Feature identification in complex fluid flows by convolutional neural networks

Recent advancements have established machine learning's utility in predicting nonlinear fluid dynamics,with predictive accuracy being a central motivation for employing neural networks.However,the pattern recognition central to the networks function is equally valuable for enhancing our dynamical insight into the complex fluid dynamics.In this paper,a single-layer convolutional neural network(CNN)was trained to recognize three qualitatively different subsonic buffet flows(periodic,quasi-periodic and chaotic)over a high-incidence airfoil,and a near-perfect accuracy was obtained with only a small training dataset.The convolutional kernels and corresponding feature maps,developed by the model with no temporal information provided,identified large-scale coherent structures in agreement with those known to be associated with buffet flows.Sensitivity to hyperparameters including network architecture and convolutional kernel size was also explored.The coherent structures identified by these models enhance our dynamical understanding of subsonic buffet over high-incidence airfoils over a wide range of Reynolds numbers.

Classification of Arrhythmia Based on Convolutional Neural Networks and Encoder-Decoder Model

As a common and high-risk type of disease,heart disease seriously threatens people’s health.At the same time,in the era of the Internet of Thing(IoT),smart medical device has strong practical significance for medical workers and patients because of its ability to assist in the diagnosis of diseases.Therefore,the research of real-time diagnosis and classification algorithms for arrhythmia can help to improve the diagnostic efficiency of diseases.In this paper,we design an automatic arrhythmia classification algorithm model based on Convolutional Neural Network(CNN)and Encoder-Decoder model.The model uses Long Short-Term Memory(LSTM)to consider the influence of time series features on classification results.Simultaneously,it is trained and tested by the MIT-BIH arrhythmia database.Besides,Generative Adversarial Networks(GAN)is adopted as a method of data equalization for solving data imbalance problem.The simulation results show that for the inter-patient arrhythmia classification,the hybrid model combining CNN and Encoder-Decoder model has the best classification accuracy,of which the accuracy can reach 94.05%.Especially,it has a better advantage for the classification effect of supraventricular ectopic beats(class S)and fusion beats(class F).

Recurrent Convolutional Neural Network MSER-Based Approach for Payable Document Processing

A tremendous amount of vendor invoices is generated in the corporate sector.To automate the manual data entry in payable documents,highly accurate Optical Character Recognition(OCR)is required.This paper proposes an end-to-end OCR system that does both localization and recognition and serves as a single unit to automate payable document processing such as cheques and cash disbursement.For text localization,the maximally stable extremal region is used,which extracts a word or digit chunk from an invoice.This chunk is later passed to the deep learning model,which performs text recognition.The deep learning model utilizes both convolution neural networks and long short-term memory(LSTM).The convolution layer is used for extracting features,which are fed to the LSTM.The model integrates feature extraction,modeling sequence,and transcription into a unified network.It handles the sequences of unconstrained lengths,independent of the character segmentation or horizontal scale normalization.Furthermore,it applies to both the lexicon-free and lexicon-based text recognition,and finally,it produces a comparatively smaller model,which can be implemented in practical applications.The overall superior performance in the experimental evaluation demonstrates the usefulness of the proposed model.The model is thus generic and can be used for other similar recognition scenarios.

Leucogranite mapping via convolutional recurrent neural networks and geochemical survey data in the Himalayan orogen

Geochemical survey data analysis is recognized as an implemented and feasible way for lithological mapping to assist mineral exploration.With respect to available approaches,recent methodological advances have focused on deep learning algorithms which provide access to learn and extract information directly from geochemical survey data through multi-level networks and outputting end-to-end classification.Accordingly,this study developed a lithological mapping framework with the joint application of a convolutional neural network(CNN)and a long short-term memory(LSTM).The CNN-LSTM model is dominant in correlation extraction from CNN layers and coupling interaction learning from LSTM layers.This hybrid approach was demonstrated by mapping leucogranites in the Himalayan orogen based on stream sediment geochemical survey data,where the targeted leucogranite was expected to be potential resources of rare metals such as Li,Be,and W mineralization.Three comparative case studies were carried out from both visual and quantitative perspectives to illustrate the superiority of the proposed model.A guided spatial distribution map of leucogranites in the Himalayan orogen,divided into high-,moderate-,and low-potential areas,was delineated by the success rate curve,which further improves the efficiency for identifying unmapped leucogranites through geological mapping.In light of these results,this study provides an alternative solution for lithologic mapping using geochemical survey data at a regional scale and reduces the risk for decision making associated with mineral exploration.

Dynamic Hand Gesture Recognition Based on Short-Term Sampling Neural Networks

Hand gestures are a natural way for human-robot interaction.Vision based dynamic hand gesture recognition has become a hot research topic due to its various applications.This paper presents a novel deep learning network for hand gesture recognition.The network integrates several well-proved modules together to learn both short-term and long-term features from video inputs and meanwhile avoid intensive computation.To learn short-term features,each video input is segmented into a fixed number of frame groups.A frame is randomly selected from each group and represented as an RGB image as well as an optical flow snapshot.These two entities are fused and fed into a convolutional neural network(Conv Net)for feature extraction.The Conv Nets for all groups share parameters.To learn longterm features,outputs from all Conv Nets are fed into a long short-term memory(LSTM)network,by which a final classification result is predicted.The new model has been tested with two popular hand gesture datasets,namely the Jester dataset and Nvidia dataset.Comparing with other models,our model produced very competitive results.The robustness of the new model has also been proved with an augmented dataset with enhanced diversity of hand gestures.

Convolutional neural network adaptation and optimization method in SIMT computing mode

For studying and optimizing the performance of general-purpose computing on graphics processing units(GPGPU)based on single instruction multiple threads(SIMT)processor about the neural network application,this work contributes a self-developed SIMT processor named Pomelo and correlated assembly program.The parallel mechanism of SIMT computing mode and self-developed Pomelo processor is briefly introduced.A common convolutional neural network(CNN)is built to verify the compatibility and functionality of the Pomelo processor.CNN computing flow with task level and hardware level optimization is adopted on the Pomelo processor.A specific algorithm for organizing a Z-shaped memory structure is developed,which addresses reducing memory access in mass data computing tasks.Performing the above-combined adaptation and optimization strategy,the experimental result demonstrates that reducing memory access in SIMT computing mode plays a crucial role in improving performance.A 6.52 times performance is achieved on the 4 processing elements case.

基于Convolutional-LSTM的蛋白质亚细胞定位研究

蛋白质亚细胞位置预测研究是目前蛋白质组学和生物信息学研究的重点问题之一。蛋白质的亚细胞定位决定了它的生物学功能,故研究亚细胞定位对了解蛋白质功能非常重要。由于蛋白质结构的序列性,考虑使用序列模型来进行亚细胞定位研究。尝试使用卷积神经网络(convolutional neural network,CNN)、长短期记忆神经网络(long short-term memory,LSTM)两种模型挖掘氨基酸序列所包含的信息,从而进行亚细胞定位的预测。随后构建了基于卷积的长短期记忆网络(Convolutional-LSTM)的集成模型进行亚细胞定位。首先通过卷积神经网络对蛋白质数据进行特征抽取,随后进行特征组合,并将其送入长短期记忆神经网络进行特征表征学习,得到亚细胞定位结果。使用该模型能达到0.816 5的分类准确率,比传统方法有明显提升。

Deep-fake video detection approaches using convolutional–recurrent neural networks

Deep-Fake is an emerging technology used in synthetic media which manipulates individuals in existing images and videos with someone else’s likeness.This paper presents the comparative study of different deep neural networks employed for Deep-Fake video detection.In the model,the features from the training data are extracted with the intended Convolution Neural Network model to form feature vectors which are further analysed using a dense layer,a Long Short-Term Memoryand Gated Recurrent by adopting transfer learning with fine tuning for training the models.The model is evaluated to detect Artificial Intelligence based Deep fakes images and videos using benchmark datasets.Comparative analysis shows that the detections are majorly biased towards domain of the dataset but there is a noteworthy improvement in the model performance parameters by using Transfer Learning whereas Convolutional-Recurrent Neural Network has benefits in sequence detection.

Audiovisual speech recognition based on a deep convolutional neural network

Audiovisual speech recognition is an emerging research topic.Lipreading is the recognition of what someone is saying using visual information,primarily lip movements.In this study,we created a custom dataset for Indian English linguistics and categorized it into three main categories:(1)audio recognition,(2)visual feature extraction,and(3)combined audio and visual recognition.Audio features were extracted using the mel-frequency cepstral coefficient,and classification was performed using a one-dimension convolutional neural network.Visual feature extraction uses Dlib and then classifies visual speech using a long short-term memory type of recurrent neural networks.Finally,integration was performed using a deep convolutional network.The audio speech of Indian English was successfully recognized with accuracies of 93.67%and 91.53%,respectively,using testing data from 200 epochs.The training accuracy for visual speech recognition using the Indian English dataset was 77.48%and the test accuracy was 76.19%using 60 epochs.After integration,the accuracies of audiovisual speech recognition using the Indian English dataset for training and testing were 94.67%and 91.75%,respectively.

Dynamic Resource Allocation in LTE Radio Access Network Using Machine Learning Techniques

Current LTE networks are experiencing significant growth in the number of users worldwide. The use of data services for online browsing, e-learning, online meetings and initiatives such as smart cities means that subscribers stay connected for long periods, thereby saturating a number of signalling resources. One of such resources is the Radio Resource Connected (RRC) parameter, which is allocated to eNodeBs with the aim of limiting the number of connected simultaneously in the network. The fixed allocation of this parameter means that, depending on the traffic at different times of the day and the geographical position, some eNodeBs are saturated with RRC resources (overused) while others have unused RRC resources. However, as these resources are limited, there is the problem of their underutilization (non-optimal utilization of resources at the eNodeB level) due to static allocation (manual configuration of resources). The objective of this paper is to design an efficient machine learning model that will take as input some key performance indices (KPIs) like traffic data, RRC, simultaneous users, etc., for each eNodeB per hour and per day and accurately predict the number of needed RRC resources that will be dynamically allocated to them in order to avoid traffic and financial losses to the mobile network operator. To reach this target, three machine learning algorithms have been studied namely: linear regression, convolutional neural networks and long short-term memory (LSTM) to train three models and evaluate them. The model trained with the LSTM algorithm gave the best performance with 97% accuracy and was therefore implemented in the proposed solution for RRC resource allocation. An interconnection architecture is also proposed to embed the proposed solution into the Operation and maintenance network of a mobile network operator. In this way, the proposed solution can contribute to developing and expanding the concept of Self Organizing Network (SON) used in 4G and 5G networks.

结合太阳辐射量计算与CNN-LSTM组合的光伏功率预测方法研究

为了提高模型预测性能,提出一种综合太阳辐射模型及深度学习的光伏功率预测模型。首先,利用太阳辐射机理建立太阳辐射模型(SRM),估算出水平面上总辐射值,再由斜面辐照度转换方法计算出光伏组件所接收的斜面辐射值。其次,通过皮尔逊相关分析法筛选出对光伏功率影响较大的主要因素,将斜面辐射计算值及主要影响因素作为输入,采用卷积神经网络(CNN)和长短期记忆网络(LSTM)建立光伏功率SRM-CNN-LSTM预测模型。分别利用春夏秋冬四季典型日的数据开展对比实验,结果表明:与几种其他方法相比,该文方法具有更好的预测效果。

基于字词向量融合的民航智慧监管短文本分类

为解决民航监管事项所产生的检查记录仅依靠人工进行分类分析导致效率低的问题,提出一种基于数据增强与字词向量融合的双通道特征提取的短文本分类模型,探讨民航监管事项的分类,包括与人、设备设施环境、制度程序和机构职责等相关问题。为解决类别不平衡问题,采用数据增强算法在原始文本上进行变换,生成新的样本,使各个类别的样本数量更加均衡。将字向量和词向量按字融合拼接,得到具有词特征信息的字向量。将字词融合的向量分别送入到文本卷积神经网络(TextCNN)和双向长短期记忆(BiLSTM)模型中进行不同维度的特征提取,从局部的角度和全局的角度分别提取特征,并在民航监管事项检查记录数据集上进行试验。结果表明:该模型准确率为0.9837,F 1值为0.9836。与一些字嵌入模型和词嵌入模型相对比,准确率提升0.4%。和一些常用的单通道模型相比,准确率提升3%,验证了双通道模型提取的特征具有全面性和有效性。

基于注意力机制与LSTM-CCN的月降水量预测

针对现有月降水量预测方法预测准确性不高的问题,提出一种基于注意力机制与LSTM-CCN的月降水量预测方法。首先,利用长短时记忆神经网络(long short-term memory neural network,LSTM)提取气象数据在时间维度的特征分布,从时间相关性方面捕获相邻时间段或长距离气象数据段中的统计分布;其次,利用因果卷积神经网络(causal convolutional network,CCN)将气象数据映射到空间维度,深层次地从空间维度捕获气象数据在空间中的特征统计分布;再次,以并联的方式将时间和空间特征作为交叉注意力网络的输入,构造融合的时空特征;最后,以长短时记忆神经网络构造解码器,并将融合的时空特征作为解码器的输入,预测的月降水量作为输出。选取河南省新乡市2001~2017年数据集进行测试,结果表明:所提出方法的均方根误差仅为13.08 mm,相比主流方法具有更低的预测误差。研究成果可为提高气象预测的准确性和实用性提供参考。

基于注意力机制的CNN-BiLSTM的IGBT剩余使用寿命预测

针对绝缘栅双极型晶体管(IGBT)可靠性问题,提出了一种融合卷积神经网络(CNN)、双向长短期记忆(BiLSTM)网络和注意力机制的剩余使用寿命(RUL)预测模型,可用于IGBT的寿命预测。模型中使用CNN提取特征参数,BiLSTM提取时序信息,注意力机制加权处理特征参数。使用IGBT加速老化数据集对提出的模型进行验证。结果表明,对比自回归差分移动平均(ARIMA)、长短期记忆(LSTM)、多层LSTM(Multi-LSTM)、 BiLSTM预测模型,在均方根误差和决定系数等评价指标方面该模型的性能最优。验证了提出的寿命预测模型对IGBT失效预测是有效的。

基于CNN-LSTM的机床滚动轴承性能退化趋势和寿命预测

滚动轴承作为机床主轴的关键部件,其剩余寿命预测直接决定着整机设备的剩余寿命。若不能及时地预知滚动轴承的健康状态或损伤情况,不仅会影响维修策略的制定,还会造成级联故障,易造成机床灾难性的事故。针对大数据下滚动轴承振动信号的自适应故障特征提取和智能诊断问题,构建卷积神经网络和长短期记忆网络(CNN-LSTM)相结合的寿命预测模型,它可以避免人工参与的影响,实现网络的互补优势。对滚动轴承的退化状态以及剩余寿命进行预测,并与卷积神经网络(CNN)、长短时记忆神经网络(LSTM)进行对比实验。结果表明:所提方法CNN-LSTM有着较高的预测准确度。

基于多源信息融合和WOA-CNN-LSTM的外脚手架隐患分类预警研究

面对施工现场外脚手架隐患信息的多样性,传统的基于传感器监测的单一信号预警研究存在容错力不佳、含有信息有限等问题。针对施工现场外脚手架“图像+监测”数据,提出一种基于数据层和特征层信息融合的脚手架隐患分类预警方法。首先,利用Revit三维建模软件建立外脚手架实体模型,对不同初始隐患下的外脚手架进行有限元分析,划分隐患预警等级;其次,利用无迹卡尔曼滤波算法(Unscented Kalman Filter,UKF)及卷积长短时记忆网络(Convolutional Neural Network-Long Short Term Memory Network,CNN-LSTM)实现脚手架同类信息数据层融合及异类信息特征层融合;最后,通过实时收集西安市某在建项目落地式双排扣件式钢管脚手架隐患信息,对其进行分类预警,并使用鲸鱼优化算法(Whale Optimization Algorithm,WOA)对CNN-LSTM网络进行参数优化,发现隐藏节点个数为30、学习率为0.0072、正则化系数为1×10^(-4)时分类效果最佳,优化后预警精度达到了91.4526%。通过可视化WOA-CNN-LSTM、CNN-LSTM、CNN-SVM(Support Vector Machine,支持向量机)及CNN-GRU(Gate Recurrent Unit,门控循环单元)分类预警结果,证实了优化后的CNN-LSTM网络在脚手架分类预警方面的优越性。

基于集群辨识和卷积神经网络-双向长短期记忆-时序模式注意力机制的区域级短期负荷预测

为了解决区域级短期电力负荷预测时输入特征过多和负荷时序性较强的问题,提出一种基于集群辨识和卷积神经网络(convolutional neural networks,CNN)-双向长短期记忆网络(bi-directional long short-term memory,BiLSTM)-时序模式注意力机制(temporal pattern attention,TPA)的预测方法。首先,将用电模式和天气作为影响因素,基于二阶聚类算法对区域内的负荷节点进行集群辨识,再从每个集群中挑选代表特征作为深度学习模型的输入,这样既能减少输入特征维度,降低计算复杂度,又能综合考虑预测区域的整体特征,提升预测精度。然后,针对区域电力负荷时序性的特点,用CNN-BiLSTM-TPA模型完成训练和预测,该模型能提取输入数据的双向信息生成隐状态矩阵,并对隐状态矩阵的重要特征加权,从多时间步上捕获双向时序信息用于预测。最后,在美国加利福尼亚州实例上分析验证了所提方法的有效性。

基于链接关系预测的弯曲密集型商品文本检测

针对商品包装文本检测任务中弯曲密集型文本导致的错检、漏检问题,提出了一种由2个子网络组成的基于链接关系预测的文本检测框架(text detection network based on relational prediction,RPTNet)。在文本组件检测网络中,下采样采用卷积神经网络和自注意力并行的双分支结构提取局部和全局特征,并加入空洞特征增强模块(DFM)减少深层特征图在降维过程中信息的丢失;上采样采用特征金字塔与多级注意力融合模块(MAFM)相结合的方式进行多级特征融合以增强文本特征间的潜在联系,通过文本检测器从上采样输出的特征图中检测文本组件;在链接关系预测网络中,采用基于图卷积网络的关系推理框架预测文本组件间的深层相似度,采用双向长短时记忆网络将文本组件聚合为文本实例。为验证RRNet的检测性能,构建了一个由商品包装图片组成的文本检测数据集(text detection dataset composed of commodity packaging,CPTD1500)。实验结果表明:RPTNet不仅在公开文本数据集CTW-1500和Total-Text上取得了优异的性能,而且在CPTD1500数据集上的召回率和F值分别达到了85.4%和87.5%,均优于当前主流算法。

基于CF-CNN-LSTM模型的滑坡易发性评价

针对滑坡灾害样本选择以及深度学习模型中的长期依赖、梯度消失、退化等问题,提出了一种结合确定性系数法(certainty factor,CF)、卷积神经网络(convolution neural network,CNN)模型和长短期记忆神经网络(long short-term memory,LSTM)模型的CF-CNN-LSTM深度学习模型。以广西壮族自治区梧州市辖区为研究区,选取高程、坡度和坡向等13种滑坡评价因子,采用CF-CNN-LSTM模型对研究区进行滑坡易发性评价,并与CNN模型、LSTM模型、循环神经网络模型和逻辑回归模型进行对比,利用受试者工作特征曲线(receiver operating characteristic,ROC)、整体准确率等6种方法对模型预测精度进行评估。结果表明:CF-CNN-LSTM模型的ROC曲线的曲线下面积(area under curve,AUC)值为0.953,高于其它单一模型,同时其它5项评估指标均优于单一模型,证明CF-CNN-LSTM模型具有更高的精度,可用于梧州市辖区的滑坡易发性评价工作,能够对该区域的滑坡风险管理提供科学的建议。

基于ConvLSTM-CNN预测太平洋长鳍金枪鱼时空分布趋势

海洋渔场的变动由空间与环境因子共同驱动,渔场时空演变信息的精准预测是海洋捕捞的关键。本研究利用1995-2018年太平洋海域长鳍金枪鱼(Thunnus alalunga)的渔业生产统计数据,结合同期海洋环境数据包括海表面温度(Sea Surface Temperature,SST)、海表面盐度(Sea Surface Salinity,SSS)、初级生产力(Primary Productivity,PP)和溶解氧浓度(Dissolved Oxygen Concentration,DO),提出了一种融合卷积长短期记忆网络(Convolutional Long Short-Term Memory Networks,ConvLSTM)和卷积神经网络(Convolutional Neural Networks,CNN)的渔场时空分布预测模型。该模型引入特征提取模块,对时空因子进行编码,提取时空特征信息,同时采用CNN提取海洋环境变量的抽象特征,采用ConvLSTM提取渔业数据的高层时空关联信息,最后融合多种特征对渔场时空演变趋势进行预测。结果表明,模型的均方根误差为0.1036,较随机森林、BP神经网络和长短期记忆网络(Long Short Term Memory,LSTM)等传统渔场预报模型的预测误差降低15%~40%,预测的高产渔区与实际作业的高渔获量区匹配度为89%。该研究构建的渔场时空预测模型能够准确地预测出太平洋长鳍金枪鱼的时空分布,为太平洋长鳍金枪鱼的延绳钓渔业提供科学参考依据。