基于多层形状特征提取与融合的城市高光谱影像解译

以前的研究往往从像素光谱的角度来解译高光谱影像,忽略了像素间的空间上下文关系。本文提出一种基于像素和对象层形状特征提取与融合的方法,把多层形状特征和光谱信息用支持向量机(SVM)输出函数方法进行融合,用于提取城市高光谱影像的形状特性,利用影像的空间关系。实验用HydICE-DC航空高光谱数据对提出的方法进行了验证,结果表明:像素级形状指数能够提供比对象级形状指数更优的结果,但像素—对象级形状特征的融合,能够给出更高的精度。     

基于混合式特征选择的高分五号影像农田识别

精准农田识别是农作物估产和粮食安全评估的基础。遥感数据作为农田识别的重要数据源,可提供动态、快速的监测结果。高光谱数据在农田识别分类方面具有巨大的应用潜力,但其中的冗余波段影响了分类效率和分类精度。因此,本研究提出了一种适用于高光谱数据农田分类的混合式特征选择算法。首先,基于变量的重要性排序或约束程度,按步长逐步进行降维;其次,寻找分类精度骤减的转折点,并将其对应的变量作为特征子集;最后,利用序列后向选择SBS（Sequential Backward Selection）方法搜索最优分类特征子集。本研究利用GF-5高光谱数据,共研究了3种降维方法（随机森林RF（Random Forest）、互信息MI（Multi-Information）和L1正则化（L1 regularization））和3种分类算法（随机森林、支持向量机SVM（Support Vector Machine）和K近邻KNN（K-Nearest Neighbor））的组合在农田分类中的表现。结果表明,基于L1正则化法得到的特征子集自相关性较低,并且包含的红边和近红外波段有效提高了农田、森林和裸土的区分度。在不同分类模型比较中发现,SVM在高维空间中表现出非常好的抗噪能力,分类精度高于RF和KNN。而RF在低维空间中的泛化能力要高于SVM和KNN。相比于第一步降维得到的特征子集,使用SBS搜索得到的最优特征子集均提高了分类精度。最终,具有23维输入的L1-SVM-SBS分类模型得到了最高的总体分类精度（94.64%）和农田召回率（95.83%）。本研究为高光谱数据特征优选提供了一种新思路,筛选出了更具代表性的特征波段,提高了农田分类精度,对高光谱遥感分类研究具有参考价值。     

基于多特征U-Net网络的工程活动图斑快速识别

为有效监测具有填挖、采剥等行为的工程活动情况,本文提出了一种用于像素级露天工程活动图斑提取的遥感多特征语义分割模型。该模型以高分二号(GF-2)光学遥感影像为数据源,采用U-Net深度神经网络架构,通过人工标注构建了反映露天工程活动的影像样本集,并提取样本的多维特征投入模型进行训练,从而实现了工程活动图斑的快速识别。试验结果显示,本文方法对露天工程活动图斑的总体识别精度可达87.36%,平均精度达86.78%,优于KNN、SVM两种传统分割方法,为工程活动自动化监管提供了技术参考。     

基于Relief F与卷积神经网络的湿地植物群落精细分类

湿地是陆地生态系统和水生生态系统之间的重要过渡带,准确高效地获取湿地植物群落分布信息对于保护湿地具有深远的意义。本文以无人机多光谱影像为数据源,首先构建包含光谱特征、植被指数和纹理特征的多维特征数据集,并采用Relief F算法进行特征优选,确定最优特征数据集;然后构建基于特征优选的卷积神经网络(CNN)分类模型,对最优特征数据集进行分类,并与基于原始多光谱影像的CNN和随机森林(RF)分类方法进行对比。结果表明：(1)随着特征个数的增加,分类精度先增加后下降,当特征数为32时分类精度最高;(2)窗口为13×13的GLCM提取的信息熵和同质性等纹理特征及GNDVI、MSAVI2、RVI等多光谱植被指数重要性较高;(3)基于最优特征数据集的CNN分类模型,能够有效提取空间光谱信息,抑制“椒盐现象”的产生,分类效果最佳,总体精度达93.40%,与未进行特征优选的RF和CNN分类模型相比分别提高了9.80%和7.40%。     

考虑光谱信息和超像素分割的高光谱解混网络

在高光谱解混的过程中考虑影像的空间信息,能够有效提高解混精度。而超像素分割能够划分空间同质区域,为此本文提出一种考虑光谱信息和超像素分割的解混网络（SSUNet）。首先需对原始影像进行超像素分割处理,获得具有空间特征的超像素分割数据,然后采用SSUNet对原始高光谱数据和超像素分割数据进行训练和解混。在线性和非线性混合模型生成的模拟数据集和两个真实数据集上的实验表明,与SUnSAL、SUnSAL-TV、SCLRSU、MTAEU、EGU-Net-pw和1DCNN的解混结果相比,所提网络具有更高的解混精度和较好的鲁棒性。     

联合局部二值模式的CNN高光谱图像分类

卷积神经网络CNN（Convolutional Neural Networks）具有强大的特征提取能力,应用于高光谱图像特征提取取得了良好的效果,双通道CNN模型能够分别提取高光谱图像的光谱特征和空间特征,并实现了特征的决策级融合。局部二值模式LBP（Local Binary Patterns）是一种简单但有效的空间特征描述算子,能够减轻CNN特征提取的压力并提高分类精度。为了充分利用CNN的特征提取能力及LBP特征的判别能力,提出一种双通道CNN和LBP相结合的高光谱图像分类方法,首先,采用1维CNN（1D-CNN）模型处理原始高光谱数据提取深层光谱特征,同时采用另一个1D-CNN模型处理LBP特征数据进一步提取深层空间特征,然后,将两个CNN模型的全连接层进行连接,实现深层光谱特征和空间特征的融合,并将融合特征输入到分类层中完成分类。实验结果表明,该方法在Indian Pines数据、Pavia University数据及Salinas数据上能够分别取得98.54%、99.73%、99.56%的分类精度,甚至在有限数量的训练样本条件下也能取得较好的分类效果。     

谱间对比学习的高光谱图像无监督特征提取EI北大核心CSCD

深度学习通过逐层抽象的方式提取输入数据的深层特征,近年来在高光谱图像分类领域得到了广泛的应用。现有的高光谱图像深度特征提取方法大多属于有监督学习模型,其训练过程需要大量标记样本,而高光谱图像逐像素的标注困难且费时。为此,本文提出了一种基于谱间对比学习的无监督深度学习模型。无须对样本进行标注,仅通过建模不同光谱波段之间的关系便可实现特征提取。具体而言,由于高光谱图像不同的光谱通道刻画了同一物体在不同电磁波段的响应程度,因此必然存在一个特征空间,使得不同通道的光谱信息具有相似的表征。受此启发,本文首先将高维光谱信息分成两组,然后利用多层卷积操作分别提取每组波段的特征,最后对比不同样本所提取的特征,通过对比损失函数来优化模型。为了测试本文方法的性能,将其应用于高光谱图像分类任务中,在Houston 2013、Pavia University和WHU-Hi-Longkou 3个常用的数据集上进行了验证。试验结果表明,在每类仅使用10个训练样本的前提下,本文所提出的无监督学习模型能够获得比主成分分析、自编码器等常见的无监督模型更优越的分类性能。     

基于EMP与混合核SVM的高光谱遥感影像分类

传统的SVM模型采用同一映射形式的单核模式对叠加的空间特征和光谱特征进行处理,往往无法得到理想的结果,为了解决该问题,提出了一种基于扩展的形态学剖面(EMP)与混合核SVM的高光谱遥感影像分类方法.该方法首先通过EMP有效提取空间信息,再采用不同的核函数处理空间信息与光谱信息,最终完成混合核SVM的高光谱影像分类.对多种组合形式的单核以及多核SVM模型进行了对比分析,结果表明,该方法具有较高的适应性,对于高光谱遥感影像的分类精度较高.     

珠海一号高光谱场景分类数据集

高空间分辨率、高光谱分辨率、大幅宽与大数据量是高光谱卫星数据发展趋势,传统高光谱影像的像素级分类面临难以处理海量数据、无法高效获取复杂海量影像中隐含信息的困境。已有研究开始关注高光谱影像的场景级分类,并逐步建立完善高光谱遥感场景分类数据集。然而,目前的数据集制作过程多参考高空间分辨率可见光遥感场景数据集的制作方法,主要采用遥感影像的空间信息进行场景类别解译,忽视了高光谱场景的光谱信息。因此,为构建高光谱影像的遥感场景分类数据集,本文利用“珠海一号”高光谱卫星拍摄的西安地区高光谱数据,使用无监督光谱聚类辅助定位、裁剪与标注待选场景样本,结合Google Earth高分影像进行目视筛选,构建6类场景类型和737幅场景样本的珠海一号高光谱场景分类数据集。并基于光谱与空间两个视角开展场景分类实验,通过视觉词袋、卷积神经网络等方法的基准测试结果,对不同算法在现有多光谱和高光谱遥感场景分类数据集下的性能进行深入分析。本研究可为后续的高光谱影像解译研究提供了有力的数据支撑。     

Approximate representation of thep-norm distribution

［1］Liu D J,Shi W Z,Tong X H,et al.Precision analysis and quality cont rol of GIS spatial data.Shanghai:Shanghai Publishing House of Scientific Documen ts,1999 ［2］Chen X R,Fang Z B,Li G Y,et al.Non_parameter statistics.S hanghai:Shanghai Publishing House of Science and Technology,1989 ［3］Li Q H,Tao B Z.Application of probability statistical theory in survey ing.Beijing:Beijing Publishing House of Surveying and Mapping,1982 ［4］Sun H Y.p_norm distribution theory and its application in surveyin g data processing:[Ph.D Thesis].Wuhan:Wuhan Technical University of Surveying and Mapping,1995     

非球形冰晶的毫米波k-Ze关系研究

针对毫米波雷达处理数据的实际需要,应用离散偶极子近似法DDA,获得了非球形冰晶的后向散射及衰减截面并进行了参数化,并主要基于细化的冰云模型,假设冰晶粒子谱为Γ分布,通过模拟取样各1330次（代表1330种粒子分布）,分别计算得到了W波段（94 GHz）与Ka波段（35 GHz）毫米波雷达探测的冰云衰减系数k及雷达反射率因子Z_e,而且利用数值模拟的方法,建立了k-Z_e关系的具体表达式。计算表明,非球形和非瑞利散射对W波段毫米波雷达衰减的影响较大,而且在同样滴谱分布条件下,W波段毫米波雷达的衰减比Ka波段毫米波雷达的大几倍,此外细化的冰云模型对k-Z_e关系具有影响。本研究对中纬度非降水性冰云的毫米波雷达的衰减订正具有参考价值,并对中国的毫米波雷达应用具有借鉴作用。     

GPS 轨道合成原理服及其实现

The algorithm of using one-day-arc to integrate n-days-arc by orbit overlaying is discussed in detail. An example is given, which proves that the orbit integration method can improve the precision of the orbit efficiently, especially in the determination of a local area＇s orbit.     

Spatially weighted supervised classification for remote sensing

A simple approach for incorporating a spatial weighting into a supervised classifier for remote sensing applications is presented. The classifier modifies the feature-space distance-based metric with a spatial weighting. This is facilitated by the use of a non-parametric (k-nearest neighbour, k-NN) classifier in which the spatial location of each pixel in the training data set is known and available for analysis. A remotely sensed image was simulated using a combined Boolean and geostatistical unconditional simulation approach. This simulated image comprised four wavebands and represented three classes: Managed Grassland, Woodland and Rough Grassland. This image was then used to evaluate the spatially weighted classifier. The latter resulted in modest increase in the accuracy of classification over the original k-NN approach. Two spatial distance metrics were evaluated: the non-centred covariance and a simple inverse distance weighting. The inverse distance weighting resulted in the greatest increase in accuracy in this case.     

Thermal re-emission effects on GPS satellites

Highly precise satellite-derived coordinates depend on accurate orbit predictions, which cannot be achieved with purely empirical models. Global positioning system (GPS) satellites undergo several periodic perturbing forces that have to be modeled and understood. In this scenario, small non-gravitational forces can no longer be neglected when the purpose of the orbital analysis is to obtain accurate results (Vilhena de Moraes 1994). Together with solar radiation pressure, thermal re-emission effects due to solar heating and Earth albedo are the two most important non-gravitational effects. While solar radiation pressure is widely understood, our knowledge about thermal re-emission effects on GPS satellites is in its infancy. Few models have been proposed in recent years and despite the interest of the scientific community, there is a lack of detailed results concerning the magnitude and the behavior of such forces. The aim of this work is to provide a thermal re-emission force model for GPS satellites, simple enough to minimize the problem of modeling a satellite of complex shape with several components on its surface, but accurate enough to provide an estimate of the magnitude and the behavior of these forces, as well as to provide some input to the present knowledge about photon thrust on GPS satellites. Some results of this work point to the fact that thermal re-emission effects are good candidates to partially explain the Y-bias for GPS satellites.     

Formal difference analysis and unification onp-norm distribution density functions

The cause of the formal difference ofp-norm distribution density functions is analyzed, two problems in the deduction ofp-norm formulating are improved, and it is proved that two different forms ofp-norm distribution density functions are equivalent. This work is useful for popularization and application of thep-norm theory to surveying and mapping. Supported by Scientific Research Fund of Human Province Education Department (No. 03C483).     

Integrating a fuzzy <Emphasis Type="Italic">k</Emphasis>-means classification and a Bayesian approach for spatial prediction of landslide hazard

A robust method for spatial prediction of landslide hazard in roaded and roadless areas of forest is described. The method is based on assigning digital terrain attributes into continuous landform classes. The continuous landform classification is achieved by applying a fuzzy k-means approach to a watershed scale area before the classification is extrapolated to a broader region. The extrapolated fuzzy landform classes and datasets of road-related and non road-related landslides are then combined in a geographic information system (GIS) for the exploration of predictive correlations and model development. In particular, a Bayesian probabilistic modeling approach is illustrated using a case study of the Clearwater National Forest (CNF) in central Idaho, which experienced significant and widespread landslide events in recent years. The computed landslide hazard potential is presented on probabilistic maps for roaded and roadless areas. The maps can be used as a decision support tool in forest planning involving the maintenance, obliteration or development of new forest roads in steep mountainous terrain.     

SAINF: A toolbox for analyzing the effect of point-like,line-like and polygon-like infrastructural features on the distribution of point-like non-infrastructural features

This paper introduces a GIS-based toolbox, called SAINF, for analyzing the effects of infrastructural features on the distribution of non-infrastructural features. One of the distinct characteristics of SAINF is that it can deal not only with point-like infrastructural features but also line-like (e.g. railways) and polygon-like (e.g. big parks) features. SAINF analyzes these effects with the conditional nearest neighbor distance method and the cross K-function method. First, the paper briefly refers to these methods. Second, the paper illustrates how to use SAINF with an actual example. SAINF can be downloaded via the Internet without charge for non-profit uses.We express our thanks to Exceed Co. Ltd. for helping us program SAINF, and to Miki Arimoto for making the data of high-class apartment buildings and parks in Kiba. This development was partly supported by Grant-in-aid for Scientific Research No. 10202201 (Spatial information science for human and social sciences) and No. 14350327 (Regional supporting network planning for the elderly) by the Ministry of Education, Culture, Sports, Science and Technology of Japan.     

Estimation of parameters in a split functional model of geodetic observations (<Emphasis Type="Italic">M</Emphasis><Subscript>split</Subscript> estimation)

The paper presents a method of estimating parameters in two competitive functional models. The models considered here are concerned with the same observation set and are based on the assumption that an observation may result from a realization of either of two different random variables. These variables differ from one another at least in the main characteristic (for example, outliers can be realizations of one variable). A quantity that describes the opportunity of identifying a single observation with one random variable is assumed to be known. That quantity, called the elementary split potential, is strictly referred to the amount of information that an observation can provide about two competitive assumptions concerning the observation distribution. Parameter assessments that maximize the global elementary split potential (concerning all observations), are called M _split estimators. A generalization of M _split estimation presented in the paper refers to the theoretical foundation of M-estimation. An erratum to this article can be found at     

油松毛虫灾害遥感监测及其影响因子分析

辽宁西部大面积的油松(Pinus tabulaeformis)人工林长期受到油松毛虫(Dendrolimus tabulaeformis)的危害,通过遥感技术,可以及时、高效、精准地对此大面积灾害进行监测,并获知地形、气象因子对其的影响。本文利用遥感和地理信息系统(GIS)技术,使用TM、ETM+数据,通过近红外与红光波段反射率的比值RVI,对油松的受灾程度进行了有效监测。前人的研究发现:油松毛虫易在干燥、温暖的环境爆发,本文将监测分类结果与地形、气象数据叠加后,分析发现结果亦与油松毛虫的生物学特性相吻合,由此逆向证明了监测结果的可靠性。通过对影像灰度直方图的分析,发现近红外波段对轻度的虫害敏感;红光波段对重度的虫害敏感。对影响因子的分析发现:油松毛虫在阳坡,坡度缓的地区危害更剧烈;在日照时数长、降雨少、积温低的地区,油松的受灾程度更重。此结论为预测虫害爆发的概率提供了依据。本研究表明:在森林灾害的遥感工作中,利用监测对象的生物生态学特性,可以在实地调查数据不足,难以直接对监测结果进行评价的情况下,判断结果的可靠性。利用此方法,一定程度上可以减少调查的工作量,降低外业的难度。