A SEM-based method to determine the mineralogical composition and the particle size distribution of suspended sediment

A robust method for characterizing the mineralogy of suspended sediment in continental rivers is introduced. It encompasses 3 steps: the filtration of a few milliliters of water, measurements of X-ray energy dispersive spectra using Scanning Electron Microscopy (SEM), and robust machine learning tools of classification. The method is applied to suspended particles collected from various Amazonian rivers. A total of more than 204,000 particles were analyzed by SEM-EDXS (Energy Dispersive X-ray Spectroscopy), i.e. about 15,700 particles per sampling station, which lead to the identification of 15 distinct groups of mineralogical phases. The size distribution of particles collected on the filters was derived from the SEM micrographs taken in the backscattered electron imaging mode and analyzed with ImageJ freeware. The determination of the main mineralogical groups composing the bulk sediment associated with physical parameters such as particle size distribution or aspect ratio allows a precise characterization of the load of the terrigenous particles in rivers or lakes. In the case of the Amazonian rivers investigated, the results show that the identified mineralogies are consistent with previous studies as well as between the different samples collected. The method enabled the evolution of grain size distribution from fine to coarse material to be described in the water column. Implications about hydrodynamic sorting of mineral particles in the water column are also briefly discussed. The proposed method appears well suited for intensive routine monitoring of suspended sediment in river systems.     

全卷积神经网络在建筑物震害遥感提取中的应用研究

为解决建筑物震害信息提取自动化程度不高的问题,本文将全卷积神经网络应用于建筑物震害遥感信息提取。以玉树地震后获取的玉树县城区0.2m分辨率航空影像作为建筑物震害信息提取试验数据源,将试验区地物划分为倒塌建筑物、未倒塌建筑物和背景3类。对427个500×500像素的子影像进行人工分类与标注,选取393个组成训练样本集,34个用于验证。利用训练样本集对全卷积神经网络进行训练,采用训练后的网络对验证样本进行建筑物震害信息提取及精度评价。研究结果表明:建筑物震害遥感信息提取总体分类精度为82.3%,全卷积神经网络方法能提高信息提取自动化程度,具有较好的建筑物震害信息提取能力。     

Inferring user tasks in pedestrian navigation from eye movement data in real-world environments

Eye movement data convey a wealth of information that can be used to probe human behaviour and cognitive processes. To date, eye tracking studies have mainly focused on laboratory-based evaluations of cartographic interfaces; in contrast, little attention has been paid to eye movement data mining for real-world applications. In this study, we propose using machine-learning methods to infer user tasks from eye movement data in real-world pedestrian navigation scenarios. We conducted a real-world pedestrian navigation experiment in which we recorded eye movement data from 38 participants. We trained and cross-validated a random forest classifier for classifying five common navigation tasks using five types of eye movement features. The results show that the classifier can achieve an overall accuracy of 67%. We found that statistical eye movement features and saccade encoding features are more useful than the other investigated types of features for distinguishing user tasks. We also identified that the choice of classifier, the time window size and the eye movement features considered are all important factors that influence task inference performance. Results of the research open doors to some potential real-world innovative applications, such as navigation systems that can provide task-related information depending on the task a user is performing.     

融合城市行道树特征选取模型的自适应深度学习分类

针对城市行道树的学习多分类问题,本文在综合分析城市行道树多分类特征的基础上,提出一种融合特征自动选取模型的自适应深度学习方法。基于随机森林法,学习行道树的特征重要性,通过特征消除方法舍弃不重要的特征,实现城市行道树多分类特征自动选取;在城市行道树分类特征工程提取的基础上,构建了城市行道树多分类问题的自适应深度学习方法,并采用交叉验证与参数搜索方法,对所提出的深度学习模型进行改进。试验结果表明,本文所提出的融合特征自动选取模型的自适应深度学习方法具有良好性能,解决了城市行道树多分类预测的准确性与泛化问题。     

深度置信网络算法反演Landsat 8 OLI气溶胶光学厚度

传统的气溶胶遥感反演算法在地表反射率较低、结构较为均一的海洋及浓密植被等区域的气溶胶反演可以达到较高的精度,而在城市、矿区等高亮度、高异质性区域的气溶胶反演中仍面临较大的挑战。当地表反射率较高时,卫星传感器获取的对气溶胶具有标识性的信息不足,导致了气溶胶反演的困难。为更大程度地挖掘卫星信号中对气溶胶具有标识性的信息,本文提出使用深度学习技术的气溶胶遥感反演算法,用于Landsat 8 OLI传感器的气溶胶反演。选择全球不同区域的AERONET站点气溶胶实测数据以及对应区域的Landsat 8 OLI传感器的观测几何角度和表观反射率数据,根据合理的时空匹配原则构建样本数据。选择深度置信网络,在合理设置训练批次和训练次数的基础上对网络进行训练和测试,生成关于卫星传感器数据的气溶胶光学厚度拟合网络模型,实现气溶胶遥感反演。使用独立的AERONET站点气溶胶实测数据对反演结果进行了验证,结果表明：该方法可反演不同地表类型区域连续空间覆盖的气溶胶光学厚度,且达到了较高的精度（R=0.8745,RMSE=0.0391,MAE=0.0616,EE=87.94%）。与传统的方法相比,本方法基于单时相卫星遥感数据即可实现气溶胶的高精度反演,简化了气溶胶反演的步骤,提高了气溶胶反演的稳定性和时空适应性。     

高分六号红边特征的农作物识别与评估

红边作为植被敏感波段,其红边特征的运用是遥感识别农作物并实现精准农业的高新手段之一。以黑龙江松嫩平原北部为研究区,以国内首个提供红边波段的多光谱高分六号影像和玉米、大豆、水稻总计82859个作物样本同时作为研究对象,从以下几个方面研究了红边波段和红边指数波段等红边特征在农作物识别中的表现,并评估了农作物的识别精度。（1） 通过作物样本辐射亮度值的统计特征,初步显示了在两红边波段0.710 μm和0.750 μm处有比其他波段更好的区分;（2） 根据传统归一化植被指数形式构建了红边归一化植被指数NDVI710和NDVI750,综合两指数在J-M距离表征的作物样本类别区分度上比传统NDVI更显著;（3） 通过多种手段筛选了有效波段并且制定了支持向量机（SVM）框架下4种农作物识别的分类策略,分别在5∶5、6∶4、7∶3、8∶2、9∶1等5套随机样本分割方案下完成研究区域农作物的分类预测。在这20类分类精度中kappa系数均高于0.9609,总体精度高于0.9742;列向上5∶5分割方案的精度最高,8∶2的精度最低;横向上分类精度排序如下：SVM-RFE > SVM-RF > SVM-有红边波段 > SVM-无红边波段,该结果表明了红边指数和红边波段的参与显著地提高了作物的识别精度;（4） 由于水域等其他样本的缺少,SVM-RFE方法和SVM-RF方法的分类图像均存在少量错分现象。但从分类精度和图像细节展示上来看,SVM-RFE方法要优于SVM-RF方法,二者分类图像的交叉验证中kappa系数为0.8060,总体精度为0.8743。总之,高分六号红边特征在作物识别中表现优越,使得识别精度显著提高。后续研究者可开发更多与红边相关的植被指数,充分发挥红边特征在精准农业中的作用。     

基于高效统计模型预测9月北极航道沿线海冰范围

The rapid decrease in Arctic sea ice cover and thickness not only has a linkage with extreme weather in the midlatitudes but also brings more opportunities for Arctic shipping routes and polar resource exploration, both of which motivate us to further understand causes of sea-ice variations and to obtain more accurate estimates of seaice cover in the future. Here, a novel data-driven method, the causal effect networks algorithm, is applied to identify the direct precursors of September sea-ice extent covering the Northern Sea Route and Transpolar Sea Route at different lead times so that statistical models can be constructed for sea-ice prediction. The whole study area was also divided into two parts: the northern region covered by multiyear ice and the southern region covered by seasonal ice. The forecast models of September sea-ice extent in the whole study area(TSIE) and southern region(SSIE) at lead times of 1–4 months can explain over 65% and 79% of the variances, respectively,but the forecast skill of sea-ice extent in the northern region(NSIE) is limited at a lead time of 1 month. At lead times of 1–4 months, local sea-ice concentration and sea-ice thickness have a larger influence on September TSIE and SSIE than other teleconnection factors. When the lead time is more than 4 months, the surface meridional wind anomaly from northern Europe in the preceding autumn or early winter is dominant for September TSIE variations but is comparable to thermodynamic factors for NSIE and SSIE. We suggest that this study provides a complementary approach for predicting regional sea ice and is helpful in evaluating and improving climate models.     

全球水冲突问题研究进展与热点的可视化分析

水冲突是发生在国家、州（省）、团体等不同利益主体之间关于水资源获取的矛盾。冷战以来,水冲突相关事件的总体烈度下降,事件数量大幅增长;与此同时,水冲突问题的研究也开始获得广泛关注,研究热度快速提升。水冲突问题关系到区域安全和可持续发展,有必要对相关研究进行梳理。为此,利用非监督机器学习、空间分析以及数据可视化方法,对国内外水冲突相关研究进行梳理,统计了相关研究的空间结构,构建了研究文献语料库,归纳了水冲突研究的主要热点议题。发现：水冲突相关研究具有明显的空间异质性,且不同的议题在时间上呈现不同的演进趋势;水冲突相关研究的主要议题包括：水冲突成因、水冲突分析方法与模型、水冲突解决机制、气候变化对水冲突的影响,以及水霸权主义与国际水法等。     

基于反射率和亮度温度的锡林郭勒积雪深度估算

积雪是我国西北干旱半干旱区重要的水资源,也是影响全球气候变化的重要因子之一。 目前光学影像反射率和雷达亮温数据是积雪遥感领域的主要数据,本文首次结合两类遥感数据估 算积雪深度,并比较偏最小二乘法和机器学习算法（人工神经网络、支持向量机和随机森林算法） 在积雪深度估算方面的表现。以锡林郭勒盟 2012—2015 年积雪深度数据为例,基于反射率和亮度 温度相结合的积雪深度估算精度优于单个数据源,且随机森林算法表现最好,均方根误差为 2.93 cm,满足实际应用的需求。研究结果对我国西北地区水资源分布、生态环境评估等研究具有重要 意义。     

喀斯特山区石漠化耕地遥感精准提取与分析——以贵州省北盘江镇与花江镇为例

选取北盘江镇与花江镇作为研究区,利用谷歌高精度遥感影像,结合分区分层分级思想,基于深度学习与传统约束方法对研究区耕地进行精准提取。结果表明：1）在数量精度上,以视觉形态差异对研究区进行分区并选取不同模型训练获得的精细地块,面积约为9 867 hm ²,与实际数量基本一致,F-Measure主要分布在[0.82, 0.98]之间,受到地形和岩石裸露率的影响,石漠化严重地区耕地提取精度较低。2）在形态精度上,预测耕地与实际耕地的GIOU主要分布在[0.7, 1]之间,分割正确率>0.85,表明预测耕地边界与实际地块边界吻合度高,提取结果符合研究区实际情况。3）利用地貌坡度等约束条件对耕地进行划分发现,研究区以喀斯特耕地为主,占比74%,并且石漠化程度较严重,其中轻度石漠化耕地与中度石漠化耕地占总耕地的32%。在石漠化地区,耕地的狭长程度、破碎程度受人类影响较大;距离居民地越近、可达性越高的地区,其土地利用率越高,斑块越破碎。文章所提出方法可用于耕地破碎、地形复杂地区的耕地提取,能够为地区的发展、耕地治理与研究、环境保护与决策等提供精准数据支持。