An introduction to the classification of the physiognomic landscape types: methodology and results of testing in the area of Kazimierz Landscape Park (Poland)

The publication of the European Landscape Convention (2000) had a stimulating effect on the development of both new systems of landscape classification and new methods of their evaluation and mapping. As an example, a new classification of physiognomic landscape types was developed in Poland in 2014. The objectives of the paper are to (1) popularize, on the international scale, the classification of physiognomic landscape types in a new, improved version, (2) present the original method of physiognomic landscape types mapping with the use of the system of basic landscape units, and (3) present the results of testing of the method in the area of the Kazimierz Landscape Park, Poland. In the area of the Kazimierz Landscape Park, 491 basic landscape units were delimited, within which, on the basis of the leading traits of land relief and cover forms, the physiognomic landscape types were identified. Maps of this type can be important tools in implementing the recommendations of the European Landscape Convention. The total number of physiognomic landscape types identified within a specific area can be one of the key indicators of landscape diversity.     

基于国产高分遥感的人工种植红树林种间分类方法研究——以广西茅尾海红树林为例

红树林的种间结构组成对红树林生态系统的健康和发展至关重要,而红树林种间分类问题一直以来都是基于遥感手段的红树林监测中的难点。针对该问题,以人工种植为特点的广西茅尾海红树林遥感种间分类为例,基于面向对象的分类思想,提出了一种现场样本与分割对象相结合的红树林种间分类方法。利用GF-2 PMS1高分辨率卫星遥感影像数据,开展了广西茅尾海红树林湿地典型植被精细分类和空间分布研究,并将分类结果与基于像素和传统面向对象SVM分类方法进行了对比。结果显示:总体上,面向对象分类方法更适合用于茅尾海红树林湿地典型植被分类;对于局部混生明显的区域使用基于像素SVM分类方法效果会更好;传统面向对象分类方法中将整个影像分割对象单元作为训练样本可能会在某种程度上造成负面影响。因此,使用文中提出的样本选择新方法进行面向对象分类精度最高,总体精度达到了93.13%,Kappa为0.89。     

高分辨率光学遥感场景分类的深度度量学习方法

针对高分辨率光学遥感影像场景具有同类型内部差异大、不同类型间相似度高导致部分场景识别困难的问题,本文提出了一种深度度量学习方法。首先在深度学习模型的特征输出层上为每类预设聚类中心,其次基于欧氏距离方法设计均值中心度量损失项,最后联合交叉熵损失项以及权重与偏置正则项构成模型的损失函数。该方法的目标是在特征空间上使同类型特征聚集并扩大类型间的距离以提高分类准确率。试验结果表明,本文方法有效地提升了分类准确率。在RSSCN7、UC Merced和NWPU-RESISC45数据集上,与现有方法相比,分类准确率分别提高了1.46%、1.09%和2.51%。     

机载多光谱LiDAR的随机森林地物分类

机载多光谱LiDAR技术利用激光进行探测和测距,不仅可以快速获取地面物体的三维坐标,还可以获得多个波段的地物光谱信息,可广泛用于地形测绘、土地覆盖分类、环境建模、森林资源调查等。本文提出了多光谱LiDAR的随机森林地物分类方法。该方法通过对LiDAR强度数据和高程数据提取分类特征,完成多光谱LiDAR的随机森林地物分类;并分析随机森林的特征贡献度特性,采用后向特征选择方法实现分类特征选择。通过对加拿大Optech Titan多光谱LiDAR数据的试验表明：随机森林方法可以获得较好的地物分类精度,而且可以适当地去除部分冗余和相关的特征,从而有效提高分类精度。     

一种高光谱遥感图像快速谱聚类算法

为获得分类效果更优良的遥感图像分类方式并解决高光谱遥感图像分类运算速度缓慢的问题,集成Lanczos算法与谱聚类算法,探讨了高光谱遥感图像谱聚类算法应用于遥感图像分类的可行性,提出了一种面向高光谱遥感图像的快速谱聚类算法;通过对比美国圣地亚哥机场高光谱遥感图像K-均值算法与谱聚类算法的分类结果,发现面向高光谱遥感图像的谱聚类算法易于识别线性地物,且分类的速度能得到较大提升。     

结合主动学习和词袋模型的高分二号遥感影像自动化分类

高分卫星遥感影像空间分辨率的提高,使得地物的光谱和纹理变得更加丰富和复杂,这给遥感影像的自动化分类带来严重挑战。因此,本文提出了一种结合主动学习和词袋模型的高分二号遥感影像分类方法。首先,对研究区域进行多尺度分割,建立影像分割对象集;然后,采用词袋模型构建影像对象的语义特征向量;最后,充分考虑位于分类边界的不确定性样本分布,迭代选择最优样本用于训练支持向量机,用于分类遥感影像。为了验证本文方法的有效性和稳健性,以山东省某市的高分二号遥感影像为试验数据进行了试验分析。结果表明,本文提出的方法可以有效地将研究区域分为水体、地面、植被和建筑物四类,正确率达到90.6%以上。     

语义因子支持的多源POI分类信息一致化处理

为了解决网络POI分类异构导致多源POI数据匹配与整合困难的问题,该文利用分类特征词提取、特征词去重与优化等关键技术,构建POI分类的语义概念格,然后通过概念格之间的语义关联关系,实现多源异构POI分类体系的映射与转换。最后通过实验验证了该文方法的有效性,并将其成功应用于省级地理信息公共服务平台(政务版)数据更新。     

Synergetic efficiency of Lidar and WorldView-2 for 3D urban cartography in Northeast Mexico

Three-dimensional urban cartography is needed for city changes’ assessment. The variety of studies using 3D calculations of urban elements grows each year. Building and vegetation volumes are necessary to assess and understand spatio-temporal urban changeable environments. However, there are technical questions as to which method can improve 3D urban cartographic accuracy. The innovative part of this current study is the creation of a six-band hybrid obtained from LIDAR and WorldView2 synergy. Two different enhancement algorithms demonstrated the most important spectral features for the urban development and vegetation classes. Results indicated an improvement in accuracy by up to 21.3%, according to the Kappa coefficient. Both infra-red band and intensity band were the most significant, according to the principal components analysis. The synergy delimited classes and polygons, as well as the direct display of information regarding heights of elements and improving the extraction of roads, buildings and vegetation classes.     

Geospatial data mining for digital raster mapping

We performed an in-depth literature survey to identify the most popular data mining approaches that have been applied for raster mapping of ecological parameters through the use of Geographic Information Systems (GIS) and remotely sensed data. Popular data mining approaches included decision trees or “data mining” trees which consist of regression and classification trees, random forests, neural networks, and support vector machines. The advantages of each data mining approach as well as approaches to avoid overfitting are subsequently discussed. We also provide suggestions and examples for the mapping of problematic variables or classes, future or historical projections, and avoidance of model bias. Finally, we address the separate issues of parallel processing, error mapping, and incorporation of “no data” values into modeling processes. Given the improved availability of digital spatial products and remote sensing products, data mining approaches combined with parallel processing potentials should greatly improve the quality and extent of ecological datasets.