可编程数据平面技术综述

在软件定义网络中,可编程数据平面提供的编程能力是网络功能虚拟化的基石。可编程数据平面技术的核心是可编程能力与数据包处理性能。首先从数据平面的可编程性出发,探讨现有数据平面的数据包处理抽象。然后,分别对数据平面实施的目标平台与对应平台上的主要流表算法进行介绍,详细论述现有数据平面技术。最后,探讨了高性能数据平面技术存在的关键挑战。     

基于双摄像头的心率测量系统的设计与改进

针对在外部干扰或昏暗环境下利用图像光电容积描记(image Photoplethysmography,iPPG)技术进行心率测量时准确度较差,提出了一种自适应心率提取算法,并在嵌入式硬件平台上进行了验证。算法根据图像中人脸与背景区域的色度关系来识别不同的场景并启动合适的摄像头进行图像采集及自适应映射,接着对提取出的信号进行滤波,在信号质量评估后输出结果。上述方法在Zynq平台上进行了验证,使用双摄像头实现实时心率测量,并对结果进行可视化输出。实验结果表明:优化后的算法在光照及运动的双重干扰下的测量误差从3.36BPM降至2.78BPM,准确率提升了17.3%。另外,所设计的系统能够实现在极端黑暗条件下的心率采集,平均误差约为2.39BPM。     

改进Harris-Susan算法下砾石土料掺配均匀度评价

砾石土料掺配均匀程度对砾石土心墙坝的安全至关重要。数字图像处理方法被广泛应用于混合物识别领域,但在处理存在黏连特性的图像时易出现过分割与欠分割问题,导致评价结果精度不足。针对上述问题,本文提出改进Harris-Susan算法下砾石土料掺配均匀度评价方法。首先,改进Harris-Susan算法可分割欠分割区域并修复过分割区域,提高二值图像精度;其次,基于砾石邻域的概念,对图像进行单分位均匀度评价;最后,采用空间分布采样构建砾石土的空间均匀度评价指标。将该方法应用于大渡河双江口砾石土高心墙堆石坝工程建设中,本文方法相对统计准确率为0.935,较Otsu算法（0.353）和分水岭算法（-3.101）提升明显。结果表明,本文方法结果与筛分法高度一致,同时可有效节省人力、物力与时间成本,评价效率更高,具有很强的工程应用价值。     

基于改进DeepLabV3+的铣床碎屑图像分割方法

针对铣床碎屑形状不规则导致图像分割中碎屑轮廓不清晰、分割精度低的问题,本文提出一种改进的DeepLabV3+铣床碎屑分割算法。首先在DeepLabV3+的Xcepetion模块中嵌入通道与空间注意力机制(convolutional block attention module, CBAM)模型,优化通道的权重和位置信息,加强碎屑图像区域的特征学习;其次将DeepLabV3+的空洞空间卷积池化金字塔(atrous spatial pyramid pooling, ASPP)模块改为密集连接(dense conolutional network, DenseNet)方式,增大碎屑图像特征点的感受野,提升铣床碎屑图像特征的复用效率;最后在解码过程中采用多尺度自适应特征融合方法,聚合多尺度特征作为解码器的输入特征,提高碎屑图像分割的精度与鲁棒性。实验结果表明,本文算法优于其他分割算法,改进后算法相比DeepLabV3+,像素准确率提高0.026,平均交并比(mean intersection over union,MIOU)提高0.020,F₁值提高了0.013。     

基于提取双选紧密特征的RGB-D显著性检测网络

针对现有算法对不同来源特征之间的交互选择关注度欠缺以及对跨模态特征提取不充分的问题,提出了一种基于提取双选紧密特征的RGB-D显著性检测网络。首先,为了筛选出能够同时增强RGB图像显著区域和深度图像显著区域的特征,引入双向选择模块(bi-directional selection module, BSM);为了解决跨模态特征提取不充分,导致算法计算冗余且精度低的问题,引入紧密提取模块(dense extraction module, DEM);最后,通过特征聚合模块(feature aggregation module, FAM)对密集特征进行级联融合,并将循环残差优化模块(recurrent residual refinement aggregation module, RAM)配合深度监督实现粗显著图的持续优化,最终得到精确的显著图。在4个广泛使用的数据集上进行的综合实验表明,本文提出的算法在4个关键指标方面优于7种现有方法。     

Intelligent Machine Learning with Metaheuristics Based Sentiment Analysis and Classification

Sentiment Analysis (SA) is one of the subfields in Natural Language Processing (NLP) which focuses on identification and extraction of opinions that exist in the text provided across reviews, social media, blogs, news, and so on. SA has the ability to handle the drastically-increasing unstructured text by transforming them into structured data with the help of NLP and open source tools. The current research work designs a novel Modified Red Deer Algorithm (MRDA) Extreme Learning Machine Sparse Autoencoder (ELMSAE) model for SA and classification. The proposed MRDA-ELMSAE technique initially performs preprocessing to transform the data into a compatible format. Moreover, TF-IDF vectorizer is employed in the extraction of features while ELMSAE model is applied in the classification of sentiments. Furthermore, optimal parameter tuning is done for ELMSAE model using MRDA technique. A wide range of simulation analyses was carried out and results from comparative analysis establish the enhanced efficiency of MRDA-ELMSAE technique against other recent techniques.     

Sentiment Analysis with Tweets Behaviour in Twitter Streaming API

Twitter is a radiant platform with a quick and effective technique to analyze users’ perceptions of activities on social media. Many researchers and industry experts show their attention to Twitter sentiment analysis to recognize the stakeholder group. The sentiment analysis needs an advanced level of approaches including adoption to encompass data sentiment analysis and various machine learning tools. An assessment of sentiment analysis in multiple fields that affect their elevations among the people in real-time by using Naive Bayes and Support Vector Machine (SVM). This paper focused on analysing the distinguished sentiment techniques in tweets behaviour datasets for various spheres such as healthcare, behaviour estimation, etc. In addition, the results in this work explore and validate the statistical machine learning classifiers that provide the accuracy percentages attained in terms of positive, negative and neutral tweets. In this work, we obligated Twitter Application Programming Interface (API) account and programmed in python for sentiment analysis approach for the computational measure of user’s perceptions that extract a massive number of tweets and provide market value to the Twitter account proprietor. To distinguish the results in terms of the performance evaluation, an error analysis investigates the features of various stakeholders comprising social media analytics researchers, Natural Language Processing (NLP) developers, engineering managers and experts involved to have a decision-making approach.     

Horizontal Voting Ensemble Based Predictive Modeling System for ColonCancer

Colon cancer is the third most commonly diagnosed cancer in the world. Most colon AdenoCArcinoma (ACA) arises from pre-existing benign polyps in the mucosa of the bowel. Thus, detecting benign at the earliest helps reduce the mortality rate. In this work, a Predictive Modeling System (PMS) is developed for the classification of colon cancer using the Horizontal Voting Ensemble (HVE) method. Identifying different patterns in microscopic images is essential to an effective classification system. A twelve-layer deep learning architecture has been developed to extract these patterns. The developed HVE algorithm can increase the system’s performance according to the combined models from the last epochs of the proposed architecture. Ten thousand (10000) microscopic images are taken to test the classification performance of the proposed PMS with the HVE method. The microscopic images obtained from the colon tissues are classified into ACA or benign by the proposed PMS. Results prove that the proposed PMS has ~8% performance improvement over the architecture without using the HVE method. The proposed PMS for colon cancer reduces the misclassification rate and attains 99.2% of sensitivity and 99.4% of specificity. The overall accuracy of the proposed PMS is 99.3%, and without using the HVE method, it is only 91.3%.     

基于动态门控特征融合的轻量深度补全算法

稠密深度图在自动驾驶和机器人等领域至关重要,但是现今的深度传感器只能产生稀疏的深度测量,所以有必要对其进行补全.在所有辅助模态中, RGB图像是常用且易得的信息.现今的许多方法都采用RGB和稀疏深度信息结合进行补全.然而它们绝大部分都是利用通道拼接或逐元素求和简单的对两种模态的信息进行融合,没有考虑到不用场景下不同模态特征的置信度.提出一种以输入深度稀疏分布为指导,结合双模态信息量的动态门控融合模块,通过动态产生融合权重的方式对两个模态特征进行更高效的结合.并且根据不同模态的数据特征设计了精简的网络结构.实验结果表明所提出模块和改进的有效性,提出的网络在两个有挑战性的公开数据集KITTI depth completion和NYU depth v2上,使用了很少的参数量达到了先进的结果,取得了性能和速度的优秀平衡.     

视觉Transformer识别任务研究综述

Transformer模型在自然语言处理领域取得了很好的效果,同时因其能够更好地连接视觉和语言,也激发了计算机视觉界的极大兴趣。本文总结了视觉Transformer处理多种识别任务的百余种代表性方法,并对比分析了不同任务内的模型表现,在此基础上总结了每类任务模型的优点、不足以及面临的挑战。根据识别粒度的不同,分别着眼于诸如图像分类、视频分类的基于全局识别的方法,以及目标检测、视觉分割的基于局部识别的方法。考虑到现有方法在3种具体识别任务的广泛流行,总结了在人脸识别、动作识别和姿态估计中的方法。同时,也总结了可用于多种视觉任务或领域无关的通用方法的研究现状。基于Transformer的模型实现了许多端到端的方法,并不断追求准确率与计算成本的平衡。全局识别任务下的Transformer模型对补丁序列切分和标记特征表示进行了探索,局部识别任务下的Transformer模型因能够更好地捕获全局信息而取得了较好的表现。在人脸识别和动作识别方面,注意力机制减少了特征表示的误差,可以处理丰富多样的特征。Transformer可以解决姿态估计中特征错位的问题,有利于改善基于回归的方法性能,还减少了三维估计时深度映射所产生的歧义。大量探索表明视觉Transformer在识别任务中的有效性,并且在特征表示或网络结构等方面的改进有利于提升性能。