结合珠海一号高光谱影像和XGBoost算法的珠江口滨海湿地分类

由于湿地类别多样且结构复杂,湿地遥感分类工作极具挑战性。本文以珠江口滨海湿地为研究区,基于珠海一号高光谱影像获取的光谱特征、形状特征、纹理特征和指数特征构建优选特征集,采用极端梯度提升(XGBoost)算法和面向对象技术提取湿地类型和空间分布,并对比分析基于支持向量机(SVM)算法和随机森林(RF)算法的湿地分类结果。结果表明：(1)珠海一号高光谱影像能够有效应用于湿地分类,且光谱特征在湿地分类中发挥了重要作用;(2)使用的机器学习算法中XGBoost算法的湿地分类效果最佳,总体精度为87.2%,Kappa系数为0.84;(3)优选的影像特征能够保证更高的湿地类型识别精度,验证了特征筛选有助于提高分类效果。本文发展了一种基于珠海一号高光谱影像和集成学习的大区域湿地类型识别方法,可为湿地资源调查提供有效的技术参考,服务于湿地的保护与开发利用。     

基于改进矩匹配算法的珠海一号高光谱卫星影像相对辐射校正

珠海一号高光谱卫星影像部分波段存在着噪声干扰,严重影响了数据的质量与信息利用.本文分析了珠海一号高光谱卫星数据噪声的特点,在矩匹配算法的基础上,使用移动窗格对影像进行分段处理,并采用距离加权计算与去除极端值统计数据的方法提高参考值的准确度,实现了基于影像的相对辐射校正.实验结果表明,与传统矩匹配法和傅里叶滤波法相比,本文方法能有效去除遥感影像中存在的噪声,并能较好保留原影像的辐射特征.     

多级融合的珠海一号高光谱影像冬小麦分类

研究基于珠海一号高光谱影像的冬小麦识别提取技术,提出基于多级融合的多时相高分辨率高光谱冬小麦提取方法。本文从珠海一号高光谱影像入手,利用高分辨率影像改善高光谱影像空间分辨率,通过主成分分析降维、多种特征提取技术,大幅减少计算量的同时提高分类精度,Kappa系数提升0.05。针对融合影像是否有效、高程特征如何正确使用、U型语义分割网络（U-Net）和深度卷积语义分割网络（DeepLab）如何选择等问题,文中以4个实验对比说明,验证了该方法可以有效改善分类结果。     

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。     

基于珠海一号高光谱数据的农作物精细化分类研究

利用高光谱遥感影像识别精细化农作物类型,已经成为农业遥感的热点领域。本文以绥化市北林区内的水稻、玉米、大豆、烤烟等典型农作物为分类对象,以ZH-1高光谱数据为主要数据源,结合外业采集样本,探讨利用高光谱数据对精细化农作物类型信息提取的方法。针对高光谱数据的冗余问题,对其进行PAC降维处理,采用支持向量机分类方法对农作物种植结构进行提取。最后,采用混淆矩阵方法对分类结果进行验证,分类总体精度达到87.08%,Kappa系数为0.86。分类结果表明,高光谱数据在精细化农业分类研究上得到较好的应用。     

WorldView-2纹理的森林地上生物量反演

使用高空间分辨率卫星WorldView-2的多光谱遥感影像,构建植被指数和纹理因子等遥感因子与森林地上生物量的关系方程,并计算模型估测精度和均方根误差,探索高分辨率数据的光谱与纹理信息在温带森林地上生物量估测应用中的潜力。以黑龙江省凉水自然保护区温带天然林及天然次生林为研究对象,通过灰度共生矩阵(GLCM)、灰度差分向量(GLDV)及和差直方图(SADH)对高分辨率遥感影像进行纹理信息提取,并利用外业调查的74个样地地上生物量与遥感因子建立参数估计模型。提取的遥感因子包括6种植被指数(比值植被指数RVI、差值植被指数DVI、规一化植被指数NDVI、增强植被指数EVI、土壤调节植被指数SAVI和修正的土壤调节植被指数MSAVI)以及3类纹理因子(GLCM、GLDV和SADH)。为避免特征变量个数较多对估测模型造成过拟合,利用随机森林算法对提取的遥感因子进行特征选择,将最优的特征变量输入模型参与建模估测。采用支持向量回归(SVR)进行生物量建模及验证,结果显示选入模型的和差直方图均值(sadh_mean)、灰度共生矩阵方差(glcm_var)和差值植被指数(DVI)等遥感因子对森林地上生物量有较好的解释效果;植被指数+纹理因子组合的模型获得较精确的AGB估算结果(R2=0.85,RMSE=42.30 t/ha),单独使用植被指数的模型精度则较低(R~2=0.69,RMSE=61.13 t/ha)。     

联合星载高光谱影像和堆栈集成学习回归算法的红树林冠层叶绿素含量遥感反演

红树林是世界上生产力最高、价值最高的湿地生态系统之一。冠层叶绿素含量CCC（Canopy Chlorophyll Content）作为红树林重要的生物物理参量,是估算其生产力和评价其健康状况的重要指标。本文利用珠海一号高光谱卫星（OHS）影像与Sentinel-2A多光谱数据计算传统植被指数与组合植被指数并构建了高维数据集,综合利用正态分布检验、最大相关系数法与变量重要性评价进行数据降维和变量优选;分别基于单一线性回归算法、机器学习回归算法和堆栈集成学习回归算法构建了红树林CCC遥感反演模型,探明北部湾红树林CCC的最佳遥感反演模型,验证OHS高光谱影像与Sentinel-2A数据反演红树林CCC的精度差异,评估SNAP-SL2P算法反演红树林CCC的适用性。研究结果表明：（1）通过数据降维和变量选择处理,从高维度OHS数据集选取了8个特征变量,其中RSI_（12,17）、DSI_（12,18）和NDSI_（6,12）组合植被指数对红树林CCC反演精度的贡献率较高;（2）联合OHS数据和最优堆栈GBRT集成学习回归模型（Score=0.999,RMSE=0.963 μg/cm²）的训练精度优于最优RF机器学习回归模型（RMSE降低了7.531 μg/cm²）,明显优于最优Lasso线性回归模型（RMSE降低了19.383 μg/cm²）;（3）在最优堆栈集成学习回归模型下,OHS数据反演红树林CCC的精度（R²=0.761,RMSE=16.738 μg/cm²）高于Sentinel-2A影像（R²=0.615,RMSE=20.701 μg/cm²）;（4）联合OHS和Sentinel-2A数据的最优堆栈集成学习回归模型反演红树林CCC的精度都明显优于SNAP-SL2P算法（R²=0.356,RMSE=49.419 μg/cm²）。研究结果论证了正态分布检验、最大相关系数法和基于XGBoost的特征选择方法有效降低了高维数据集的维度,并得到了最优特征变量;OHS数据的最优堆栈GBRT集成学习回归模型训练精度最高,是估算红树林CCC的最优反演模型;OHS和Sentinel-2A数据都能有效反演红树林CCC（R²均大于0.61）,而OHS数据的估算精度更高（R²大于0.75）;SNAP-SL2P算法不能有效反演红树林CCC（R²小于0.4）,且对红树林CCC数值存在系统性低估。     

基于集成学习方法的中国近地面臭氧浓度时空分布

自2013年大气污染防治行动以来,PM_2.5、PM₁₀、SO₂、NO₂等空气污染物浓度都有不同程度下降,但臭氧污染仍有上升趋势,臭氧污染已成为制约中国空气质量持续改善的关键问题。地基站点可以提供空间上特定点的臭氧浓度,但无法获得近地面臭氧连续的空间分布。由于臭氧大量分布于平流层,遥感卫星反演的臭氧柱浓度产品仅能反映整层臭氧柱浓度,但整层臭氧柱浓度与近地面浓度无明显相关性,因而无法体现近地面臭氧浓度。本文综合地基监测数据、再分析资料、卫星产品,采用不同的模型方法,得到近地面臭氧浓度的时空分布,结果表明集成学习方法可以准确估算近地面臭氧在空间上的分布状况和在时间上的变化趋势。本文对比了梯度提升回归树(GBRT)、极端随机树(ERT)、极端梯度提升器(XGBoost) 3种不同的集成学习方法在近地面臭氧污染估算的效果表现,3种集成学习方法在2019年—2020两年的十折交叉验证R2都在0.89以上,极端梯度提升器(XGBoost)方法在RMSE、MAE指标上有最好的表现,2019年—2020年两年的平均R...     

利用Sentinel-2影像超分辨率重建的红树林冠层氮含量反演

氮素是植被整个生命周期的必要元素,红树林冠层氮素含量（CNC）遥感估算对红树林健康监测具有重要意义。以广东湛江高桥红树林保护区为研究区,本文旨在基于Sentinel-2影像超分辨率重建技术进行红树林CNC估算和空间制图。研究首先基于三次卷积重采样、Sen2Res和SupReMe算法实现Sentinel-2影像从20 m分辨率到10 m的重建;然后以重建后的影像和原始20 m影像为数据源构建40个相关植被指数,采用递归特征消除法（SVM-RFE）确定CNC估算的最优变量组合,进而构建CNC反演的核岭回归（KRR）模型;最后选取最优模型实现CNC制图。研究结果表明：基于Sen2Res和SupReMe超分辨率算法的重建影像不仅与原始影像具有很高的光谱一致性,且明显提高了影像的清晰度和空间细节。红树林CNC反演波段主要集中在红（B4）、红边（B5）、近红外波段（B8a）以及短波红外波段（B11和B12）,与“红边波段”相关的植被指数（RSSI和TCARIre1/OSAVI）也是红树林CNC反演的有效变量。基于3种方法重建后10 m的影像构建的模型反演精度（R²val>0.579）均优于原始20 m的影像（R²val=0.504）;基于Sen2Res算法重建影像构建的反演模型拟合精度（R²val=0.630,RMSE_val=5.133,RE_val=0.179）与基于三次卷积重采样重建影像的模型拟合精度（R²val=0.640,RMSE_val=5.064,RE_val=0.179）基本相当,前者模型验证精度（R²cv=0.497,RMSE_cv=5.985,RE_cv=0.214）较高且模型变量选择数量最为合理。综合重建影像光谱细节及模型精度,基于Sen2Res算法重建的Sentinel-2影像在红树林CNC估算中具有良好的应用潜力,能为区域尺度红树林冠层健康状况的精细监测提供有效的方法借鉴和数据支撑。     

基于高分一号宽幅多光谱影像的鄱阳湖湿地分类研究

典型湿地类型分类对于湿地生态环境保护和生态环境建设具有重要意义。本研究以高分一号宽幅多光谱影像数据作为数据源,以鄱阳湖典型湿地作为研究对象,进行典型湿地类型分类。研究采用主成分分析、归一化植被指数、比值植被指数和归一化差异水体指数等方法对不同湿地类型进行光谱特征分析,结果表明:高分一号宽幅多光谱影像反演的归一化植被指数和归一化水体指数能够较好地区分水体、泥滩、挺水植物和湿生植物等常见湿地类型。通过归一化植被指数将湿地类型分为植被和非植被,归一化差异水体指数将植被类型进一步分为挺水植物和湿生植物,将非植被分为水体和泥滩。构建决策树对典型湿地类型进行自动分类,经过精度评价和分析验证,该方法针对高分一号宽幅数据进行典型湿地分类总体精度能够达到82.28%,Kappa系数0.7346,优于常规的监督分类和非监督分类。     

无人机遥感与XGBoost的红树林物种分类

无人机遥感数据会衍生大量的光谱、纹理与结构特征,如何提取优势特征是提高红树林物种分类效率和精度的关键问题.针对深圳福田红树林自然保护区缓冲区获取的无人机高光谱影像和LiDAR点云数据,本研究旨在利用极端梯度提升算法(XGBoost)的“特征重要性”属性筛选出适合红树林物种分类的8类优势特征:基于无人机高光谱影像的单一特...     

WorldView-2影像与OLI影像协同岩性模糊分类

Landsat系列、ASTER等中等空间分辨率遥感数据（中分数据）覆盖了碳酸盐矿物、粘土矿物、铁氧化物矿物等矿物的诊断光谱区间,广泛应用于矿物、岩石信息提取,但受限于空间分辨率,混合像元现象明显,严重制约了其岩性分类精度。WorldView-2、QuickBird等高空间分辨率遥感数据（高分数据）提供了岩石地层表面丰富的空间结构信息,同时空间分辨率的提高也是缓解混合像元效应的最有效途径,但高分数据覆盖的光谱区间往往较窄,难以满足大多数特征吸收谱段位于短波红外、热红外区间的矿物、岩石信息提取。在岩性自动分类方法上,前人研究中仍以采用基于像元的分类方法为主,分类结果的“椒盐现象（Salt-and-pepper,出现在分类结果图中大量孤立的错分点或小图斑）”严重。为结合中分数据的光谱信息优势和高分数据的空间结构信息优势,同时减少基于像元的岩性分类方法中的“椒盐现象”,提高岩性自动分类精度,本文以Landsat 8 OLI数据和WorldView-2数据为例,提出了一种协同中、高分遥感数据进行面向对象的岩性模糊分类的方法。首先通过“结构协同”和“光谱协同”方案对WorldView-2数据和OLI数据进行信息协同,利用主成分变换对协同后数据的纹理信息和光谱信息进行压缩和增强,然后将增强后的纹理信息和光谱信息进行波段绑定,并进行多尺度分割。根据岩性单元间的光谱特征和纹理特征的差异,构建各岩性单元的模糊逻辑隶属度函数,实现对研究区岩性的模糊分类。实验结果表明,该方法成功划分了岩性单元的分布,总体岩性分类精度为89.35％。     

A rule-based parameter aided with object-based classification approach for extraction of building and roads from WorldView-2 images

Roads and buildings constitute a significant proportion of urban areas. Considerable amount of research has been done on the road and building extraction from remotely sensed imagery. However, a few of them have been concentrating on using only spectral information. This study presents a comparison between three object-based models for urban features’ classification, specifically roads and buildings, from WorldView-2 satellite imagery. The three applied algorithms are support vector machines (SVMs), nearest neighbour (NN) and proposed rule-based system. The results indicated that the proposed rules in this study, despite the spectral complexity of land cover types, performed a satisfactory output with an overall accuracy of 92.92%. The advantages offered by the proposed rules were not provided by other two applied algorithms and it revealed the highest accuracy compared to SVM and NN. The overall accuracy for SVM was 76.76%, which is almost similar to the result achieved by NN (77.3%).     

Improving classification accuracy of spectrally similar land covers in the rangeland and plateau areas with a combination of WorldView-2 and UAV images

This study aims to increase the accuracy of the object based classification approach to differentiate the spectrally similar land cover types to create thematic maps depicting the current land use status in rangeland. Firstly, the multispectral and panchromatic bands of a WorldView-2 MS and Pan images are fused. The fused WV-2 image is then classified with object based approach using Support Vector Machines (SVMs) classifier (Method 1). The overall classification accuracy for Method 1 is found to be 88.6%. Secondly, UAV ortho-image is utilised for segmentation process, which is required for the object based SVM classification of the WV-2 MS image (Method 2). The overall classification accuracy for Method 2 is obtained as 92.4%. It is realised that the Method 2 increases the object based classification accuracy by 4%, compared to Method 1. This result reveals that the object based classification of the UAV and WV-2 MS images makes significant contribution to the classification accuracy.     

A comparative analysis of pansharpening techniques on QuickBird and WorldView-3 images

Nowadays, different image pansharpening methods are available, which combine the strengths of different satellite images that have different spectral and spatial resolutions. These different image fusion methods, however, add spectral and spatial distortions to the resultant images depending on the required context. Therefore, a careful selection of the fusion method is required. Simultaneously, it is also essential that the fusion technique should be efficient to cope with the large data. In this paper, we investigated how different pansharpening algorithms perform, when applied to very high-resolution WorldView-3 and QuickBird satellite images effectively and efficiently. We compared these 27 pansharpening techniques in terms of quantitative analysis, visual inspection and computational complexity, which has not previously been formally tested. In addition, 12 different image quality metrics available in literature are used for quantitative analysis purpose.     

The use of WorldView-2 satellite imagery to model urban drainage system with low impact development (LID) Techniques

Within a wide range of best management practices for stormwater management in urban areas, there has been an increasing interest in source control measures. Source controls such as low-impact development (LID) techniques are potentially attractive as retrofit options for older developed areas that lack available land to implement conventional measures such as stormwater management ponds. Hence, distributed urban drainage models requiring detailed representation of developed drainage areas should be developed to accurately estimate the benefits that LIDs may provide. This study (1) presents a two-stage classification process on a high-resolution WorldView-2 image, and (2) demonstrates how to use the extracted land cover information in the subsequent hydrologic modelling and assessment of different LIDs’ performance. The proposed two-stage classification method achieved an overall accuracy of 80.6%, whereas a traditional pixel-based achieved 68.4% in classifying the same urban area into six land cover classes. From the classification results, the hydrologic properties of micro-subcatchments were imported in the United States Environmental Protection Agency Storm Water Management Model to assess the performance of LIDs. A reduction of run-off volume 18.2% and 37.1% was found with the implementation of porous pavement and bioretention, respectively, in a typical low-rise residential area located in the city of San Clemente, California, US. The study demonstrates the use of high-resolution remote sensing image to aid in evaluating LID retrofit options, and thus benefits in situations where detailed drainage area information is not available.     

Assessing the utility WorldView-2 imagery for tree species mapping in South African subtropical humid forest and the conservation implications: Dukuduku forest patch as case study

Indigenous forest biome in South Africa is highly fragmented into patches of various sizes (most patches < 1 km²). The utilization of timber and non-timber resources by poor rural communities living around protected forest patches produce subtle changes in the forest canopy which can be hardly detected on a timely manner using traditional field surveys. The aims of this study were to assess: (i) the utility of very high resolution (VHR) remote sensing imagery (WorldView-2, 0.5–2 m spatial resolution) for mapping tree species and canopy gaps in one of the protected subtropical coastal forests in South Africa (the Dukuduku forest patch (ca.3200 ha) located in the province of KwaZulu-Natal) and (ii) the implications of the map products to forest conservation. Three dominant canopy tree species namely, Albizia adianthifolia, Strychnos spp. and Acacia spp., and canopy gap types including bushes (grass/shrubby), bare soil and burnt patches were accurately mapped (overall accuracy = 89.3 ± 2.1%) using WorldView-2 image and support vector machine classifier. The maps revealed subtle forest disturbances such as bush encroachment and edge effects resulting from forest fragmentation by roads and a power-line. In two stakeholders’ workshops organised to assess the implications of the map products to conservation, participants generally agreed amongst others implications that the VHR maps provide valuable information that could be used for implementing and monitoring the effects of rehabilitation measures. The use of VHR imagery is recommended for timely inventorying and monitoring of the small and fragile patches of subtropical forests in Southern Africa.     

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。     

Integration of WorldView-2 and airborne LiDAR data for tree species level carbon stock mapping in Kayar Khola watershed,Nepal

Integration of WorldView-2 satellite image with small footprint airborne LiDAR data for estimation of tree carbon at species level has been investigated in tropical forests of Nepal. This research aims to quantify and map carbon stock for dominant tree species in Chitwan district of central Nepal. Object based image analysis and supervised nearest neighbor classification methods were deployed for tree canopy retrieval and species level classification respectively. Initially, six dominant tree species (Shorea robusta, Schima wallichii, Lagerstroemia parviflora, Terminalia tomentosa, Mallotus philippinensis and Semecarpus anacardium) were able to be identified and mapped through image classification. The result showed a 76% accuracy of segmentation and 1970.99 as best average separability. Tree canopy height model (CHM) was extracted based on LiDAR’s first and last return from an entire study area. On average, a significant correlation coefficient (r) between canopy projection area (CPA) and carbon; height and carbon; and CPA and height were obtained as 0.73, 0.76 and 0.63, respectively for correctly detected trees. Carbon stock model validation results showed regression models being able to explain up to 94%, 78%, 76%, 84% and 78% of variations in carbon estimation for the following tree species: S. robusta, L. parviflora, T. tomentosa, S. wallichii and others (combination of rest tree species).