Point Cloud Classification Using Content-Based Transformer via Clustering in Feature Space

Recently, there have been some attempts of Transformer in 3D point cloud classification. In order to reduce computations, most existing methods focus on local spatial attention,but ignore their content and fail to establish relationships between distant but relevant points. To overcome the limitation of local spatial attention, we propose a point content-based Transformer architecture, called PointConT for short. It exploits the locality of points in the feature space(content-based), which clusters the sampled points with similar features into the same class and computes the self-attention within each class, thus enabling an effective trade-off between capturing long-range dependencies and computational complexity. We further introduce an inception feature aggregator for point cloud classification, which uses parallel structures to aggregate high-frequency and low-frequency information in each branch separately. Extensive experiments show that our PointConT model achieves a remarkable performance on point cloud shape classification. Especially, our method exhibits 90.3% Top-1 accuracy on the hardest setting of ScanObjectN N. Source code of this paper is available at https://github.com/yahuiliu99/PointC onT.

BSTFNet:An Encrypted Malicious Traffic Classification Method Integrating Global Semantic and Spatiotemporal Features

While encryption technology safeguards the security of network communications,malicious traffic also uses encryption protocols to obscure its malicious behavior.To address the issues of traditional machine learning methods relying on expert experience and the insufficient representation capabilities of existing deep learning methods for encrypted malicious traffic,we propose an encrypted malicious traffic classification method that integrates global semantic features with local spatiotemporal features,called BERT-based Spatio-Temporal Features Network(BSTFNet).At the packet-level granularity,the model captures the global semantic features of packets through the attention mechanism of the Bidirectional Encoder Representations from Transformers(BERT)model.At the byte-level granularity,we initially employ the Bidirectional Gated Recurrent Unit(BiGRU)model to extract temporal features from bytes,followed by the utilization of the Text Convolutional Neural Network(TextCNN)model with multi-sized convolution kernels to extract local multi-receptive field spatial features.The fusion of features from both granularities serves as the ultimate multidimensional representation of malicious traffic.Our approach achieves accuracy and F1-score of 99.39%and 99.40%,respectively,on the publicly available USTC-TFC2016 dataset,and effectively reduces sample confusion within the Neris and Virut categories.The experimental results demonstrate that our method has outstanding representation and classification capabilities for encrypted malicious traffic.

Comparative study on the performance of ConvLSTM and ConvGRU in classification problems-taking early warning of short-duration heavy rainfall as an example

卷积长短期记忆单元ConvLSTM和卷积门控循环单元ConvGRU是两种广泛应用的深度学习单元,通过将循环机制与卷积运算相结合,常常用于时空序列的预测.为了明确上述两种模型的收敛速度和分类能力,需要使用相同的模型架构对相同的分类问题进行预测.本研究将北京短时强降水区级预警问题看作深度学习中的二分类问题,使用京津冀雷达网的组合反射率数据和北京区域内的自动气象站降雨数据进行深度学习模型的训练和评估.结果表明,ConvGRU的收敛速度比ConvLSTM快约25%.ConvLSTM和ConvGRU的预警性能随地区,时间,降雨强度的变化趋势相似,但大部分ConvLSTM的得分较高,少数情况下ConvGRU的得分较高.

Classification of Coalbed Methane Enrichment Units of Qinshui Basin Based on Geological Dynamical Conditions

Coalbed methane enrichment will be controlled by many good macro geological dynamical conditions; there is evident difference of enrichment grade in different area and different geological conditions.This paper has studied tectonic dynamical conditions, thermal dynamical conditions and hydraulic conditions, which affect coalbed methane enrichment in Qinshui basin.Coalbed methane enrichment units have been divided based on tectonic dynamical conditions of Qinshui basin,combined with thermal dynamical conditions and hydraulic conditions.

Classification and Denominationof Flow Units for Clastic Reservoirsof Continental Deposit

On the basis of other researchers' achievements and the authors' understanding of flow units, a proposal on classification and denomination of flow units for clastic reservoirs of continental deposit is put forward according to the practical need of oilfield development and relevant theories. The specific implications of development and geology are given to each type of flow units, which has provided a scientific basis for oil development.

Clastic compaction unit classification based on clay content and integrated compaction recovery using well and seismic data

Compaction correction is a key part of paleogeomorphic recovery methods. Yet, the influence of lithology on the porosity evolution is not usually taken into account. Present methods merely classify the lithologies as sandstone and mudstone to undertake separate porositydepth compaction modeling. However, using just two lithologies is an oversimplification that cannot represent the compaction history. In such schemes, the precision of the compaction recovery is inadequate. To improve the precision of compaction recovery, a depth compaction model has been proposed that involves both porosity and clay content. A clastic lithological compaction unit classification method, based on clay content, has been designed to identify lithological boundaries and establish sets of compaction units. Also, on the basis of the clastic compaction unit classification, two methods of compaction recovery that integrate well and seismic data are employed to extrapolate well-based compaction information outward along seismic lines and recover the paleo-topography of the clastic strata in the region. The examples presented here show that a better understanding of paleo-geomorphology can be gained by applying the proposed compaction recovery technology.

Three-dimensional Extension of the Unit-Feature Spatial Classification Method for Cloud Type

We describe how the Unit-Feature Spatial Classification Method(UFSCM) can be used operationally to classify cloud types in satellite imagery efficiently and conveniently.By using a combination of Interactive Data Language(IDL) and Visual C++(VC) code in combination to extend the technique in three dimensions(3-D),this paper provides an efficient method to implement interactive computer visualization of the 3-D discrimination matrix modification,so as to deal with the bi-spectral limitations of traditional two dimensional(2-D) UFSCM.The case study of cloud-type classification based on FY-2C satellite data (0600 UTC 18 and 0000 UTC 10 September 2007) is conducted by comparison with ground station data, and indicates that 3-D UFSCM makes more use of the pattern recognition information in multi-spectral imagery,resulting in more reasonable results and an improvement over the 2-D method.

Green Architecture for Dense Home Area Networks Based on Radio-over-Fiber with Data Aggregation Approach

The high-density population leads to crowded cities. The future city is envisaged to encompass a large-scale network with diverse applications and a massive number of interconnected heterogeneous wireless-enabled devices. Hence, green technology elements are crucial to design sustainable and future-proof network architectures. They are the solutions for spectrum scarcity, high latency, interference, energy efficiency, and scalability that occur in dense and heterogeneous wireless networks especially in the home area network （HAN）. Radio-over-fiber （ROF） is a technology candidate to provide a global view of HAN＇s activities that can be leveraged to allocate orthogonal channel communications for enabling wireless-enabled HAN devices transmission, with considering the clustered-frequency-reuse approach. Our proposed network architecture design is mainly focused on enhancing the network throughput and reducing the average network communications latency by proposing a data aggregation unit （DAU）. The performance shows that with the DAU, the average network communications latency reduces significantly while the network throughput is enhanced, compared with the existing ROF architecture without the DAU.

Statistical Tools for Estimation of Threshold Values at Data Classification Task Solution

The paper contains a summary of some results of original research total aggregates. The main idea is determining the boundaries of the groups for classification of fuzzy and threshold aggregates using the method of decomposing a mixture of probability distributions. The article presents the experience of partitions of a real aggregate as a finite mixture of probability distributions on private aggregates. Threshold value defined by the boundaries of private aggregates, will match the value of the phenomenon at the intersection of the curves of probability distributions, which extracted from the mixture. The proposed scheme of identification threshold aggregates has found practical application in the research of aggregate of Russian employees by level of payroll and establishing the optimal minimum value monthly wage. The official data of the Federal State Statistics Service were used.

Multi⁃Scale Dilated Convolutional Neural Network for Hyperspectral Image Classification

Aiming at the problem of image information loss,dilated convolution is introduced and a novel multi⁃scale dilated convolutional neural network(MDCNN)is proposed.Dilated convolution can polymerize image multi⁃scale information without reducing the resolution.The first layer of the network used spectral convolutional step to reduce dimensionality.Then the multi⁃scale aggregation extracted multi⁃scale features through applying dilated convolution and shortcut connection.The extracted features which represent properties of data were fed through Softmax to predict the samples.MDCNN achieved the overall accuracy of 99.58% and 99.92% on two public datasets,Indian Pines and Pavia University.Compared with four other existing models,the results illustrate that MDCNN can extract better discriminative features and achieve higher classification performance.

融合二连通模体结构信息的节点分类算法

节点表示学习将图结构数据信息编码到低维的潜在空间中,在节点分类、聚类、链路预测等机器学习任务中被广泛应用。在复杂网络中,节点与节点之间不仅存在直接相连的低阶结构,也存在以特殊连接模式形成的高阶结构,称为模体。提出一种融合二连通模体结构信息的节点分类算法(FMI),利用节点间高阶二连通模体信息学习节点表示,完成节点分类任务。首先,统计网络中的二连通模体,利用其中信息提出一个节点重要性的度量指标——模体比值。根据模体比值计算采样概率进行邻域采样;构造一个带权辅助图以融合网络节点连接的低阶关系与高阶关系,对节点进行加权邻域聚合以得到节点表示。在5个数据集Cora、Citeseer、Pubmed、Wiki和DBLP上执行节点分类任务,与5种经典基准算法进行对比,所提算法FMI在准确度和F1-分数等指标上表现良好。

水泥稳定碎石基层取芯芯样分类与整体性评价技术标准

为了实现水泥稳碎石基层整体性质量客观和公正的评价,消除7 d取芯芯样质量评判的随意性、弥补因不考虑气候条件和公路等级等因素对施工质量评价的局限性,选择高速、一级、二级和三级共5条公路,对7 d龄期的水泥稳定粒料类基层进行了大量的取芯试验,系统地对取芯芯样的形态进行了调研、统计和分析,根据芯样的完整程度,将其分为完整类、残缺类和松散类3类,芯样完整性的差异代表其扩散荷载应力能力的不同,根据芯样的致密程度,将芯样进一步分成8级,芯样致密性的不同体现服役功能性差别,提出了评价路段芯样完整率的计算方法,确定了芯样完整率技术标准的确定原则,对于季冻区宜采用F(Ⅰ+Ⅱ)作为评价指标,对于非季冻区宜采用F(Ⅰ)作为评价指标,并根据回归曲线,提出了不同区域不同等级道路7 d龄期内的芯样完整率技术标准。结果表明:芯样完整率与道路等级呈线性关系,道路等级对其影响幅度为2%~9%,养生模式的影响幅度为10%左右,气候因素的影响幅度为5%左右。研究成果实现了对半刚性基层取芯芯样质量和整体性质量的定量化评价。

基于超图和MuSig2聚合签名的联盟链主从多链共识机制

针对多链式区块链采用主链最终共识机制,导致主链负载大,制约从链性能等问题,论文提出一种基于超图和MuSig2聚合签名的联盟链主从多链共识机制.首先根据超图理论,构建以横贯超图为主链,子超图为从链的联盟链主从多链架构;然后借鉴分治思想,结合“背书-排序-验证”的共识方式,构建分层分类共识机制,通过分类处理交易降低主链负载压力;最后构建基于MuSig2聚合签名的联盟链多方背书签名方法,提升背书签名的验证效率.性能分析表明:基于MuSig2聚合签名的联盟链多方背书签名安全可靠,基于超图和MuSig2聚合签名的分层分类共识机制具有强一致性和线性时间复杂度.实验结果表明:基于MuSig2聚合签名的多方背书方法的总效率是椭圆曲线数字签名算法(Elliptic Curve Digital Signature Algorithm,ECDSA)的1.55倍,分层分类共识机制能够提升12.5%的共识效率.该机制具有较高性能,可满足企业多样化业务需求.

基于残差卷积与多头自注意力的CXR图像分类

为了提高新型冠状病毒肺炎(COVID-19)检测的效率和准确性,本文提出一种自动识别COVID-19胸部X射线(CXR)图像的网络模型(MHRA-RCNet)。在ResNet50模型的基础上,首先,采用残差卷积对CXR图像中形状复杂的感染区域进行局部特征提取。其次,选择在ResNet50的第2、3阶段引入多头关系聚合模块,以增强对全局信息的建模能力;为了进一步将局部信息和全局信息进行融合,以提高特征的表达能力和特征之间位置的相关性,在ResNet50的最后阶段引入了空洞视觉Transforme模块,有助于识别CXR图像中复杂的病变区域。最后,将融合后的特征以串联方式输入全局平均池化层进行全局空间信息整合,通过多层感知机进行图像分类并进行可视化分析。在公开访问的COVID-19 Radiography Database数据集与其他深度学习模型进行实验对比。实验结果表明:本文模型在多项分类指标上具有较好的分类精度;另外,从精确度、灵敏度和特异性上也可以直观地看出本文模型能够较好地识别新冠肺炎,进一步证明了本文模型在图像分类任务中的优越性和有效性。

基于用户性格和语义-结构特征的文本评论情感分类方法

由于传统文本评论情感分类方法通常忽略用户性格对于情感分类结果的影响,提出一种基于用户性格和语义-结构特征的文本评论情感分类方法(User Personality and Semantic-structural Features based Sentiment Classification Method for Text Comments,BF_Bi GAC).依据大五人格模型能够有效表达用户性格的优势,通过计算不同维度性格得分,从评论文本中获取用户性格特征.利用双向门控循环单元(Bidirectional Gated Recurrent Unit,Bi GRU)和卷积神经网络(Convolutional Neural Network,CNN)可以有效提取文本上下文语义特征和局部结构特征的优势,提出一种基于Bi GRU、CNN和双层注意力机制的文本语义-结构特征获取方法.为区分不同类型特征的影响,引入混合注意力层实现对用户性格特征和文本语义-结构特征的有效融合,以此获得最终的文本向量表达.在IMDB、Yelp-2、Yelp-5及Ekman四个评论数据集上的对比实验结果表明,BF_Bi GAC在分类准确率(Accuracy)和加权macro F_(1)值(F_(w))上均获得较好表现,相对于拼接Bi GRU、CNN的情感分类方法(Sentiment Classification Method Concatenating Bi GRU and CNN,Bi G-RU_CNN)在Accuracy值上分别提升0.020、0.012、0.017及0.011,相对于拼接CNN、Bi GRU的情感分类方法(Sentiment Classification Method Concatenating CNN and Bi GRU,Conv Bi LSTM)F_(w)值上分别提升0.022、0.013、0.028及0.023;相对于预训练模型BERT和Ro BERTa,BF_Bi GAC在保证分类精度的情况下获得了较高的运行效率.

KENN:线性结构熵的图核神经网络

图神经网络(graph neural network,GNN)是一种利用深度学习直接对图结构数据进行表征的框架,近年来受到人们越来越多的关注.然而传统的基于消息传递聚合的图神经网络(messaging passing GNN,MP-GNN)忽略了不同节点的平滑速度,无差别地聚合了邻居信息,易造成过平滑现象.为此,研究并提出一种线性结构熵的图核神经网络分类方法,即KENN.它首先利用图核方法对节点子图进行结构编码,判断子图之间的同构性,进而利用同构系数来定义不同邻居间的平滑系数.其次基于低复杂度的线性结构熵提取图的结构信息,加深和丰富图数据的结构表达能力.通过将线性结构熵、图核和图神经网络三者进行深度融合提出了图核神经网络分类方法.它不仅可以解决生物分子数据节点特征的稀疏问题,也可以解决社交网络数据以节点度作为特征所产生的信息冗余问题,同时还使得图神经网络能够自适应调整对图结构特征的表征能力,使其超越MP-GNN的上界(WL测试).最后,在7个公开的图分类数据集上实验验证了所提出模型的性能优于其他的基准模型.

面向节点分类任务的节点级自适应图卷积神经网络

图神经网络通过对图中节点的递归采样与聚合以学习节点嵌入,而现有方法中节点采样与聚合的模式较固定,对局部模式的多样性捕获存在不足,从而降低模型性能.因此,文中提出节点级自适应图卷积神经网络(Node-Level Adaptive Graph Convolutional Neural Network,NA-GCN).设计基于节点重要性的采样策略,自适应地确定各节点的邻域规模.同时,提出基于自注意力机制的聚合策略,自适应地融合给定邻域内的节点信息.在多个基准图数据集上的实验表明,NA-GCN在节点分类任务上具有较优性能.

基于融合模型与语义网络的App用户意图识别研究

随着手机应用软件的流行,应用市场上出现了大量非结构化的中文用户评论。基于用户评论识别App用户意图,可以帮助开发人员对App软件进行有针对性的维护和改善。为了从中准确识别用户意图,提出一种基于融合模型和语义网络的App用户意图识别方法FSAUIR。使用百度工具Senta判断评论的情感倾向,构建基于RoBERTa的融合意图分类模型RBMS,通过RoBERTa模型将用户评论转化为语义特征表示,并将其输入到双向门控循环单元中,以提取评论的全局上下文语义信息,同时利用多头自注意力机制和SoftPool获取关键的特征信息,保留主要特征,通过Softmax进行归一化处理,得到意图分类结果。在意图分类的基础上,引入PositionRank模型提取各意图类别下评论的关键词,计算关键词之间的共现关系,构建关键词语义网络,从而更细粒度地识别用户意图。实验结果表明,相比BERT、RoBERTa、RoBERTa-CNN等模型,RBMS模型在人工标注数据集上具有较优的分类性能,准确率、精确率、召回率、F1值分别为87.75%、88.09%、87.80%、87.88%。此外,在意图分类的结果集中,FSAUIR构建的语义网络可以高效地挖掘出用户评论中有价值的信息。

不平衡数据集下基于多粒度近邻图的智能电表故障分类方法

智能电表故障的准确预测对实现计量设备精准主动运维、保障电网稳定运行具有重要意义。电表各故障类型样本的出现频次不同,且不同故障类型样本在高维特征空间中的分布存在重叠,这极大增加了故障预测的难度。现有不平衡分类方法通过构建单一样本信息与其对应类别标签的映射关系来划分样本类型,导致对具有相似表征信息的重叠区样本难以准确判别,降低了整体分类精度。该文提出一种基于多粒度近邻图的智能电表故障分类方法。首先,选择原始数据集中样本作为目标样本,以目标样本及其近邻样本作为节点、目标样本与其近邻样本连线作为边构建近邻图。根据选择的近邻样本数量不同构建多粒度近邻图,实现目标样本的信息扩充和训练样本的数量扩增,更有利于模型稳定训练。构建编码器挖掘近邻图节点特征,利用图注意力机制,根据近邻图节点编码特征和节点邻接关系将近邻样本信息自适应地聚合到目标样本,实现对相似样本差异的有效挖掘。对于给定测试样本,通过集成测试样本多粒度近邻图的分类结果,得到更精准、更鲁棒的智能电表故障预测结果。在20个KEEL(knowledge extraction based on evolutionary learning)和UCI(UC Irvine machine learning repository)不平衡分类公开数据集和智能电表实际故障数据集上的大量实验结果表明,与17种典型方法相比,该文所提算法在处理智能电表故障分类问题上具有显著优势。

CINO双通道结合多头注意力机制藏文情感分类方法

为了解决藏文情感分类任务中现有的模型对文本语义信息理解和深层文本特征提取能力不足的问题,该文使用CINO(Chinese Minority PLM)预训练模型来获取动态词向量,通过TextCNN和BiGRU融合的双通道情感分类模型,分别实现获取文本局部特征和深层全局特征,并引入多头自注意力机制引导模型学习更重要的信息。实验结果表明,该文提出的双通道模型准确率高达92.84%,相较于该文的其他对比模型效果更佳。