The coherence cube technology has become an important technology for the seismic attribute interpretation, which extracts the discontinuities of the events through analyzing the similarities of adjacent seismic channe...The coherence cube technology has become an important technology for the seismic attribute interpretation, which extracts the discontinuities of the events through analyzing the similarities of adjacent seismic channels to identify the fault form. The coherence cube technology which uses constant time window lengths can not balance the shallow layers and the deep layers, because the frequency band of seismic data varies with time. When analyzing the shallow layers, the time window will crossover a lot of events, which will lead to weak focusing ability and failure to delineate the details. While the time window will not be long enough for analyzing deep layers, which will lead to low accuracy because the coherences near the zero points of the events are heavily influenced by noise. For solving the problem, we should make a research on the coherence cube technology with self-adaptive time window. This paper determines the sample points' time window lengths in real time by computing the instantaneous frequency bands with Wavelet Transformation, which gives a coherence computing method with the self-adaptive time window lengths. The result shows that the coherence cube technology with self-adaptive time window based on Wavelet Transformation improves the accuracy of fault identification, and supresses the noise effectively. The method combines the advantages of long time window method and short time window method.展开更多
针对在生物医学领域中命名实体数据标注成本高、难以获取大量有标签数据的问题,提出了一个两阶段学习框架实现低资源下的中文生物医学命名实体识别。在第一阶段,利用Word2Vec和BERT为基础模型预训练并进行微调,获得特定领域的词向量表示...针对在生物医学领域中命名实体数据标注成本高、难以获取大量有标签数据的问题,提出了一个两阶段学习框架实现低资源下的中文生物医学命名实体识别。在第一阶段,利用Word2Vec和BERT为基础模型预训练并进行微调,获得特定领域的词向量表示;在第二阶段,将生成的词向量输入到由BiLSTM和条件随机场(Conditional random field,CRF)组成的神经网络中用于最终任务的训练。本文在Yidu-S4k数据集进行实验,结果表明本文算法在少量标签的情况下取得80.94%的准确率,具有较优性能。展开更多
文摘The coherence cube technology has become an important technology for the seismic attribute interpretation, which extracts the discontinuities of the events through analyzing the similarities of adjacent seismic channels to identify the fault form. The coherence cube technology which uses constant time window lengths can not balance the shallow layers and the deep layers, because the frequency band of seismic data varies with time. When analyzing the shallow layers, the time window will crossover a lot of events, which will lead to weak focusing ability and failure to delineate the details. While the time window will not be long enough for analyzing deep layers, which will lead to low accuracy because the coherences near the zero points of the events are heavily influenced by noise. For solving the problem, we should make a research on the coherence cube technology with self-adaptive time window. This paper determines the sample points' time window lengths in real time by computing the instantaneous frequency bands with Wavelet Transformation, which gives a coherence computing method with the self-adaptive time window lengths. The result shows that the coherence cube technology with self-adaptive time window based on Wavelet Transformation improves the accuracy of fault identification, and supresses the noise effectively. The method combines the advantages of long time window method and short time window method.
文摘针对在生物医学领域中命名实体数据标注成本高、难以获取大量有标签数据的问题,提出了一个两阶段学习框架实现低资源下的中文生物医学命名实体识别。在第一阶段,利用Word2Vec和BERT为基础模型预训练并进行微调,获得特定领域的词向量表示;在第二阶段,将生成的词向量输入到由BiLSTM和条件随机场(Conditional random field,CRF)组成的神经网络中用于最终任务的训练。本文在Yidu-S4k数据集进行实验,结果表明本文算法在少量标签的情况下取得80.94%的准确率,具有较优性能。