Toward accountable land use mapping: Using geocomputation to improve classification accuracy and reveal uncertainty

The classification of satellite imagery into land use/cover maps is a major challenge in the field of remote sensing. This research aimed at improving the classification accuracy while also revealing uncertain areas by employing a geocomputational approach. We computed numerous land use maps by considering both image texture and band ratio information in the classification procedure. For each land use class, those classifications with the highest class-accuracy were selected and combined into class-probability maps. By selecting the land use class with highest probability for each pixel, we created a hard classification. We stored the corresponding class probabilities in a separate map, indicating the spatial uncertainty in the hard classification. By combining the uncertainty map and the hard classification we created a probability-based land use map, containing spatial estimates of the uncertainty. The technique was tested for both ASTER and Landsat 5 satellite imagery of Gorizia, Italy, and resulted in a 34% and 31% increase, respectively, in the kappa coefficient of classification accuracy. We believe that geocomputational classification methods can be used generally to improve land use and land cover classification from imagery, and to help incorporate classification uncertainty into the resultant map themes.     

增强型DeepLab算法和自适应损失函数的高分辨率遥感影像分类

高分辨率遥感影像地物复杂,分类难度大,而深度学习方法可以提取地物更多更深层次的特征信息,适用于高分辨率遥感影像的地物分类。本文研究对高分辨率影像中不透水地面、建筑、低矮植被、树、车辆等地物的高精度分类。结合遥感多地物分类的特点,以DeepLab v3+网络模型为基础,提出E-DeepLab网络模型。主要改进为：（1）改进编码器和解码器的结合方式,使用简洁有效的加成连接方式。（2）缩小单次上采样倍数,增加上采样层,提高编码器与解码器连接的紧密性。（3）使用改进的自适应权重损失函数,自动调节地物损失权重。同时根据数据特点,提出结合DSM、NDVI数据等多通道训练方式。使用两个地区数据进行实验,结果表明,两地区精度均明显优于原始DeepLab v3+模型和其他相关模型,Potsdam地区总体提取精度达到93.2%,建筑物提取精度达到97.8%,Vaihingen地区总体提取精度达到90.7%,建筑物提取精度达到96.3%。目视对比分类图和标准标记图,两者具有高度的一致性。本文所提出的E-DeepLab网络在高分辨率遥感影像地物高精度提取和分类中有较好的应用价值。     

A novel image-fusion method based on the un-mixing of mixed MS sub-pixels regarding high-resolution DSM

A major reason for the spectral distortions of fused images generated by current image-fusion methods is that the fused versions of mixed multispectral (MS) sub-pixels (MSPs) corresponding to panchromatic (PAN) pure pixels remain mixed. The MSPs can be un-mixed spectrally to pure pixels having the same land cover classes in a fine classification map during the fusion process. Since it is difficult to produce such a land cover classification map using only MS and PAN images, a Digital Surface Model (DSM) derived from airborne Light Detection And Ranging data were employed in this study to facilitate the classification. In a novel fusion method proposed in this paper, MSPs near and across boundaries between vegetation and non-vegetation are identified using MS, PAN, and normalized Digital Surface Model (nDSM). The identified MSPs then are fused to pure pixels with respect to the corresponding land cover class in the classification map. In a test on WorldView-2 images over an urban area and the corresponding nDSM, the fused image generated by the proposed method was visually and quantitatively compared with fused images obtained using common image-fusion methods. The fused images generated by the proposed method yielded minimal spectral distortions and sharpened boundaries between vegetation and non-vegetation.     

Phenological metrics-based crop classification using HJ-1 CCD images and Landsat 8 imagery

Crop type data are an important piece of information for many applications in agriculture. Extracting crop type using remote sensing is not easy because multiple crops are usually planted into small parcels with limited availability of satellite images due to weather conditions. In this research, we aim at producing crop maps for areas with abundant rainfall and small-sized parcels by making full use of Landsat 8 and HJ-1 charge-coupled device (CCD) data. We masked out non-vegetation areas by using Landsat 8 images and then extracted a crop map from a long-term time-series of HJ-1 CCD satellite images acquired at 30-m spatial resolution and two-day temporal resolution. To increase accuracy, four key phenological metrics of crops were extracted from time-series Normalized Difference Vegetation Index curves plotted from the HJ-1 CCD images. These phenological metrics were used to further identify each of the crop types with less, but easier to access, ancillary field survey data. We used crop area data from the Jingzhou statistical yearbook and 5.8-m spatial resolution ZY-3 satellite images to perform an accuracy assessment. The results show that our classification accuracy was 92% when compared with the highly accurate but limited ZY-3 images and matched up to 80% to the statistical crop areas.     

Mapping of aggregated floodplain plant communities using image fusion of CASI and LiDAR data

Combined optical and laser altimeter data offer the potential to map and monitor plant communities based on their spectral and structural characteristics. A problem unresolved is, however, that narrowly defined plant communities, i.e. plant communities at a low hierarchical level of classification in the Braun-Blanquet system, often cannot be linked directly to remote sensing data for vegetation mapping. We studied whether and how a floristic dataset can be aggregated into a few major discrete, mappable classes without substantial loss of ecological meaning. Multi-source airborne data (CASI and LiDAR) and floristic field data were collected for a floodplain along the river Waal in the Netherlands. Mapping results based on floristic similarity alone did not achieve highest levels of accuracy. Ordination of floristic data showed that terrain elevation and soil moisture were the main underlying environmental drivers shaping the floodplain vegetation, but grouping of plant communities based on their position in the ordination space is not always obvious. Combined ordination-based grouping with floristic similarity clustering led to syntaxonomically relevant aggregated plant assemblages and yielded highest mapping accuracies.     

通过训练样本采样处理改善小宗作物遥感识别精度

训练样本质量是决定农作物遥感识别精度的关键因素,虽然高空间分辨率卫星的发展有效地解决了农作物遥感识别过程中的混合像元问题,但是当区域内不同作物种植面积差异较大时,训练集中不同类别样本数量往往相差较大,这样的不均衡数据集影响分类器的训练,导致少数类别的识别精度不理想。为研究作物遥感识别过程中的不均衡样本问题,本文基于GF-2号卫星数据,首先挖掘了地物的光谱信息、纹理信息,用特征递归消除RFE (Recursive Feature Elimination)方法进行特征优选,然后从数据处理的角度采用了5种采样算法对不均衡训练集进行处理,最后使用采样后的均衡数据集训练分类器,对比数据采样前后决策树与Adaboost(Adaptive Boosting)两种分类器的识别结果,发现:(1)经过采样处理后两种分类算法明显提升了小宗作物的分类精度;(2)经过ADASYS (Adaptive synthetic sampling)采样处理后,分类器性能提升最多,决策树的Kappa系数提高了14.32%,Adaboost的Kappa系数提高了10.23%,达到最高值0.9336;(3)过采样的处理效果优于欠采样,过采样对分类器的性能提升更多。综上所述,选择合适的采样方法和分类方法是提高不均衡数据集遥感分类精度的有效途径。     

综合非光谱信息的荒漠化土地CART分类

用遥感手段对荒漠化进行监测是当前荒漠化研究的热点问题,传统的荒漠化遥感信息自动提取方法是基于光谱特征的图像分割,受多种因素的影响,分类精度的提高遇到瓶颈,因此基于知识的分类方法应运而生。CART是一种非参数化的分类与回归方法,在用于遥感影像自动分类时,可以方便地应用多源知识,提高分类精度。本文在分析了CART方法原理的基础上,针对荒漠化地区各种地物的特点,将包括地物光谱知识、纹理知识、植被盖度等在内的多种知识融入CART模型,克服了单纯利用光谱特征进行分类的不足,取得了85.94%的精度。     

A semi-ellipsoid-model based fuzzy classifier to map grassland in Inner Mongolia,China

Remote sensing techniques offer effective means for mapping plant communities. However, mapping grassland with fine vegetative classes over large areas has been challenging for either the coarse resolutions of remotely sensed images or the high costs of acquiring images with high-resolutions. An improved hybrid-fuzzy-classifier (HFC) derived from a semi-ellipsoid-model (SEM) is developed in this paper to achieve higher accuracy for classifying grasslands with Landsat images. The Xilin River Basin, Inner Mongolia, China, is chosen as the study area, because an acceptable volume of ground truthing data was previously collected by multiple research communities. The accuracy assessment is based on the comparison of the classification outcomes from four types of image sets: (1) Landsat ETM+ August 14, 2004, (2) Landsat TM August 12, 2009, (3) the fused images of ETM+ with CBERS, and (4) TM with CBERS, respectively, and by three classifiers, the proposed HFC-SEM, the tetragonal pyramid model (TPM) based HFC, and the support vector machine method. In all twelve classification experiments, the HFC-SEM classifier had the best overall accuracy statistics. This finding indicates that the medium resolution Landsat images can be used to map grassland vegetation with good vegetative detail when the proper classifier is applied.     

顾及纹理特征贡献度的变化影像对象提取算法

遥感影像变化检测是全球变化研究的重要内容。基于两期遥感影像的变化检测方法存在数据条件要求苛刻、难以充分利用快速发展的多源遥感影像数据等问题。目前许多变化检测的参考数据中包含了一期分类矢量数据,矢量数据中往往包含了位置、形状、大小和类别属性等先验信息,充分利用这些先验信息将可提高变化检测精度。提取变化影像对象是结合矢量数据和遥感影像进行变化检测的核心步骤。本文提出了一种顾及纹理特征贡献度的变化影像对象提取方法。该方法利用矢量数据分割遥感影像,获取影像对象,计算影像对象纹理特征值。根据信息增益原理计算纹理特征参数的特征贡献度,选择特征参数。由贡献度指数大小确定纹理特征参数权重,计算影像对象与先验要素类别的相似度系数,提取变化影像对象。试验结果表明,基于纹理特征贡献度的特征参数选择,能有效地提高变化影像对象提取结果的精度。     

New vegetation type map of India prepared using satellite remote sensing: Comparison with global vegetation maps and utilities

A seamless vegetation type map of India (scale 1: 50,000) prepared using medium-resolution IRS LISS-III images is presented. The map was created using an on-screen visual interpretation technique and has an accuracy of 90%, as assessed using 15,565 ground control points. India has hitherto been using potential vegetation/forest type map prepared by Champion and Seth in 1968. We characterized and mapped further the vegetation type distribution in the country in terms of occurrence and distribution, area occupancy, percentage of protected area (PA) covered by each vegetation type, range of elevation, mean annual temperature and precipitation over the past 100 years. A remote sensing-amenable hierarchical classification scheme that accommodates natural and semi-natural systems was conceptualized, and the natural vegetation was classified into forests, scrub/shrub lands and grasslands on the basis of extent of vegetation cover. We discuss the distribution and potential utility of the vegetation type map in a broad range of ecological, climatic and conservation applications from global, national and local perspectives. We used 15,565 ground control points to assess the accuracy of products available globally (i.e., GlobCover, Holdridge’s life zone map and potential natural vegetation (PNV) maps). Hence we recommend that the map prepared herein be used widely. This vegetation type map is the most comprehensive one developed for India so far. It was prepared using 23.5 m seasonal satellite remote sensing data, field samples and information relating to the biogeography, climate and soil. The digital map is now available through a web portal (http://bis.iirs.gov.in).     

从高光谱遥感影像提取植被信息

遥感可以快速有效地监测大面积植被的种类、特性、长势等各类信息。高光谱遥感数据因其特有的高光谱分辨率特性使其在植被生态环境领域具有极大的应用潜力。植被信息作为生态环境评价的重要参数对区域生态环境的监测和建设具有重要的意义。本文基于云南省鹤庆县北衙的高光谱遥感数据用SAM方法对植被信息进行了提取,参考光谱使用ASD光谱辐射仪采集的植被光谱曲线。文中对高光谱遥感影像的辐射定标和大气校正进行了研究,针对影响光谱辐射仪采集的主要因素采取了相应的措施,并对光谱曲线分类及参考光谱曲线的选取进行了研究。将选取出的参考光谱曲线与大气校正后的遥感影像进行SAM匹配提取出植被信息,经过与实地调查资料比较并计算总体精度和kappa系数,计算结果达到预期精度。最后将分类结果转换为矢量图,经过投影转换为大地坐标后制作出北衙植被分布图。     

Mapping Wildfire Burn Severity in Southern California Forests and Shrublands Using Enhanced Thematic Mapper Imagery

Abstract

Wildfire is a major disturbance agent in Mediterranean Type Ecosystems (MTEs). Providing reliable, quantitative information on the area of burns and the level of damage caused is therefore important both for guiding resource management and global change monitoring. Previous studies have successfully mapped burn severity using remote sensing, but reliable accuracy has yet to be gained using standard methods over different vegetation types. The objective of this research was to classify burn severity across several vegetation types using Landsat ETM imagery in two areas affected by wildfire in southern California in June 1999. Spectral mixture analysis (SMA) using four reference endmembers (vegetation, soil, shade, non‐photosynthetic vegetation) and a single (charcoal‐ash) image endmember were used to enhance imagery prior to burn severity classification using decision trees. SMA provided a robust technique for enhancing fire‐affected areas due to its ability to extract sub‐pixel information and minimize the effects of topography on single date satellite data. Overall kappa classification accuracy results were high (0.71 and 0.85, respectively) for the burned areas, using five canopy consumption classes. Individual severity class accuracies ranged from 0.5 to 0.94.     

利用矢量影像法进行土地利用变化自动检测

为解决土地利用矢量图与遥感影像的变化检测问题,提出了一种基于类别的矢量图与遥感影像变化检测方法。在矢量图约束下,对遥感影像进行影像分割获取像斑;提取像斑在遥感影像上的直方图特征,采用G统计量度量像斑之间的特征距离;利用像斑与其他相同类别像斑之间的特征距离,构建单波段上像斑的类别异质度,自适应加权组合各波段上像斑的类别异质度构建像斑的类别异质度;依据最大熵方法获取各地物类别对应的异质度阈值,以类别为单位对各像斑进行变化判别,获取变化检测结果。在QuickBird遥感影像上的试验验证了本文方法的有效性,实现了矢量图与遥感影像的自动变化检测。     

面向对象与卷积神经网络模型的GF-6 WFV影像作物分类

GF-6 WFV影像是中国首颗带有红边波段的中高分辨率8波段多光谱卫星的遥感影像,对于其影像及红边波段对作物分类影响的研究利用亟待展开。本文结合面向对象和深度学习提出一种适用于GF-6 WFV红边波段的卷积神经网络(RE-CNN)遥感影像作物分类方法。首先采用多尺度分割和ESP工具选择最佳分割参数完成影像分割,通过面向对象的CART决策树消除椒盐现象的同时提取植被区域,并转化为卷积神经网络的输入数据,最后基于Python和Numpy库构建的卷积神经网络模型(RE-CNN)用于影像作物分类及精度验证。有无红边波段的两组分类实验结果表明:在红边波段组,卷积神经网络(RE-CNN)作物分类识别取得了较好的效果,总体精度高达94.38%,相比无红边波段组分类精度提高了2.83%,验证了GF-6 WFV红边波段对作物分类的有效性。为GF-6 WFV红边波段影像用于作物的分类研究提供技术参考和借鉴价值。     

地理信息系统支持下的知识获取及其在遥感影像植被分类中的应用研究

本文根据植被类型分布与地理环境因子的关系,在地理信息系统和遥感技术支持下,通过ＧＩＳ叠加、统计分析操作,建立植被分布与年积温、降水量、海拔高度、土壤类型等环境因子的定量化知识向量表。综合应用所得到的地学知识向量表和植被光谱特征值进行分类试验,得到研究区的植被分布图。文章以贺兰山地区为例,详细介绍该方法的应用。     

基于高光谱数据和专家决策法提取红树林群落类型信息

高光谱遥感是进行地表植被观测的强有力工具,研究并验证有效的算法和数据支撑技术,对于合理利用高光谱数据进行地表植被监测与分析至关重要。在光谱特征分析和地面调查的基础上,基于决策树方法和高光谱分析方法的组合,以深圳市福田国家级自然保护区为例,利用高光谱数据进行红树林群落信息提取的实证研究。结果证实了Hymap数据对于红树林群落类型信息提取的数据支撑能力,以及相关方法用于红树林分类研究方面的有效性。     

利用Sentinel-2A多光谱成像仪与Landsat8陆地成像仪影像进行普陀山岛植被分类效果比较

卫星遥感技术可用于海岛资源调查。Sentinel-2A与Landsat 8两颗卫星都可免费提供空间分辨率较高的多光谱遥感影像,在海岛调查中的应用潜力较大。本文以浙江舟山普陀山岛为例开展了针对这两种影像在海岛植被分类中的应用效果的研究,分别利用Sentinel-2A多光谱成像仪（MSI）和Landsat 8陆地成像仪（OLI）影像基于最大似然法分类获得了该岛阔叶林、针阔混交林、针叶林、灌丛、草丛等植被及其他地物的分布情况,并进行了精度检验,结果表明MSI的总体分类精度略高于OLI。     

高分辨率遥感影像土地利用变化检测方法研究

提出一种利用高分辨率遥感影像进行土地利用变化检测的方法。以土地利用图为辅助数据,通过土地利用图和遥感影像的配准套合,获取影像像斑;同时,对遥感影像进行基于像素的监督分类,获取概略的类别图;再根据像斑内像素的类别编码完成子像斑的划分。以子像斑为影像分析的基本单位提取特征,以相关系数为相似性测度衡量不同时期子像斑的特征相似性,用ROC曲线(接受者操作特性曲线)代替经验选取的方法自动获取变化阈值,确定像斑是否发生变化。以武汉市区局部QuickBird 2002年和2005年多光谱影像、相同地区2002年1∶10 000土地利用图为实验数据进行了算法的实验,结果显示绝大部分的变化区域都可以被提取出来,实验方法可行。     

遥感图像质量等级分类的深度卷积神经网络方法

遥感图像应用发展对图像质量的要求越来越高,不同质量的遥感图像往往需要不同的处理方法和参数。通过遥感图像质量等级分类研究,不仅能够为遥感图像的处理提供先验信息,还能够对遥感图像的客观质量评价和传感器的成像效果进行评估。为了克服现有的遥感图像质量等级分类方法计算参数获取困难、等级数量少的缺点,利用深度学习方法的分类机能,通过改进特征提取网络和等级分类设计,建立了一种基于深度卷积神经网络的遥感图像质量等级分类模型。通过质量等级分类预处理后,利用经典的深度学习方法进行目标检测实验。结果表明,所提方法在西北工业大学遥感图像数据集上质量等级分类的准确率、召回率、精确率和F1最高能达到0.976、0.972、0.974和0.973, 优于传统算法。利用卷积神经网络实现遥感图像质量等级分类,既拓展了深度学习的应用领域,又为遥感图像质量评估提供了一个新方法。     

Satellite remote sensing of forest type and landcover in the subalpine forest region,Kananaskis Valley,Alberta

Landsat Thematic Mapper (TM) imagery and a digital elevation model (DEM) of the Kananaskis Valley in southwestern Alberta have been used to separate three forest types and eight landcover classes with mapping accuracies up to 76% overall. Image transformations based on a principal components analysis (PCA) were used to distinguish vegetation type and separate surface features in visual interpretations, and to reduce the 10 channel data set (TM 1–7, elevation, slope and incidence) to a more manageable 7 channel data set (PCA 1–4, elevation, slope and incidence). The DEM was shown to be critical in providing explanation of surface cover variability even though the original model was produced from medium scale aerial photography on a relatively coarse 100 metre grid. Discrimination increased up to 50% for pure stands of Lodgepole Pine (Pinus contorta Dougl.) and Englemann Spruce (Picea englemanii Parry) based on analysis of 100 pixels in test areas. Overall increases in map accuracy were between 2 and 11%. Success at this level of classification is required prior to detailed ecological study and modelling of mountain vegetation productivity at the community level using current satellite and aerial remote sensing technology.