基于LBSN和多图融合的兴趣点推荐EI北大核心CSCD

兴趣点推荐作为推荐领域的一个重要分支一直备受研究者青睐。本文提出一种基于位置的社交网络(LBSN)和多图融合的兴趣点推荐方法GraphPOI。综合分析用户和兴趣点的内在因素和外部表征,首先,对用户-兴趣点的评分矩阵进行学习得到用户和兴趣点的内部潜在向量;其次,根据评分矩阵构造用户-兴趣点交互图,得到兴趣点在用户空间的表征向量以及用户在兴趣点空间的表征向量;然后,对兴趣点按其地理位置进行聚类,得到兴趣点在位置空间的表征向量,结合兴趣点在用户空间的表征向量进而得到兴趣点的外部表征向量;对用户社交图中的信息扩散现象进行建模,捕获用户的朋友关系,得到用户在社交空间的表征向量,结合用户在兴趣点空间的表征向量进而得到用户的外部表征向量;最后,结合用户和兴趣点的内部潜在向量与外部表征向量,得到用户和兴趣点的最终向量表示,并将其输入到多层神经网络模型中进行评分预测。在Yelp数据集上对所提模型进行验证,结果表明本文方法能够有效提升兴趣点推荐的准确性。     

A pan-sharpening method based on NSCT and formulated as compressive sensing problem

In this paper, a pan-sharpening method, using non-subsampled contourlet transform (NSCT) and the theory of compressive sensing (CS), is proposed. The NSCT is used for sparse image representation to perform a multiscale and directional decomposition of source images in order to express their detail and express the sparsity of their high frequency. The CS is used to merge the multispectral (MS) and panchromatic (pan) images from partial random measurements. Two different fusion rules are then applied. The final pan-sharpened image is obtained by inverse NSCT. Experimental results show the efficiency of the proposed method, compared with pan sharpening based on standard NSCT, in terms of visual quality and objective assessment. Moreover, the proposed technique is very effective when the storage and transmission bandwidth are limited.     

移动机器人SLAM位姿估计的改进四元数无迹卡尔曼滤波EI北大核心CSCD

在全自主运动控制的移动机器人系统中,自身位姿的估计和校正对于移动机器人的运动至关重要。卡尔曼滤波是解决移动机器人同步定位与地图构建(SLAM)常用方法。相较于卡尔曼滤波,无迹卡尔曼滤波(UKF)无须对复杂的非线性函数进行雅可比矩阵运算。本文基于无迹卡尔曼滤波,根据先验协方差的平方根选择sigma点,计算协方差以及加权均值。用四元数表示姿态,将四元数矢量转换为旋转空间进行矩阵运算,在此基础上设计了一种位姿估计算法——基于四元数平方根的无迹卡尔曼滤波(QSR-UKF)算法。试验将EKF、QSR-UKF、SR-UKFEKF 3种算法的位姿估计结果进行仿真分析,并通过相关定量指标进行了描述,验证了本文算法的有效性。     

空-谱信息与稀疏表示相结合的高光谱遥感影像分类

针对传统的高光谱遥感影像分类中多依赖光谱信息而忽视空间信息以及提取的特征维数高的问题,提出了一种空-谱信息与稀疏表示相结合的分类算法。首先,利用最小噪声分离对原始影像进行降维,在此基础上,对主成分图上局部影像块内的所有像素进行重组,并用排序的方法得到旋转不变的空-谱特征。然后,对空-谱特征进行监督学习得到字典,并将提取的测试样本的空-谱特征编码到字典中以得到测试样本的稀疏表示。最后,使用支持向量机分类器(SVM)对高光谱影像进行分类。3组高光谱数据试验表明,与传统的分类方法比较,本文方法能有效提高分类精度。     

张量组稀疏表示的高光谱图像去噪算法

提出了一种基于张量组稀疏表示的高光谱遥感影像降噪。高光谱影像数据可视为三阶张量。首先,高光谱图像被划分为小的张量分块,然后,对相似的张量分块进行聚类,并对聚类分组进行稀疏表示。基于高光谱图像的空间非局部自相似性和光谱相关性,将张量组稀疏表示模型分解为一系列无约束低秩张量的近似问题,进而通过张量分解进行求解。对模拟和真实高光谱数据进行试验,验证了该算法的有效性。     

一种基于四元数变换的彩色遥感影像检索方法

遥感影像常存在旋转和缩放等变化关系,这就要求检索时图像特征具有旋转和尺度不变等性质。目前,彩色遥感影像检索在提取这些特征时通常对色彩通道单独进行处理,导致3个通道之间的关系信息丢失,且没有充分利用数据本身所特有的几何特征,影响了检索精度。结合四元数与正交傅里叶-梅林矩的优点,提出了一种基于四元数变换的遥感影像检索方法。针对已有工作中存在的基于四元数傅里叶-梅林矩所提取的纹理特征只具有旋转不变性的问题,利用四元数正交傅里叶-梅林矩构造出具有旋转和尺度不变的纹理特征,并使用四其对图像进行边缘检测,得到边缘色彩图像并提取边缘色彩直方图,综合多种特征进行图像检索。实验结果表明,使用此方法进行遥感影像检索对于图像旋转和尺度变化具有良好的鲁棒性,检索性能明显提高。     

Remote Targets Recognition Based on Adaptive Weighting Feature Dictionaries and Joint Sparse Representations

With the improvement in resolution, more and more useful information is contained in the space of remote sensing images, which makes the processing of remote sensing data become more complex, and it is easy to cause the curse of dimensionality and the poor recognition effect. In this paper, a remote target recognition approach named AJRC is proposed, which uses joint feature dictionary for sparse representation based on different feature information for adaptive weighting. Firstly, the features of the images are extracted to calculate the contribution weight of each eigenvalue in sparse representation, and each eigenvalue contribution weight is calculated in sparse representation. Through the adaptive method, the contribution ability of each feature value in sparse representation is strengthened, and new atoms are formed to construct feature dictionary, which makes the dictionary more discriminative. Then, the common features of each category image and the private features of a single image are extracted from the feature vector, and a joint dictionary is formed to represent the test image sparse and recognize the output of the target. Aiming at the problem that the target visual contrast difference, the low resolution and the rotation of the target with different angles, the experiment is carried out by different feature extraction methods. At the same time, we use the PCA method to reduce the feature dictionary in order to avoid dimensionality. Experiments show that compared with the existing SRC method and JSM method, this method has better recognition rate.     

联合稳健跨域映射和渐进语义基准修正的零样本遥感影像场景分类

零样本影像分类技术旨在通过学习数据集的部分类别（可见类）,获得识别在训练阶段未出现类别（不可见类）的能力。该技术在遥感大数据时代具有重要现实意义。目前,遥感领域的零样本场景分类方法对于映射后的语义空间优化关注很少,导致已有方法的整体分类性能较差。基于这一考虑,本文提出了一种基于稳健跨域映射和渐进语义基准修正的零样本遥感影像场景分类方法。在训练的有监督学习模块,基于可见类的类别语义向量和场景影像样本,实现深度特征提取器学习和视觉空间到语义空间的稳健映射。在训练的无监督学习阶段,基于全体类别的类别语义向量和不可见类遥感影像样本,分别通过协同表示学习和k近邻算法来渐进修正不可见类类别的语义向量,从而缓解可见类语义空间与不可见类语义空间的漂移问题和自编码跨域映射模型映射后不可见类语义空间与协同表示后不可见类语义空间的偏移问题。在测试阶段,基于学习所得的深度特征提取器、自编码跨域映射模型和修正后的不见类语义向量,实现对不可见类遥感影像场景的分类。本文整合多个已有公开的遥感影像场景数据集,组建了一个新的遥感影像场景数据集,在此数据集上进行试验。试验结果表明本文提出的算法在多种不同的可见类与不可见类的划分情况下都明显优于已有公开零样本分类方法。     

Tetrolet域利用时空相关性的云图压缩感知重建

针对卫星云图数据量大,但传输通道和存储空间相对狭小的问题,本文基于Tetrolet变换,利用相邻时次云图的时空相关性,实现了一种高重构质量的卫星云图压缩感知方法。该方法将善于表达图像方向纹理及边缘信息的Tetrolet变换引入压缩感知的稀疏表示环节,从而很好地体现了卫星云图细节丰富、纹理结构复杂的特性;同时,考虑到卫星云图序列间的相关性,将时间相邻的卫星云图组成图像组,以中间时刻云图作为参考图像,计算其与相邻时次云图的差异,通过在参考图片及序列差异图片间合理分配采样率,获取测量数据,在压缩感知框架下,采用带平滑处理的投影Landweber算法,重构出相邻时次的图像组。实验结果表明,Tetrolet变换适用于卫星云图的稀疏表示,而且图像组时空相关性的利用可显著提高重构云图的视觉效果及客观评价指标;在采样率低于0.2时,红外1、水汽和可见光3个通道重构云图的峰值信噪比(PSNR)较传统方法平均提高了7.48 dB,13.51 dB和6.15 dB。由此可见,本文方法可以通过获取少数随机测量值,重构出高质量的卫星云图,不仅为云图数据的低比特率压缩提供了一种可行的解决方案,而且对于其他序列图像的压缩采样具有借鉴意义。     

Wide-band multi-spectral space for color representation

This paper develops a wide-band multi-spectral space for color representation with Aitken PCA algorithm. This novel mathematical space using the broad-band spectra matching method aims at improving the accuracy of color representation as well as reducing costs for processing and storing multi-spectral images. The results show that the space can present our experimental original spectral spaces (i. e. Munsell color matt and DIN-6164 color chips) with high efficiency, and that the spanning space with three eigenvectors can present the original space at more than 98% CSCR, and when 5 eigenvectors are used it can cover almost the whole original spaces.     

Wide—Band Multi—spectral Space for Color Representation

This paper develops a wide-band multi-spectral space for color representation with Aitken PCA al-gorithm. This novel mathematical space using the broad-band spectra matching metod aims at improving the accuracy of color representation as well as reducing costs for processing and storing multi-spectral images.The results show that the space can present our experimental original spectral spaces(i,e.Munsell color matt and DIN-6164 color chips)with high efficiency,and that the spanning space with three eigenvectors can present the original space at more than 98% CSCR,and when 5eigenvectors are used it can cover almost the whole original spaces.     

面向对象的多尺度加权联合稀疏表示的高空间分辨率遥感影像分类

针对高空间分辨率遥感影像中的地物具有多尺度特性,以及各个尺度的对象特征对地物分类精度的影响具有较强的尺度效性,并结合面向对象影像分析方法和多尺度联合稀疏表示方法在高空间分辨率遥感影像分类中的各自优点,提出了一种面向对象的多尺度加权稀疏表示的高空间分辨率遥感影像分类算法。首先,采用多尺度分割算法获得多尺度分割结果并提取对象的多尺度特征;然后,根据影像对象的多尺度分割质量测度计算各尺度的对象权重,构建面向对象的多尺度加权联合稀疏表示模型;最后,采用2个国产GF-2高空间分辨率遥感数据集和1个高光谱-高空间分辨率航空遥感数据集（WashingtonD.C.数据）验证该算法的有效性。试验结果表明,与SVM、像素级稀疏表示、单尺度和多尺度对象级稀疏表示和深度学习等算法相比较,本文算法获得了较高的OA和Kappa分类精度,提高了各个尺度地物的分类精度,有效抑止了地物分类结果中的椒盐噪声现象,同时保持大尺度地物的区域性和小尺度地物的细节信息。     

对偶四元数用于航天线阵遥感影像光束法平差解算

利用对偶四元数可以同时描述位置与姿态的特性,在简化四元数球面线性插值算法的基础上,建立了基于对偶四元数的航天线阵遥感影像的外方位元素模型,且利用该模型设计并实现了光束法平差解算方法。该方法可使用具有约束条件的参数平差法进行迭代求解。利用两个地区的SPOT-5 HRS立体影像进行了对比试验分析,结果表明,提出的基于对偶四元数的光束法平差算法正确可靠,相比于基于欧拉角的平差算法和基于单位四元数的平差算法,有更高的平差计算精度。     

基于解析字典的图像压缩方法

随着高分辨率图像采集技术的发展,采集图像的数据量大幅增加给图像的压缩、存储和传输带来了严重的挑战。近年来,压缩感知理论以能够通过少量观测信号高精度重建原始信号的优势给图像压缩带来了新的思路。目前压缩感知理论的研究主要集中于综合模型,然而综合模型下图像数据的稀疏表示却存在着计算量大、耗时长的问题。提出了基于解析字典的图像压缩模型。在分析综合模型字典学习方法的基础上介绍了解析模型字典学习的流程,然后详细分析了基于解析字典的图像压缩的总流程。以标准测试库自然图像为数据对模型的重构质量和系统耗时进行了分析,结果表明,本文所提模型在采样率相同的情况下,相对其他模型不仅能够缩短系统的耗时,也提高了图像重构的准确性。     

遥感卫星影像K-SVD稀疏表示去噪

常规的去噪方法在去除遥感卫星影像噪声时,往往会造成影像模糊和空间分辨率下降。本文将稀疏表示理论应用于遥感卫星影像去噪,提出了一种改进算法,能够保留低频信息不变,仅对影像的高频信息进行稀疏重建。算法核心是按照是否能够从过完备字典中选择较少原子进行稀疏表示的原则来区分高频信息中的有效信息和噪声。通过3组实验对改进算法与传统去噪方法进行对比检测,实验结果表明,改进算法在去除噪声的同时,能较好地突出影像的边缘和细节信息。     

稀疏差异先验信息支持的高光谱图像稀疏解混算法

基于光谱库的高光谱稀疏解混技术近年来得到了人们的关注,该技术利用光谱库中光谱样本作为端元,将解混问题转化为稀疏表示问题。然而,由于测量环境的差异,待解混图像的实际端元往往与光谱库中相应光谱信号存在差异。本文提出了一种光谱差异稀疏约束的联合稀疏回归解混算法。首先,假设光谱差异具有稀疏特性,建立了光谱库校正模型,使得在解混过程中可对光谱库进行自适应地调整;然后,将光谱库校正模型与联合稀疏回归解混模型结合,建立了考虑光谱差异的稀疏解混模型;最后,基于交替方向乘子法得到了迭代优化解决方案。分别利用仿真和真实高光谱数据进行了试验验证,结果表明,在光谱库不匹配的情形下,本文方法能够有效提高稀疏解混算法的解混性能。     

The spherical horizontal and spherical vertical boundary value problem – vertical deflections and geoidal undulations – the completed Meissl diagram

 In a comparison of the solution of the spherical horizontal and vertical boundary value problems of physical geodesy it is aimed to construct downward continuation operators for vertical deflections (surface gradient of the incremental gravitational potential) and for gravity disturbances (vertical derivative of the incremental gravitational potential) from points on the Earth's topographic surface or of the three-dimensional (3-D) Euclidean space nearby down to the international reference sphere (IRS). First the horizontal and vertical components of the gravity vector, namely spherical vertical deflections and spherical gravity disturbances, are set up. Second, the horizontal and vertical boundary value problem in spherical gravity and geometry space is considered. The incremental gravity vector is represented in terms of vector spherical harmonics. The solution of horizontal spherical boundary problem in terms of the horizontal vector-valued Green function converts vertical deflections given on the IRS to the incremental gravitational potential external in the 3-D Euclidean space. The horizontal Green functions specialized to evaluation and source points on the IRS coincide with the Stokes kernel for vertical deflections. Third, the vertical spherical boundary value problem is solved in terms of the vertical scalar-valued Green function. Fourth, the operators for upward continuation of vertical deflections given on the IRS to vertical deflections in its external 3-D Euclidean space are constructed. Fifth, the operators for upward continuation of incremental gravity given on the IRS to incremental gravity to the external 3-D Euclidean space are generated. Finally, Meissl-type diagrams for upward continuation and regularized downward continuation of horizontal and vertical gravity data, namely vertical deflection and incremental gravity, are produced. Received: 10 May 2000 / Accepted: 26 February 2001     

基于 GVF 和 Snake 模型的高分辨率遥感图像四元数空间道路提取

基于Snake模型的优势及缺陷,提出一种高分辨率多光谱图像道路提取方法。通过建立多光谱图像信息的四元数表达式,解决Snake模型应用于多光谱图像的局限性;通过多光谱图像四元数空间标记分水岭分割,获取道路初始轮廓,解决道路初始边界设置问题;通过梯度矢量流模型迭代求解,解决道路凹陷位置逼近问题,最终实现道路提取。用QuickBird 高分辨率多光谱遥感数据验证方法的可行性,与直接采用Snake 模型进行道路提取结果相比,本文所提出的方法对于高分辨率影像上的高等级道路提取具有较好的精度。     

Taxonomy of space tessellation

When we map an area or create a digital database for it, the first task is often to partition the space into smaller units. There are traditionally two methods of partitioning: vector and raster. A vector partition delineates the boundary of features by polylines while a raster partition subdivides the space into a regular matrix of square or rectangular pixels. These two are complementary methods of subdividing the space either by features or by unconstrained space cells. In the third dimension, they are extended to polyhedra and voxels, respectively. We will argue in this paper that the terms “vector” and “raster” cannot describe all cases of tessellation. With advances in data modelling, variations of the two traditional methods have been developed, such as the representation of a feature by pixels and not by polylines. At present, there is a lack of systematic terminology to describe the various methods of tessellation. In this paper, we will propose a taxonomy for three-dimensional space tessellation. Its essential feature is to distinguish between abstract concepts of tessellation and their encoding methods. We recognise that tessellation of geographic space is carried out in different stages with increasingly precise mathematical meaning. This provides us with an insight into the process of spatial tessellation and a model to systematically describe the various structures. These concepts could form a basis for spatial data models.     

对偶四元数法在稳健点云配准中的应用

针对点云特征点含有粗差的问题,提出了基于对偶四元数的稳健点云配准方法。遵循加权最小二乘原则构建了基于对偶四元数的稳健点云配准模型,运用拉格朗日乘数法推导了旋转矩阵和平移向量求解公式,研究了新权计算与坐标转换迭代步骤。实例证明,该方法能够有效识别并剔除粗差,提高点云配准精度。