摘要：【中文】本发明提供了一种河流自然清洁系统，涉及水利疏浚工程技术领域，该装置架设在河流中以及河流两岸，包括截流分流机构、砂子收集引出机构、石料收集输送机构，其中，砂子收集引出机构和石料收集输送机构分别设置两套，对称设置在截流分流机构两侧，本发明可充分利用河流落差产生的势能，并辅助螺旋输送机和卷扬机等设备，在耗能较少的情况下实现砂石的清理，有利于江河环境的维护，并且清理的砂石可直接用于建筑行业，社会效益和经济效益并存；系统本身的规模可依据需求进行调整，适应性较强。 【EN】The invention provides a natural river cleaning system, which relates to the technical field of water conservancy dredging engineering, and is erected in a river and on two banks of the river, and comprises an interception and diversion mechanism, a sand collection and extraction mechanism and a stone collection and conveying mechanism, wherein the sand collection and extraction mechanism and the stone collection and conveying mechanism are respectively provided with two sets and are symmetrically arranged on two sides of the interception and diversion mechanism; the scale of the system can be adjusted according to the requirement, and the adaptability is strong.

摘要：【中文】本发明提供了一种基于图的动态分类器选择方法，包括以下步骤：生成分类器池‑分类器能力评估‑分类器序列选择‑决策层融合。传统的分类器序列选择可分为动态方法和静态方法，其中动态方法可根据测试样本的具体情况选择出不同的分类器序列对样本进行识别。然而这种方法容易受到分类器邻域内样本的干扰。因此本发明在此基础上重构样本特征，并使用测地距离计算样本与样本之间的距离，为动态集成打下了坚实基础；通过构建必须连接图及不可连接图来评定分类器池中每个成员识别特定未知样本时可能表现出的能力，并根据其得分对分类器排序，进而达到动态选择的目的；抑制测试样本邻域构成方式不合理带来的评估误差。 【EN】The invention provides a dynamic classifier selection method based on a graph, which comprises the following steps: generating a classifier pool-classifier capability evaluation-classifier sequence selection-decision layer fusion. Traditional classifier sequence selection can be divided into a dynamic method and a static method, wherein the dynamic method can select different classifier sequences to identify samples according to the specific situation of the test samples. However, this method is susceptible to interference from samples in the neighborhood of the classifier. Therefore, the invention reconstructs the sample characteristics on the basis, and calculates the distance between the samples by using the geodesic distance, thereby laying a solid foundation for dynamic integration; the capacity which is possibly shown when each member in the classifier pool identifies a specific unknown sample is evaluated by constructing a necessary connection graph and a non-connection graph, and the classifiers are sorted according to the scores of the members, so that the aim of dynamic selection is fulfilled; and inhibiting the evaluation error caused by unreasonable neighborhood constitution mode of the test sample.

摘要：【中文】本发明公开了一种基于对抗多任务学习的医疗咨询命名实体识别方法。所述方法包括以下步骤：采集医疗咨询数据，对医疗咨询数据进行预处理，并对其中一部分数据进行实体的标注，得到有标注的医疗咨询数据；构建双向语言模型和掩码语言模型，利用无标注的医疗咨询数据，分别预训练双向语言模型和掩码语言模型；将双向语言模型和掩码语言模型的预训练特征引入到命名实体识别模型；对命名实体识别模型进行对抗多任务训练，得到训练好的命名实体识别模型；输入一段文本到训练好的命名实体识别模型的目标标注模型中，实现文本命名实体识别。本发明引入了迁移学习、对抗学习、多任务学习等技术，有效地提高了医疗咨询文本命名实体识别的效果。 【EN】The invention discloses a medical consultation named entity identification method based on anti-multitask learning. The method comprises the following steps: acquiring medical consultation data, preprocessing the medical consultation data, and carrying out entity marking on a part of the data to obtain marked medical consultation data; constructing a bidirectional language model and a mask language model, and respectively pre-training the bidirectional language model and the mask language model by using the medical consultation data without labels; introducing pre-training characteristics of a bidirectional language model and a mask language model into a named entity recognition model; carrying out anti-multi-task training on the named entity recognition model to obtain a trained named entity recognition model; and inputting a section of text into a target labeling model of the trained named entity recognition model to realize the recognition of the named entity of the text. The invention introduces the technologies of transfer learning, counterstudy, multi-task learning and the like, and effectively improves the effect of identifying the named entities of the medical consultation texts.

摘要：【中文】本发明公开了一种基于目标动态训练的深度神经网络人脸表情识别方法。所述方法包括以下步骤：输入训练样本集和测试样本集；采用多个损失函数作为训练的目标函数；将损失函数排序；根据排序后的损失函数依次在训练样本集上重复训练深度神经网络，获得表情识别模型；根据表情识别模型对输入的测试样本进行表情分类。本发明让模型在初期能尽快拟合训练数据，在后期通过多个损失的动态权重变化拉大不同类别的特征空间距离以达到更好的泛化性能。本发明能够有效地提升神经网络在多个人脸表情识别数据集上的准确率。 【EN】The invention discloses a deep neural network facial expression recognition method based on target dynamic training. The method comprises the following steps: inputting a training sample set and a test sample set; adopting a plurality of loss functions as a training target function; sorting the loss functions; repeatedly training the deep neural network on the training sample set in sequence according to the sorted loss function to obtain an expression recognition model; and carrying out expression classification on the input test sample according to the expression recognition model. The method ensures that the model can fit the training data as soon as possible in the initial stage, and enlarges the characteristic space distances of different types through a plurality of lost dynamic weight changes in the later stage so as to achieve better generalization performance. The method can effectively improve the accuracy of the neural network on a plurality of facial expression recognition data sets.

摘要：【中文】本发明涉及云计算技术领域，公开了一种基于云上用户日志的自适应过滤方法及装置。从多个主流的云操作平台上收集用户行为日志，并存储下来作为日志数据源；将不同云操作平台上的日志转换成统一格式，将日志按照种类解析分割，进行日志归类；在过滤之前将日志条目聚类，用EM算法给出最优聚类簇的数量K，将得到的K带入K‑means算法，对日志进行聚类；根据聚类结果将噪音日志丢弃，将丢弃后留下的有效数据按照原始产生顺序聚集在一起，并去除重复日志条目。上述方案在解析日志格式的基础上过滤掉无用的用户日志，过滤的核心是将比较大的聚簇过滤掉，将不断改变聚簇的日志保留。同时此方案中运用了大数据的聚类算法来提高日志过滤的高效性。 【EN】The invention relates to the technical field of cloud computing, and discloses a self-adaptive filtering method and device based on a user log on a cloud. Collecting user behavior logs from a plurality of mainstream cloud operation platforms, and storing the user behavior logs as log data sources; converting logs on different cloud operation platforms into a uniform format, analyzing and dividing the logs according to types, and classifying the logs; clustering log entries before filtering, giving the number K of the optimal clustering clusters by using an EM algorithm, substituting the obtained K into a K-means algorithm, and clustering the logs; and discarding the noise log according to the clustering result, clustering the effective data left after discarding together according to the original generation sequence, and removing repeated log entries. According to the scheme, useless user logs are filtered on the basis of analyzing the log format, the filtering core is that large clusters are filtered, and logs of constantly changing clusters are reserved. Meanwhile, the clustering algorithm of big data is applied to improve the efficiency of log filtering.

摘要：【中文】本发明涉及计算机信息系统技术领域，公开了基于图计算技术的云平台数据泄漏路径识别方法。包括数据泄漏触发集合设置、事件影响路径集合获取、数据泄漏行为路径查找几个步骤。上述技术方案先设置一个可能触发泄露的触发行为集合，再基于关系度构建子图的方法，筛选出资源和最远行为，获取影响路径元素集合，最终采用起点、必经点、终点的方式找出数据泄露的行为路径；通过该过程能够发现隐藏在正常行为序列中的数据泄漏行为序列，最大限度还原数据泄漏场景的行为。 【EN】The invention relates to the technical field of computer information systems, and discloses a cloud platform data leakage path identification method based on a graph computing technology. The method comprises the steps of setting a data leakage trigger set, acquiring an event influence path set and searching a data leakage behavior path. According to the technical scheme, a triggering behavior set which can trigger leakage is set, a method for constructing a subgraph based on the relation degree is adopted, resources and the farthest behavior are screened out, a set of influencing path elements is obtained, and finally a behavior path of data leakage is found out in a mode of adopting a starting point, a must-pass point and an end point; through the process, the data leakage behavior sequence hidden in the normal behavior sequence can be found, and the behavior of a data leakage scene is restored to the maximum extent.