山脊线和山谷线自动提取的一种新方法

针对当前山脊线和山谷线的提取都是从其几何特征或物理特征的单一方面进行研究和设计,不能够利用隐含在数字化资料中的山脊线和山谷线的有效信息的弊端;本文利用几何分析方法Douglas-Peuker提取山脊线和山谷线的特征点,并结合提取区域内地形概略的分水线和汇水线,用概略的分水线和汇水线对特征点进行识别、归类、顺序提取各条山脊线和山谷线,把山脊线的几何特性与物理特性结合起来,克服了各自的弊端。实践表明提取的山脊线和山谷线与实际基本符合。     

一种微地形探测回波信号处理的新算法

针对钴结壳具体赋存环境和开采技术要求,作者提出了利用Morlet小波信号包络和高斯函数曲线及能量相关性算法对拟设计的微地形探测方法接收到的超声回波信号进行分析与计算。由实验初步验证,利用该算法计算出来的被测对象位置精度及误差稳定性均高于传统的超声波单波束多探头探测方法。     

一种新的全球海量地形实时绘制算法

本文为实现全球海量地形数据的实时可视化,提出了一种新算法。算法不使用几何数据而是利用球面特征进行地形多分辨率模型初建,然后基于视锥与节点关系对初建结果进行扩展来得到完整的地形网格。此外设计了能消除具有复杂邻接关系的节点间裂缝的拼接方式,构造了简洁的方法消除GPU32位浮点精度导致的"wob-bling"现象。实现的算法在普通微机上平均漫游速度达每秒95帧以上。     

Integration of remote sensing,county-level census,and machine learning for century-long regional cropland distribution data reconstruction

The Lower Mississippi Alluvial Valley (LMAV) was home to about ten million hectare bottomland hardwood (BLH) forests in the Southern U.S. It experienced over 80 % area loss of the BLH forests in the past centuries and large-scale afforestation in recent decades. Due to the lack of a high-resolution cropland dataset, impacts of land use change (LUC) on the LMAV ecosystem services have not been fully understood. In this study, we developed a novel framework by integrating the machine learning algorithm, county-level agricultural census, and satellite-based cropland products to reconstruct the LMAV cropland distribution during 1850–2018 at a 30-m resolution. Results showed that the LMAV cropland area increased from 0.78 × 10⁴ km² in 1850 to 6.64 × 10⁴ km² in 1980 and then decreased to 6.16 × 10⁴ km² in 2018. Cropland expansion rate was the largest in the 1960s (749 km² yr⁻¹) but decreased rapidly thereafter, whereas cropland abandonment rate increased substantially in recent decades with the largest rate of 514 km² yr⁻¹ in the 2010s. Our dataset has three notable features: (1) the depiction of fine spatial details, (2) the integration of the county-level census, and (3) the inclusion of a machine-learning algorithm trained by satellite-based land cover product. Most importantly, our dataset well captured the continuous increasing trend in cropland area from 1930–1960, which was misrepresented by other cropland datasets reconstructed from the state-level census. Our dataset would be important to accurately evaluate the impacts of historical deforestation and recent afforestation efforts on regional ecosystem services, attribute the observed hydrological changes to anthropogenic and natural driving factors, and investigate how the socioeconomic factors control regional LUC pattern. Our framework and dataset are crucial to developing managerial and policy strategies for conserving natural resources and enhancing ecosystem services in the LMAV.     

Context-based automatic reconstruction and texturing of 3D urban terrain for quick-response tasks

Highly detailed 3D urban terrain models are the base for quick response tasks with indispensable human participation, e.g., disaster management. Thus, it is important to automate and accelerate the process of urban terrain modeling from sensor data such that the resulting 3D model is semantic, compact, recognizable, and easily usable for training and simulation purposes. To provide essential geometric attributes, buildings and trees must be identified among elevated objects in digital surface models. After building ground-plan estimation and roof details analysis, images from oblique airborne imagery are used to cover building faces with up-to-date texture thus achieving a better recognizability of the model. The three steps of the texturing procedure are sensor pose estimation, assessment of polygons projected into the images, and texture synthesis. Free geographic data, providing additional information about streets, forest areas, and other topographic object types, suppress false alarms and enrich the reconstruction results.     

Embedding scale: new thinking of scale in machine learning and geographic representation

Journal of Geographical Systems - Concepts of scale are at the heart of diverse scientific endeavors that seek to understand processes and how observations and analyses influence our understanding....     

张量分类算法的遥感影像目标探测

提出了一种基于张量学习机的遥感影像目标探测方法。该方法基于张量数据模型和张量代数运算, 针对遥感影像数据多维或高维的特点, 将基于向量的监督法学习机扩展为基于张量的监督法学习机, 然后利用凸函数最优化理论和交互投影迭代法求得张量学习机的最优解。最后分别以高光谱遥感影像和高分辨率遥感影像为例, 使用张量学习机进行目标探测。实验表明, 与支持向量机等方法相比, 本文的方法在保持较高探测成功率的同时更好的抑制了虚警。     

A wavelet-extreme learning machine for low-cost INS/GPS navigation system in high-speed applications

The combined navigation system consisting of both global positioning system (GPS) and inertial navigation system (INS) results in reliable, accurate, and continuous navigation capability when compared to either a GPS or an INS stand-alone system. To improve the overall performance of low-cost micro-electro-mechanical systems (MEMS)-based INS/GPS by considering a high level of stochastic noise on low-cost MEMS-based inertial sensors, a highly complex problems with noisy real data, a high-speed vehicle, and GPS signal outage during our experiments, we suggest two approaches at different steps: (1) improving the signal-to-noise ratio of the inertial sensor measurements and attenuating high-frequency noise using the discrete wavelet transform technique before data fusion while preserving important information like the vehicle motion information and (2) enhancing the positioning accuracy and speed by an extreme learning machine (ELM) which has the characteristics of quick learning speed and impressive generalization performance. We present a single-hidden layer feedforward neural network which is employed to optimize the estimation accuracy and speed by minimizing the error, especially in the high-speed vehicle and real-time implementation applications. To validate the performance of our proposed method, the results are compared with an adaptive neuro-fuzzy inference system (ANFIS) and an extended Kalman filter (EKF) method. The achieved accuracies are discussed. The results suggest a promising and superior prospect for ELM in the field of positioning for low-cost MEMS-based inertial sensors in the absence of GPS signal, as it outperforms ANFIS and EKF by approximately 50 and 70%, respectively.     

基于SRTM DEM的InSAR高分辨率山区地表高程重建算法

山体的叠掩和阴影现象造成的信号去相关,一直是InSAR重建山区地表高程的瓶颈之一.为此,提出了一种新的基于粗分辨率SRTM DEM(约90m分辨率)辅助InSAR数据重建山区地表高程的方法.利用SRTM DEM模拟的干涉相位,对ERS-1/2干涉相位做去地形相位处理,得到残余相位.通过对解缠后的残余相位计算方差提取叠掩和阴影区域的噪声,并用平均相位近似恢复噪声区域的相位,然后将其转换为高程,并用SRTM DEM作高程补偿处理,从而实现地表高程重建.最后,定量比较了该方法与传统InSAR技术生成的DEM精度.实验表明,这种方法能有效提高传统InSAR技术生成地表高程的精度,这对提高星载雷达数据的使用效率具有重要意义.     

气溶胶定量遥感的机器学习方法综述

机器学习方法近年来取得突破进展,其遥感应用从目标识别和地物分类领域,发展到定量化反演的多个领域。气溶胶定量遥感因其机理复杂,反演参数的种类和精度受到限制,机器学习为气溶胶遥感带来了新的研究和应用技术手段。本文汇总现有研究进展将气溶胶机器学习方法归纳为卫星遥感反演气溶胶光学厚度AOD（Aerosol Optical Depth）、卫星遥感反演其他气溶胶参数、卫星遥感反演颗粒物浓度（PMx）、地基气溶胶遥感4类。结合作者研究工作,通过分析讨论,归纳机器学习用于气溶胶定量遥感的条件为：（1）物理模型无法使用;（2）已有模型卫星产品精度低;（3）已有模型精度高但计算速度低。从应用的角度来说,可以借助于更多的具有相关性的输入信息,发挥机器学习在反演产品种类、反演精度、计算效率等方面的优势;而对定量遥感来说,应该同时重视挖掘遥感数据本身的信息来提高反演能力,并通过误差分析等手段反馈对遥感机理的理解,使机器学习与遥感机理研究相互促进。     

顾及测区地形的水准网定权新方法研究

针对目前水准网平差中高差定权方法未能较好地考虑地形环境对高差观测误差的影响关系的问题,提出一种顾及测区地形起伏并根据测段往返测高差较差的水准网高差分类定权新方法。理论分析和实际数据证明,在水准网中既有地形起伏较大的测段,又有地形起伏不大的平缓测段时,采用文中提出的定权方法比传统定权方法更能真实反映水准网测量的精度情况,弥补了传统定权方法的不足。     

面向地上下无缝集成建模的新一代三维地理信息系统

地上下无缝集成三维建模是新一代3DGIS的主要标志,地上下无缝集成建模已成为当务之急。在三维空间建模技术现状与存在问题、目标层次与功能需求及三维空间模型现状与趋势分析的基础上,介绍了地上下集成建模的两个层次和无缝集成的基本原理;阐述以CD-TIN为纽带、以BRep-TIN-GTP为核心、以三层混合模型为成份的地上下集成空间数据模型的概念结构与典型逻辑关系。并结合城市与矿山应用,介绍该集成模型在GeoMo3D系统中初步实现后的可视化效果。同时,指出了与地上下集成建模相关的近年3DGIS的主要攻关方向和重点理论难题。     

基于LiDAR的农田地形环境三维重建方法

为克服传统农田土地平整测量方法耗时费力的特点,提出采用LiDAR技术对农田地形进行重建的探索性研究。通过HDL-32E型激光雷达等搭建了系统的硬件平台,应用C++语言编写了系统数据的采集程序;在此基础上对激光雷达所采集数据进行了标定,研究了农田地形重建系统中不同坐标系的转换方法;同时基于最小值去噪法设计了更适用于农田地形点云去噪的均值限差去噪法。通过对比在农田起伏较大区域不同坡度范围内RTK与激光雷达所测单元个数,对系统精度进行了评价;最后实现了车载农田地形重建系统的界面显示、应用与精度评估。结果表明,在10°~15°、25°~30°大坡度范围内激光雷达所获农田地形更为丰富,精度更高。该方法重建的农田地形模型点云数据和原始农田地形点云数据投影面积逼近度可达93%,验证了本文研究方法应用于农田地形环境重建的可行性,同时为今后的土地精细平整工作提供了理论参考与依据。     

Improving settlement selection for small-scale maps using data enrichment and machine learning

Acquiring and formalizing cartographic knowledge still is a challenge, especially when the generalization process concerns small-scale maps. We concentrate on the settlement selection process for small-scale maps, with the aim of rendering it more holistic, and making methodological contributions in four areas. First, we show how written specifications and rules can be validated against the actual published map products, thus pointing to gaps and potential improvements. Second, we use data enrichment based on supplementing information extracted from point-of-interest data in order to assign functional importance to particular settlements. Third, we use machine learning (ML) algorithms to infer additional rules from existing maps, thus making explicit the deep knowledge of cartographers and allowing to extend the cartographic rule set. And fourth, we show how the results of ML can be transformed into human-readable form for potential use in the guidelines of national mapping agencies. We use the case of settlement selection in the small-scale maps published by the Polish national mapping agency (GUGiK). However, we believe that the methods and findings of this paper can be adapted to other environments with minor modifications.     

Geographically weighted regression with the integration of machine learning for spatial prediction

Journal of Geographical Systems - Conventional methods of machine learning have been widely used to generate spatial prediction models because such methods can adaptively learn the mapping...     

地形特征提取的一种简易算法

地形特征作为基本地理要素,从DEM数据中提取其隐含的地形结构信息是非常重要的。本方法采用距离倒数加权法建立格网DEM,提出一种直接比较高程值来提取地形特征的简易算法,并利用差分算法计算坡度和坡向,通过V isual C++软件平台,编程实现格网DEM地形特征的自动提取,进行三维立体显示。该算法易于编程实现,数据结构简单,程序运行效率高,提取结果与真实地形基本拟合。     

一种基于机器学习算法的区域/单站ZTD组合预测模型

针对天顶对流层总延迟(ZTD)具有一定的时空变化特性,提出了一种基于BP神经网络、长短期记忆网络(LSTM)算法的区域/单站ZTD组合预测模型. 以连续14天香港连续运行参考站(CORS)网络18个监测站观测数据为例,利用BP神经网络、LSTM及本文算法进行了区域、单站及二者组合ZTD预测模型研究. HKWS测站的预测结果表明:利用前13天数据预报第14天数据,区域、单站、组合模型ZTD预测的均方根误差（RMSE）分别为10.2 mm、10.4 mm、8.5 mm,组合模型相对于区域、单站模型预测精度分别提升了17.2%、18.4%.      

A support vector machine classifier based on a new kernel function model for hyperspectral data

The kernel function is a key factor to determine the performance of a support vector machine (SVM) classifier. Choosing and constructing appropriate kernel function models has been a hot topic in SVM studies. But so far, its implementation can only rely on the experience and the specific sample characteristics without a unified pattern. Thus, this article explored the related theories and research findings of kernel functions, analyzed the classification characteristics of EO-1 Hyperion hyperspectral imagery, and combined a polynomial kernel function with a radial basis kernel function to form a new kernel function model (PRBF). Then, a hyperspectral remote sensing imagery classifier was constructed based on the PRBF model, and a genetic algorithm (GA) was used to optimize the SVM parameters. On the basis of theoretical analysis, this article completed object classification experiments on the Hyperion hyperspectral imagery of experimental areas and verified the high classification accuracy of the model. The experimental results show that the effect of hyperspectral image classification based on this PRBF model is apparently better than the model established by a single global or local kernel function and thus can greatly improve the accuracy of object identification and classification. The highest overall classification accuracy and kappa coefficient reached 93.246% and 0.907, respectively, in all experiments.     

AMSR-E地表温度数据重建深度学习方法

地表温度对于全球气候变化等研究具有重要意义。被动微波遥感传感器AMSR-E (Advanced Microwave Scanning Radiometer for EOS)可以获得全天候地表温度,可作为多云条件下热红外地表温度数据的补充;但轨道扫描间隙限制了该数据在全球或区域尺度上的实际应用。鉴于地表温度的高时空异质性和AMSR-E LST轨道间隙数据的特点,本文提出了一种多时相特征连接卷积神经网络地表温度双向重建模型（MTFC-CNN）,利用深度学习在处理复杂非线性问题上的优势,重建轨道间隙区域的地表温度值。将2010年中国大陆四季的AMSR-E LST数据（数据未含港澳台区域）,分为白天和夜晚,形成共8个数据子集进行实验。在模拟实验中,重建结果与原始反演地表温度值平均均方根误差在1.0 K左右,决定系数R2在0.88以上,优于传统的样条空间插值和时间线性回归方法;真实实验结果具有较好的目视效果,且与对应MODIS LST产品对比发现,重建区LST值和未重建区LST值与MODIS LST产品间具有相近的平均均方根误差和决定系数。因此,本文提出的MTFC-CNN方法能有效重建AMSR-...     

基于Sentinel-1A数据的多种机器学习算法识别冰山的比较

冰山识别对于海洋环境监测和船只安全运行等具有重要的意义,是北极航道开通和北极开发过程中的重要内容。采用合成孔径雷达(SAR)影像进行冰山识别具有独特的优势,多种机器学习算法均可用于SAR影像的冰山识别中。为了最大限度地发挥机器学习算法的性能,有必要对不同机器学习算法及其搭配使用的特征与特征标准化方法进行评估,从而进行最优冰山识别方法的选择。因此,本文基于Sentinel-1A SAR影像,采用多种机器学习方法、多种特征组合及多种特征标准化方法进行冰山识别,并比较各流程方法的识别性能差异。采用的机器学习算法包括贝叶斯分类器(Bayes)、反向神经网络(BPNN)、线性判别分析(LDA)、随机森林(RF)以及支持向量机(SVM);特征标准化方法包括Min-max标准化、Z-score标准化及log函数标准化;数据集是含有12个SAR影像特征的969个冰山与非冰山样本,样本主要位于格陵兰岛东海岸。分类效果采用接收者操作特性(ROC)曲线下的面积(AUC)进行衡量。结果显示,最佳搭配下的RF的AUC值最高,达到了0.945,比最差的Bayes高出0.09。从识别率上来看,RF在冰山查全率为80%的情况下非冰山查全率达到92.6%,效果最好,比第2位的BPNN高出1.4%,比最差的Bayes高出2.6%;BPNN在冰山查全率为90%的情况下非冰山查全率达到87.4%,比第2位的RF高出0.8%,比最差的Bayes高出2.7%。上述结果表明,对冰山识别而言,选择最优的机器学习算法和最佳的特征与特征标准化方法都是十分重要的。