of primary importance in Chinese reading. Global and local measures indicated that sentences with unfamiliar word spaced format were as easy to read as visually familiar unspaced text. Nonword spacing and a space between every character produced longer reading times. In Experiment 2, highlighting was used to create analogous conditions: normal Chinese text, highlighting that marked words, highlighting that yielded nonwords, and highlighting that marked each character. The data from both experiments clearly indicated that words, and not individual characters, are the unit of primary importance in Chinese reading. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Attribution difficulty and memory for attribution-relevant information.

Compared the processing and retrieval of attribution-relevant information when the attributional inference is easy or difficult to make. Ss attributed behavioral events to the person or to the situation, based on several items of context information. Each context sentence implied either the person or the entity as causal agent. When the attributional inference was difficult to make (an equal number of context sentences implied actor and entity as the causal agent), Ss recalled more of the behavioral events, recalled more context sentences, and were less confident in their attributions than when the attributional inference was easy to make (most context sentences implied the same causal agent). Ss also recalled context information that was implicationally incongruent with the majority of the other context sentences with a higher probability than when that same information was implicationally congruent. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

基于改进遗传算法优化Elman网络的板形识别方法

针对目前的板形缺陷识别方法精度不高、识别速度慢的问题,根据Elman神经网络模型可以反映系统动态特性,而且可以逼近任意非线性函数的特点,提出了一种利用改进的遗传算法优化Elman神经网络,使其泛化能力强、学习速度快、识别精度高,并建立板形缺陷模式识别模型的方法。为了验证该方法的识别能力,在隐层节点数与学习次数相同的条件下,分别与遗传算法优化的Elman网络和BP网络模型进行板形识别仿真对比分析。试验结果表明,改进遗传算法优化的Elman神经网络模型对板形缺陷识别精度高于BP网络等模型,并且具有收敛速度快的优点。     

A neural basis for the retrieval of conceptual knowledge

Both clinical reports and systematic neuropsychological studies have shown that patients with damage to selected brain sites develop defects in the retrieval of conceptual knowledge for various concrete entities, leading to the hypothesis that the retrieval of knowledge for entities from different conceptual categories depends on partially segregated large-scale neural systems. To test this hypothesis, 116 subjects with focal, unilateral lesions to various sectors of the telencephalon, and 55 matched controls, were studied with a procedure which required the visual recognition of entities from three categories--unique persons, non-unique animals and non-unique tools. Defective recognition of persons was associated with maximal lesion overlap in right temporal polar region; defective recognition of animals was associated with maximal lesion overlap in right mesial occipital/ventral temporal region and also in left mesial occipital region; and defective recognition of tools was associated with maximal lesion overlap in the occipital-temporal-parietal junction of the left hemisphere. The findings support the hypothesis that the normal retrieval of knowledge for concrete entities from different conceptual domains depends on partially segregated neural systems. These sites may operate as catalysts for the retrieval of the multidimensional aspects of knowledge which are necessary and sufficient for the mental representation of a concept of a given entity.     

弱光照条件下交通标志检测与识别

针对弱光照条件下交通标志易发生漏检和定位不准的问题，本文提出了增强YOLOv3（You only look once）检测算法，一种实时自适应图像增强与优化YOLOv3网络结合的交通标志检测与识别方法。首先构建了大型复杂光照中国交通标志数据集；然后针对复杂的弱光照图像提出自适应增强算法，通过调整图像亮度和对比度强化交通标志与背景之间的差异；最后采用YOLOv3网络框架检测交通标志。为了降低先验锚点框设置精度以及图像中背景与前景比例严重失衡对检测精度造成的影响，优化了先验锚点框聚类算法和网络的损失函数。对比实验结果表明，在实时性大致相当的情况下，本文提出的增强YOLOv3检测算法较标准YOLOv3算法对交通标志有更高的回归精度和置信度，召回率和准确率分别提高0.96%和0.48%。      

基于改进降噪自编码器半监督学习模型的热轧带钢水梁印识别算法

针对水梁印识别困难且工作量大问题,提出一种基于改进降噪自编码器半监督学习模型的热轧带钢水梁印识别算法。该算法在降噪自编码器（Denoising auto-encoder, DAE）的基础上对编码层的每一层添加随机噪声,在隐藏层后添加分类层,并对数据添加伪标签,在解码的同时进行分类训练,使得DAE具有半监督学习能力。通过提取热轧带钢粗轧出口温度数据中的温差特征,用相应特征对模型进行训练。实验结果表明,算法能够准确识别出带钢的水梁印,在模型精确度上,与主流分类识别模型对比,提出的模型在带标签样本数量较小时,分类精度相比其他模型高5.0%～10.0%;在带标签样本数量较大时,提出的模型分类精度达到93.8%,现场能够根据模型的识别结果提高生产效率。