Modulation Recognition in Maritime Multipath Channels: A Blind Equalization-Aided Deep Learning Approach

Modulation recognition has been long investigated in the literature,however,the performance could be severely degraded in multipath fading channels especially for high-order Quadrature Amplitude Modulation(QAM)signals.This could be a critical problem in the broadband maritime wireless communications,where various propagation paths with large differences in the time of arrival are very likely to exist.Specifically,multiple paths may stem from the direct path,the reflection paths from the rough sea surface,and the refraction paths from the atmospheric duct,respectively.To address this issue,we propose a novel blind equalization-aided deep learning(DL)approach to recognize QAM signals in the presence of multipath propagation.The proposed approach consists of two modules:A blind equalization module and a subsequent DL network which employs the structure of ResNet.With predefined searching step-sizes for the blind equalization algorithm,which are designed according to the set of modulation formats of interest,the DL network is trained and tested over various multipath channel parameter settings.It is shown that as compared to the conventional DL approaches without equalization,the proposed method can achieve an improvement in the recognition accuracy up to 30%in severe multipath scenarios,especially in the high SNR regime.Moreover,it efficiently reduces the number of training data that is required.     

基于E-CNN的情绪原因识别方法

文本情绪原因识别作为一个新型的研究方向在文本情绪分析领域占据重要地位。该文结合卷积神经网络,提出了一种基于集成卷积神经网络的情绪原因识别方法。该方法通过词向量、卷积、池化等操作充分融合了句子的语义信息,利用多个CNN集成降低数据不平衡性对情绪原因识别的影响,克服了传统情绪原因识别方法的繁琐规则制定、特征抽取、特征空间降维等过程。实验结果表明,该文的方法在情绪原因识别方面取得了较好的效果,对于情绪归因的方法研究具有一定的指导作用。     

基于高低维度特征融合的双通道卷积神经网络

为了充分利用图像中所隐藏的特征信息,提出将低级维度特征融合在全连接层,构建出融合了高低级维度特征的双通道卷积神经网络。首先构建一个传统的双通道卷积神经网络,在两通道上设置不同大小的卷积核,将双通道的池化层分别连接到全连接层,同时将两通道卷积神经网络的第一池化层提取的特征也直接送到全连接层,使提取得到的初级和高级特征图在全连接层上进行融合,融合后的数据输入到Softmax分类器进行分类。不同算法在fashion-mnist和CIFAR-10数据库上的对比仿真结果表明,本文模型获得了较高的分类准确率。     

结合结构自相似性和卷积网络的单幅图像超分辨率

针对单幅图像超分辨率（SR）复原病态逆问题,在重建过程边缘细节丢失导致的模糊,提出一种结合结构自相似和卷积网络的单幅图像超分辨率算法。首先,通过将尺度分解获得待重构图片样本的自身结构相似性,结合外部数据库样本结合作为训练样本,可以解决样本过于分散的问题;其次,将样本输入卷积神经网络（CNN）进行训练学习,得到单幅图像超分辨率的先验知识;然后,利用非局部约束项自适应选择最优字典重建图像;最后,利用迭代反投影算法对图像超分辨率效果进一步提升。实验结果表明,与双三次插值（Bicubic）方法、K-SVD算法和基于卷积神经网络的图像超分辨率（SRCNN）方法等优秀算法相比,所提算法可以得到边缘更为清晰的超分辨率重建效果。     

一种轻量化油田危险区域入侵检测算法

油田危险区域入侵是油田安防领域的核心问题,以目标检测的方式捕获实时发生的危险是区域入侵任务的重点。为了提高模型的实时性,本文提出结合跨阶段线性瓶颈模块和通道注意力机制的轻量化YOLO检测算法。首先以轻量化卷积模块与跨阶段局部残差模块级联的跨阶段线性瓶颈模块搭建特征提取网络,大大减少了模型的参数量。在特征金字塔的特征融合模块前使用改进的通道注意力机制,增强特征的表达能力与特征的全局的关联性。在特征推理模块,使用中心归一化非极大值抑制方法进行输出优化,避免了对邻近目标的错误抑制。本算法在VOC2007数据集实验,精确率可达74.9%,优于大多轻量化检测算法,已在冀东油田部署应用,有效保证了油田作业人员的生命财产安全。     

基于网络结构搜索的工业过程自动故障诊断方法

针对现有基于深度神经网络的工业过程故障诊断方法存在网络结构设计烦琐及参数寻优耗时等问题,提出了一种基于网络结构搜索的工业过程自动故障诊断方法(automatic fault diagnosis, AutoFD),该方法采用AutoFD网络结构搜索算法,来自动完成卷积神经网络的网络结构设计和网络参数寻优。在此基础上,首先通过在原始数据上施加操作生成新通道;接着利用表现预测加速获取通道适应性排序的过程;然后依据通道适应性排序,通过表现预测来快速选取最优卷积通道数;最终根据最优卷积通道来搜索表现最优的多通道卷积神经网络模型用于工业过程自动故障诊断。采用田纳西—伊斯曼(Tennessee Eastman, TE)工业过程和数值系统对提出方法进行验证,结果表明该方法可以实现网络结构自动设计及网络参数的自动寻优,并且具有优良的故障诊断性能。     

无锚点的遥感图像任意角度密集目标检测方法

针对基于深度学习的遥感图像目标检测方法密集目标漏检率高、分类不准确的问题,建立了一种基于深度学习的无锚点的遥感图像任意角度的密集目标检测方法。首先采用CenterNet作为基线模型,经过主干网络提取特征,并改造原有检测器结构,即加入角度回归分支进行目标角度回归;然后提出一种基于非对称卷积的特征增强模块,并将主干网络提取到的特征图输入特征增强模块,从而增强目标的旋转不变性特征,消除由于目标的旋转、翻转带来的影响,进一步提升目标中心点、尺寸信息的回归精度。采用HourGlass-101作为主干网络时,该方法在DOTA数据集上的平均精度均值（mAP）比旋转区域候选网络（RRPN）提升了7.80个百分点,每秒处理帧数（FPS）提升了7.5;在自建数据集Ship3上,该方法的mAP比RRPN提升了8.68个百分点,FPS提升了6.5。结果表明,所提方法能获得检测精度和速度的平衡。     

基于多尺度网络的运动模糊图像复原算法

动态场景的非均匀盲去模糊一直是图像复原领域中的一个难题。针对目前的模糊图像复原算法不能很好地解决多样性模糊源的问题,提出了一种端到端的基于多尺度网络的运动模糊图像复原算法。所提算法使用修剪过的残差块作为基本单元,且在每一级尺度上都采用相同的非对称编解码网络。为了更好地提取输入图像特征,在编码端使用引入注意力机制的残差模块,还加入了空间金字塔池化层。编码端和解码端中间的循环单元可以获取图像的空间信息,从而利用图像空间的连续性来进行非均匀运动模糊图像的复原。测试结果显示,在GoPro数据集上所提算法的峰值信噪比（PSNR）达到33.69 dB,结构相似性（SSIM）达到0.953 7,且能够更好地复原模糊图像的细节信息,而在Blur数据集上所提算法的PSNR为31.47 dB,SSIM为0.904 7。实验结果表明,与尺度递归网络和深度层次化多patch网络相比,所提算法取得了更优的模糊图像复原效果。     

基于注意力机制的轻量型人体姿态估计

针对高分辨率人体姿态估计网络存在参数量大、运算复杂度高等问题,提出一种基于高分辨率网络（HRNet）的轻量型沙漏坐标注意力网络（SCANet）用于人体姿态估计。首先引入沙漏（Sandglass）模块和坐标注意力（CoordAttention）模块;然后在此基础上构建了沙漏坐标注意力瓶颈（SCAneck）模块和沙漏坐标注意力基础 （SCAblock）模块两种轻量型模块,在降低模型参数量和运算复杂度的同时,获取特征图空间方向的长程依赖和精确位置信息。实验结果显示,在相同图像分辨率和环境配置的情况下,在COCO（Common Objects in COntext）校验集上,SCANet模型与HRNet模型相比参数量降低了52.6%,运算复杂度降低了60.6%;在MPII（Max Planck Institute for Informatics）校验集上,SCANet模型与HRNet模型相比参数量和运算复杂度分别降低了52.6%和61.1%;与常见的人体姿态估计网络如堆叠沙漏网络（Hourglass）、级联金字塔网络（CPN）和SimpleBaseline相比,SCANet模型在拥有更少的参数量与运算复杂度的情况下,仍能实现对人体关键点的高准确度预测。     

Series arc fault diagnosis method of photovoltaic arrays based on GASF and improved DCGAN

In recent years, the methods of machine learning are widely investigated to resolve the series arc fault (SAF) diagnosis problem in photovoltaic (PV) arrays. However, owing to the factors such as weak signal characteristics, long algorithm execution time, and sample imbalance in practical applications, these methods may have difficulties of detecting the SAF. To address these problems, a method based on the Gramian angular summation field (GASF) combined with the squeeze and excitation-deep convolution generative adversarial network (SE-DCGAN) is proposed. Firstly, the absolute difference of margin factor (ADMF) of the current signal is calculated to accurately extract the transient current data when the SAF occurs. Thereafter, the GASF is used to convert transient current data into two-dimensional images to amplify the universal characteristics of the SAF. Subsequently, the SE-DCGAN is adopted to augment the GASF images of the SAF to solve the problem of limited SAF samples. Finally, a convolutional neural network (CNN) is trained to identify the SAF. Also, a fusion sample training method is proposed in this research, that is, normal samples of different PV systems are added to the training set to enhance the generalization ability of CNN. The advantages of the proposed method are that the identification of SAF is improved by converting one-dimensional signals into two-dimensional images, and the generalization ability of the detection model is improved by exploiting the common features of SAFs and fusion training. The validity and generalization ability of the proposed method are verified by three datasets under different PV systems. Experimental results reveal that the proposed method can achieve high recognition accuracy for the measured data; moreover, no misjudgments occurred in identifying the interference events such as maximum power point tracking (MPPT) adjustment and irradiance mutation (IM). In addition, the experiments confirm that the fusion training method enables the model more universal and applicable.