基于数据扩维的SAR目标识别性能提升技术

卷积神经网络的出现使得深度学习在视觉领域取得了巨大的成功,并逐渐延伸到合成孔径雷达(SAR)图像识别领域。然而,SAR图像样本量不足,难以支撑卷积神经网络的训练需求,并且SAR图像包含大量相干斑噪声及不确定性,网络结构的设计较为困难。所以,深度学习在SAR图像识别领域的应用受到阻碍。针对上述问题,文中提出一种基于数据扩维的SAR目标识别性能提升方法,通过对原始SAR 图像进行相关预处理操作并把处理后图像与原始图像结合,从而将一维的原始数据扩充成多维数据来作为训练样本。该扩维方法不仅间接扩充了样本量来支撑网络训练,同时也在网络训练前加入了“主动学习冶影响,所以无需针对SAR图像特性来构建复杂卷积网络,而采用成熟、简单的网络进行训练就可以达到理想的测试精度。最后,使用MSTAR 数据对该方法进行了性能验证,实验结果显示了所提方法的有效性。     

改进的全卷积神经网络的脑肿瘤图像分割

针对现有机器学习算法分割脑肿瘤图像精度不高的问题,提出一种基于改进的全卷积神经网络的脑肿瘤图像分割算法。算法首先将FLAIR、T2和T1C三种模态的MR脑肿瘤图像进行灰度归一化,随后利用灰度图像融合技术得到肿瘤信息更加全面的预处理图像;然后采用融合三次脑肿瘤特征信息的改进全卷积神经网络对预处理图像进行粗分割,并且在每个卷积层后加入批量正则化层以加快网络训练的收敛速度,提高训练模型精度;最后融合全连接条件随机场细化粗分割结果中的脑肿瘤边界。实验结果表明,相较于传统的卷积神经网络脑肿瘤图像分割算法,本算法在分割精度和稳定性上有了较大提升,平均Dice可达91.29%,实时性较好,利用训练模型平均1s内可完成单张脑肿瘤图像的分割。      

一种改进的噪声图像语义分割方法

在图像的获取和传输过程中,可能会出现噪声, 它不仅破坏了图像的真实信息,而且严重影响了图像的视觉效果。因此, 噪声图像的语义分割成为图像分析中最具挑战性的问题之一。为了提高噪声图像的分割性能 ,本文在分析全卷积网络(FCN)的 基础上,提出一种改进的FCN模型(IFCN)对噪声图像语义分割。该算法采用一种新的中值 池化方法代替卷积神经网络的最大值 池化,可以在去除噪声的同时保留更多边缘信息。在训练整个深度网络时,通过反向传播算 法以一种直接的端到端,像素到像素 的方式映射。实验结果表明,提出的模型在PASCAL VOC2012数据集上对噪声图像语义分割 可以获得比较好的分割效果,准确率mean IU达到86.5%。     

基于“点到面”策略的图像检索

图像检索是计算机视觉领域的一个重要分支。其主要目的是从图像数据库中找出与查询图像相似的语义图像。传统的图像检索方法是在查询图像和数据库图像之间进行“点到点”检索。但是,单个查询图像包含的类别提示较少,即类别信息较弱,使得检索结果并不理想。为了解决这个问题,本文提出了一种基于“点到面”的类别检索策略来扩展一个图像（点）到一个图像类别（面）,这意味着从单个查询图像到整个图像类别的语义扩展。该方法挖掘了查询图像的类别信息。在两个常用的数据集上对所提出方法的性能进行了评估。实验表明,该方法可以显著提高图像检索的性能。      

基于图像差分与卷积深度置信网络的表情识别

针对传统的人脸表情识别方法中提取表情特征时 没有去除个体性差异及突出表情关键部位的高层次 特征,本文提出一种将图像差分与改进的卷积深度置信网络(CDBN)相结合的表情识别方法 。首先对人 脸表情图片进行裁剪、降维等预处理,之后将各类表情图像与中性表情图像做差分运算提取 各类表情的差 分图像,为了提取表情关键部位的深层次特征,本文将卷积受限玻尔兹曼机(CRBM)的可见层 单元划分为多 个区域,分区进行特征学习,并将此CRBM堆叠起来,形成分区卷积深度置信网络(PCDBN), 之后将各表 情的差分图像作为PCDBN可视层的输入,并利用对比散度算法逐层训练网络,最后在顶层添 加softmax 分类器作为输出层以实现表情识别。在JAFFE和CK+表情库上的实验结果均达到了95%以上的识别率,扩 大训练样本后,在CK+表情库上的识别率可达99%以上。     

Energy-efficient skyline query optimization in wireless sensor networks

With the deployment of wireless sensor networks (WSNs) for environmental monitoring and event surveillance, WSNs can be treated as virtual databases to respond to user queries. It thus becomes more urgent that such databases are able to support complicated queries like skyline queries. Skyline query which is one of popular queries for multi-criteria decision making has received much attention in the past several years. In this paper we study skyline query optimization and maintenance in WSNs. Specifically, we first consider skyline query evaluation on a snapshot dataset, by devising two algorithms for finding skyline points progressively without examining the entire dataset. Two key strategies are adopted: One is to partition the dataset into several disjoint subsets and produce the skyline points in each subset progressively. Another is to employ a global filter that consists of some skyline points in the processed subsets to filter out unlikely skyline points from the rest of unexamined subsets. We then consider the query maintenance issue by proposing an algorithm for incremental maintenance of the skyline in a streaming dataset. A novel maintenance mechanism is proposed, which is able to identify which skyline points from past skylines to be the global filter and determine when the global filter is broadcast. We finally conduct extensive experiments by simulations to evaluate the performance of the proposed algorithms on both synthetic and real sensing datasets, and the experimental results demonstrate that the proposed algorithms significantly outperform existing algorithms in terms of network lifetime prolongation.     

Speckle noise removal based on structural convolutional neural networks with feature fusion for medical image

Application of convolutional neural networks (CNNs) for image additive white Gaussian noise (AWGN) removal has attracted considerable attentions with the rapid development of deep learning in recent years. However, the work of image multiplicative speckle noise removal is rarely done. Moreover, most of the existing speckle noise removal algorithms are based on traditional methods with human priori knowledge, which means that the parameters of the algorithms need to be set manually. Nowadays, deep learning methods show clear advantages on image feature extraction. Multiplicative speckle noise is very common in real life images, especially in medical images. In this paper, a novel neural network structure is proposed to recover noisy images with speckle noise. Our proposed method mainly consists of three subnetworks. One network is rough clean image estimate subnetwork. Another is subnetwork of noise estimation. The last one is an information fusion network based on U-Net and several convolutional layers. Different from the existing speckle denoising model based on the statistics of images, the proposed network model can handle speckle denoising of different noise levels with an end-to-end trainable model. Extensive experimental results on several test datasets clearly demonstrate the superior performance of our proposed network over state-of-the-arts in terms of quantitative metrics and visual quality.     

Efficient Continuous Skyline Query Processing Scheme over Large Dynamic Data Sets

Performing continuous skyline queries of dynamic data sets is now more challenging as the sizes of data sets increase and as they become more volatile due to the increase in dynamic updates. Although previous work proposed support for such queries, their efficiency was restricted to small data sets or uniformly distributed data sets. In a production database with many concurrent queries, the execution of continuous skyline queries impacts query performance due to update requirements to acquire exclusive locks, possibly blocking other query threads. Thus, the computational costs increase. In order to minimize computational requirements, we propose a method based on a multi‐layer grid structure. First, relational data object, elements of an initial data set, are processed to obtain the corresponding multi‐layer grid structure and the skyline influence regions over the data. Then, the dynamic data are processed only when they are identified within the skyline influence regions. Therefore, a large amount of computation can be pruned by adopting the proposed multi‐layer grid structure. Using a variety of datasets, the performance evaluation confirms the efficiency of the proposed method.     

多级跳线连接的深度残差网络超分辨率重建

由于快速的卷积神经网络超分辨率重建算法(FSRCNN)卷积层数少、相邻卷积层的特征信息之间缺乏关联性,因此难以提取到图像深层信息导致图像超分辨率重建效果不佳。针对此问题,该文提出多级跳线连接的深度残差网络超分辨率重建方法。首先,该方法设计了多级跳线连接的残差块,在多级跳线连接的残差块基础上构造了多级跳线连接的深度残差网络,解决相邻卷积层的特性信息缺乏关联性的问题;然后,使用随机梯度下降法(SGD)以可调节的学习率策略对多级跳线连接的深度残差网络进行训练,得到该网络超分辨率重建模型;最后,将低分辨率图像输入到多级跳线连接的深度残差网络超分辨率重建模型中,通过多级跳线连接的残差块得到预测的残差特征值,再将残差图像和低分辨率图像组合在一起转化为高分辨率图像。该文方法与bicubic, A+, SRCNN, FSRCNN和ESPCN算法在Set5和Set14测试集上进行了对比测试,在视觉效果和评价指标数值上该方法都优于其它对比算法。     

采用深度卷积神经网络方法的毫米波图像目标检测

通过收集大量的毫米波图像并建立相应的人体数据集进行检测,提出基于Faster R-CNN深度学习的方法检测隐藏于人体上的危险物品。该方法将区域建议网络和VGG19训练卷积神经网络模型相结合,构建了面向毫米波图像目标检测的深度卷积神经网络。为了提高毫米波图像的处理能力,采用Caffe深度学习框架在图形处理单元上进行训练和测试。实验结果证明了基于Faster R-CNN深度卷积神经网络的目标检测方法能有效检测毫米波图像中的危险物品,并且目标检测的平均准确率约94%,检测速度约为6 frame/s,对毫米波安检系统的智能化发展有着极其重要的参考价值。     

基于深度学习的复杂背景图像分类方法研究

复杂背景图像受背景干扰后不易被识别。针对这一问题,文中提出了基于前景分割机制的卷积神经网络图像分类方法。采用全卷积神经网络对图像前景区域进行自动分割,通过图像中前景区域周围的最小边界框对其进行定位。对于定位的前景区域,构建卷积神经网络对其进行处理以区分不同的类别,从而实现复杂背景图像的分类。将提出方法在公开数据集中提取的单一背景和复杂背景图像数据集上进行对比实验,并使用迁移学习与数据增强等方法优化模型。实验结果表明,所提方法使用前景区域分割相比于仅分类CNN具有更高的准确度,且复杂背景图像上的准确度提升幅度要远大于单一背景图像。该结果说明引入前景区域分割对于复杂背景图像分类模型准确度的提升具有一定帮助,能够显著前景区域特征并减少背景因素的干扰。     

Gravel coverage rate measurement in synchronous chip seal based on deep convolutional neural network

Synchronous chip seal is an advanced road constructing technology, and the gravel coverage rate is an important indicator of the construction quality. In this paper, a novel approach for gravel coverage rate measurement is proposed based on deep learning. Convolutional neural network (CNN) is used to segment the image of ground covered with gravels, and the gravel coverage rate is computed by the percentage of gravel pixels in the segmented image. The gravel coverage rate dataset for model training and testing is built. The performance of fully convolutional neural network (FCN) and U-Net model in the dataset is tested. A better model named GravelNet is constructed based on U-Net. The scaled exponential linear unit (SELU) is employed in the GravelNet to replace the popular combination of rectified linear unit (ReLU) and batch normalization (BN). Data augmentation and alpha dropout are performed to reduce overfitting. The experimental results demonstrate the effectiveness and accuracy of our proposed method. Our trained GravelNet achieves the mean gravel coverage rate error of 0.35% on test dataset.     

Color image retrieval based on hidden Markov models

In this correspondence, a new approach to retrieving images from a color image database is proposed. Each image in the database is represented by a two-dimensional pseudo-hidden Markov model (2-D PHMM), which characterizes the chromatic and spatial information about the image. In addition, a flexible pictorial querying method is used, by which users can paint the rough content of the desired images in a query picture. Image matching is achieved by comparing the query picture with each 2-D PHMM in the database. Experimental results show that the proposed approach is indeed effective.     

基于卷积神经网络迁移学习的飞机目标识别

飞机目标识别是地面情报系统的一项重要关键技术。近年来火热的深度学习方法,如卷积神经网络,展现出对于图像识别任务的优越性能。但是,训练卷积神经网络需要大量的带标签样本以估计规模庞大的模型参数,因而限制了其在雷达目标识别领域中的应用。针对飞机目标识别中的小样本问题,文中引入适用于有限数据场景的迁移学习技术,预先在其他大样本高分辨距离像数据上训练一个初始卷积神经网络模型,再结合当前飞机目标识别任务调优模型参数。在实测数据上的实验结果显示,与仅使用卷积神经网络的方法相比,所提方法可显著提升识别准确率,验证了方法的有效性。     

基于兴趣目标的图像检索

当前图像检索算法通常针对整体图像提取特征以完成检索任务.然而,在很多情况下用户只会关注图像的一部分,即他们的兴趣目标.此时,从整体图像提取的特征一部分是有效的,另一部分则是无效的且会对检索过程带来消极影响.为此,本文提出基于兴趣目标的图像检索方案,并借助于现有的显著性检测、图像分割、特征提取等技术实现一款有效的图像检索算法.首先采用HS （Hierarchical Saliency,分层显著性）检测算法分析用户的兴趣目标并应用SC （Saliency-based Image Cut,基于显著性的图像分割）算法将其分割,然后针对兴趣目标提取HSV （Hue、Saturation、Value,色调、饱和度、明度）颜色特征、SIFT （Scale Invariant Feature Transform,尺度不变特征变换）局部特征和CNN （Convolutional Neural Network,卷积神经网络）语义特征,最后计算其与数据库图像的相似度并根据相似度排序返回检索结果.仿真实验结果表明,本文算法在解决"这是什么东西"这类图像检索任务时明显优于现有的图像检索算法.     

基于3D卷积神经网络的活体人脸检测

非法入侵者通过伪装人脸骗取系统认证,给人脸认证系统带来了严重的威胁。因此,活体人脸检测成了人脸认证系统走向实用必须解决的一个重要课题。现有活体人脸检测方法多为基于照片的人脸攻击方面的研究成果,对于基于视频的人脸攻击,效果并不理想。3D卷积神经网络（Convolutional Neural Network,CNN）具有深度学习的特点,能自动学到图像的分布式特征表示;与2D卷积相比,它能学到连续视频帧的动作信息。本文结合3D卷积神经网络的特性,提出利用3D卷积实现视频人脸伪装检测。通过提取3D卷积神经网络最后全连接层学到的时间空间特征,训练SVM（Support Vector Machine）分类器,实现真实人脸和伪装人脸的分类。实验采用两个人脸伪装公开数据库ReplayAttack和CASIA,实现多尺度内部数据库测试和交叉数据库测试。实验结果相对于纹理特征及2D卷积方法有较大提高,可应用于视频人脸攻击的活体人脸检测。      

Blind quality assessment of tone-mapped images using multi-exposure sequences

The tone mapping operator (TMO) enables high dynamic range (HDR) images to be presented on low dynamic range (LDR) consumer electronic devices. However, the results obtained by this method are not always ideal due to the reduced number of bits. In comparison, the multi-exposure image fusion (MEF) bypasses the intermediate HDR image composition and directly produces an image presented on standard devices. Inspired by this, this paper proposes a quality assessment method for tone-mapped image (TMI) based on generating multi-exposure sequences. Specifically, the method uses a generative adversarial network (GAN) to generate a set of sequences with different exposure levels based on the TMIs. Then a two-branch convolutional neural network (CNN) is used to extract features from the tone-mapped images and the multi-exposure reference sequences, respectively. Finally, the transformer is used to mine the intrinsic connections between TMIs and multi-exposure sequences and learn the mapping relationships from feature space to quality space. We conducted extensive experiments on the ESPL-LIVE HDR database. The applicability and effectiveness of the proposed method are verified by comparing and analyzing relevant features and model configurations with existing mainstream evaluation algorithms.     

基于分块CNN的多尺度SAR图像目标分类算法

针对合成孔径雷达（Synthetic Aperture Radar, SAR）图像目标分辨率差异大,多尺度SAR图像目标分类准确率不高的问题,提出了一种基于迁移学习和分块卷积神经网络（Convolutional Neural Network, CNN）的SAR图像目标分类算法。首先通过大量与目标域相近的源域数据对分块CNN的参数进行训练,得到不同尺度下的CNN特征提取网络;其次将CNN的卷积和池化层迁移到新的网络结构中,实现目标特征的提取;最后用超限学习机（Extreme Learning Machine, ELM）网络对提取的特征进行分类。实验数据采用美国MSTAR数据库以及多尺度SAR图像舰船目标数据集,实验结果表明,该方法对多尺度SAR图像的分类效果优于传统CNN。     

基于点云极化表征与孪生网络的智能车定位

基于3维激光雷达(LiDAR)的智能车定位在地图存储空间与匹配效率、准确率等方面仍存在诸多问题。该文提出一种轻量级点云极化地图构建方法：采用多通道图像模型对3维点云进行编码生成点云极化图,利用孪生网络结构提取并训练点云极化指纹,结合轨迹位姿信息构建点云极化地图。还提出一种基于点云极化地图匹配的智能车定位方法：采用孪生网络对查询指纹与地图指纹进行相似度建模实现快速的地图粗匹配,采用基于2阶隐马尔可夫模型(HMM2)的地图序列精确匹配方法获取最近的地图节点,通过点云配准计算车辆位姿。使用实地数据集和公开的KITTI数据集进行测试。实验结果表明,地图匹配准确率高于96%,定位平均误差约为30 cm,并对不同类型的LiDAR传感器与不同的场景具有较好的鲁棒性。     

Screen content image quality assessment based on convolutional neural networks

Screen content image (SCI) is a composite image including textual and pictorial regions resulting in many difficulties in image quality assessment (IQA). Large SCIs are divided into image patches to increase training samples for CNN training of IQA model, and this brings two problems: (1) local quality of each image patch is not equal to subjective differential mean opinion score (DMOS) of an entire image; (2) importance of different image patches is not same for quality assessment. In this paper, we propose a novel no-reference (NR) IQA model based on the convolutional neural network (CNN) for assessing the perceptual quality of SCIs. Our model conducts two designs solving problems which benefits from two strategies. For the first strategy, to imitate full-reference (FR) CNN-based model behavior, a CNN-based model is designed for both FR and NR IQA, and performance of NR-IQA part improves when the image patch scores predicted by FR-IQA part are adopted as the ground-truth to train NR-IQA part. For the second strategy, image patch qualities of one entire SCI are fused to obtain the SCI quality with an adaptive weighting method taking account the effect of the different image patch contents. Experimental results verify that our model outperforms all test NR IQA methods and most FR IQA methods on the screen content image quality assessment database (SIQAD). On the cross-database evaluation, the proposed method outperforms the existing NR IQA method in terms of at least 2.4 percent in PLCC and 2.8 percent in SRCC, which shows high generalization ability and high effectiveness of our model.