基于局部空间自相关分析的时空数据融合

由于受到16d重访周期与云等对数据质量的影响,具有时间与空间连续性的Landsat 8OLI观测数据难以直接获取。考虑地物分布的空间自相关性,提出一种基于STARFM模型改进的局部自相关时空数据融合模型(LASTARFM),以新疆维吾尔族自治区喀什地区叶城县为研究区,利用Landsat 8OLI数据和MODIS数据的红光波段和近红外波段进行融合方法测试。结果表明:利用LASTARFM模型得到的融合影像,与真实影像NDVI相关系数达到0.92;在局部空间自相关性低的区域比STARFM模型影像反映出更多地物细节,具有更高的融合精度;在土地利用类型发生显著变化的区域与真实影像具有一定差异。     

基于深度学习与超分辨率重建的遥感高时空融合方法

针对遥感影像的“时空矛盾”,提出一种改进STARFM的遥感高时空融合方法。利用SRCNN对低分辨率影像进行超分辨率重建,由于所融合的2组影像分辨率差距过大,网络训练困难,先将2组影像均采样至某一中间分辨率,使用高分辨率影像作为低分辨率影像的先验知识进行SRCNN重建,再将得到的中间分辨率影像重采样后以原始高分辨率影像作为先验知识进行第2次SRCNN重建,得到的最终重建图像相比原先使用插值法重采样所得图像,在PSNR和SSIM上均有提升,缓解了传感器差异所造成的系统误差。STARFM融合方法在筛选相似像元与计算权重时均使用专家知识提取人工特征,基于STARFM时空融合的基本思想,以SRCNN作为基本框架 自动提取特征,实验结果表明,其MSE值相比原方法更低,进一步提高了遥感时空融合的质量,有利于充分利用遥感影像。     

数据融合技术在提高NPP估算精度中的应用

针对现有遥感数据不能同时满足在时间和空间上精确监测植被动态变化的问题,提出利用时空适应性反射率融合模型（STARFM）的方法对MODIS-NDVI和TM-NDVI影像数据进行融合处理获得30 m较高时空分辨率的融合NDVI影像,进而将多种尺度的MODIS-NDVI和融合NDVI数据分别输入到CASA模型,对锡林浩特地区进行植被净初级生产力（NPP）的多尺度估算。将不同尺度的NPP估算结果与地上生物量地面实测值进行验证比较,结果表明：随着输入NDVI空间分辨率的提高,NPP估算值与实测地上生物量之间的相关性也逐渐增大,[r]最大值达到了0.915。此外以融合NDVI影像作为输入数据之一的NPP估算值与实测地上生物量的相关性均比未融合NDVI的相关性高,说明融合NDVI估算NPP的效果较未融合NDVI好,并且以融合NDVI影像作为模型输入数据可提高NPP估算精度。     

STARFM算法生成湿地类型TM反射率数据的应用评价

当前数据获取条件下,很难直接获得兼具高时间与高空间分辨率的多光谱遥感数据,提出利用STARFM(Spatial and Temporal Adaptive Reflection Fusion Model)算法来合成高时间序列的高空间分辨率数据。该算法在我国地理区域的适用性与预测精度验证等工作尚未充分展开。为此,以内蒙古呼伦湖湿地自然保护区为研究样区,并借助于Landsat-5 TM(Thematic Mapper)与高时序MODIS反射率产品,利用STARFM算法生成具有高时序特征的TM数据,进而将其与真实TM数据进行对比验证分析。结果表明:STARFM算法能够在空间上保持一定预测精度的条件下,对湿地区域内不同地物类别随时相的变化特征具有较好的预测能力,尤其适用于对反射特征随时相变化较小的湿地区域进行时空拟合或数据预测研究。     

基于遥感时空融合的升金湖湿地生态水文结构分析

利用遥感技术开展湖泊湿地生态水文结构分析对维持其生态服务功能具有重要意义,但受大气状况的影响会造成特定水位下可用高分辨率遥感影像的缺失,而遥感时空融合技术为弥补这一缺陷提供了重要途径。以安徽省升金湖湿地为研究区,根据改进后的时空自适应反射率融合模型（ESTARFM）模拟生成高时空分辨率遥感影像,评价模拟遥感影像的数值精度,进而分析了升金湖湿地的生态水文结构。结果表明： ① ESTARFM模型能够有效模拟不同水位下湖泊湿地的高分辨率遥感影像,融合影像与真实影像在近红外波段、短波红外波段反射率的相关系数分别达到0.93和0.91,且输入数据与融合数据的日期间隔越短,模拟精度越高;② 基于不同水体指数的水体提取效果表明,新型组合水体指数（NCWI）更适用于湖泊湿地的水体信息提取;③ 对升金湖湿地生态水文结构分析可知,湿地中心区、适宜活动区和非适宜区分别约占该湿地总面积的32.8%、12.1%和55.1%。     

基于多输入密集连接神经网络的遥感图像时空融合算法

为了解决地表反射率遥感卫星Landsat和MODIS影像的时空融合问题,文中提出基于多输入密集连接网络的遥感图像时空融合算法.首先提出多输入的密集连接网络,学习包含连续时刻间差异信息的过渡遥感影像.基于差异相似假设,融合网络学习得到的2幅过渡影像与已知的2幅高空间分辨率影像,得到最终的预测影像.对Landsat遥感影像和MODIS遥感影像的融合实验表明,文中算法在各项定量指标中均较优,最终的预测图像也可表明,文中算法对噪声具有较好的鲁棒性,能较好地恢复细节信息.     

面向云覆盖的遥感影像时空融合深度学习方法及其应用

遥感影像时空融合是一种获取高时空分辨率数据的有效手段,但现有方法在选定基础数据对时要求预测时间低分辨率数据无云覆盖影响,这极大地限制了其应用潜力。为此,提出一种面向云覆盖的遥感影像时空融合方法,即在深度学习框架下,构建重建子网络恢复预测时刻云下缺失信息,将重建后的低分辨率影像与前后相邻时刻高、低分辨率数据对构建时空融合子网络,得到最终的融合影像。以安徽淮南采煤沉陷区Landsat和MODIS反射率数据为例,对预测时刻MODIS数据模拟不同缺失率的云污染;利用所提方法进行时空融合实验,进而比较深度学习与传统方法融合数据对水体信息的提取效果。结果表明：该方法融合结果各波段的RMSE和SSIM均取得较好的定量评价效果,且总体优于传统方法;沉陷区水体提取实验表明本方法水体提取结果更加接近真实观测影像。因此,该方法降低了时空融合对数据的限制要求,且具有更高的融合精度和更有效的应用性。     

基于地表参数真实性的GF-1和SPOT-7多光谱影像质量评价

目的 遥感影像中地表信息表达真实程度决定了影像信息提取和定量化应用水平,传统的从像素灰度和视觉特性角度的影像质量评价方法难以评价影像对地表信息表达能力,本文从地表反射率和NDVI（normalized difference vegetation index）两种地表参数真实性角度评价GF-1和SPOT-7多光谱影像质量。方法 提出了一种基于地表参数真实性的多光谱影像质量评价方法,完成GF-1和SPOT-7卫星对实验区同步成像,地面同步测量大气光学特性和典型地物样区光谱,获取同步观测数据并对多光谱影像进行辐射误差处理,计算地物样区在影像上的反射率和NDVI,通过与地面实测光谱数据比较分析了地表参数真实性,评价GF-1和SPOT-7多光谱影像质量。结果 人工靶标中GF-1影像在4个波段反射率误差均在5%内,精度优于SPOT-7;植被地物中SPOT-7影像在蓝绿红波段反射率误差在4%内,近红外波段误差在15%内,NDVI误差在16%内,反射率和NDVI精度均优于GF-1;硬地地物中GF-1影像在4个波段反射率误差在6%内,精度优于SPOT-7;评价结果表明SPOT-7多光谱影像对植被类地物光谱表达真实度更高,GF-1对硬地类地物光谱表达真实度更高。结论 提出的基于地表参数真实性的遥感影像质量评价方法,能够有效地从地物光谱信息表达精度的角度评价影像质量。     

GF-5卫星高光谱数据大气校正反射率精度评价

反射率是高光谱遥感数据应用的基础,直接关系到高光谱应用效能和质量。目前,对国产GF-5卫星高光谱数据的精确大气校正反射率精度评价方法尚未有全面深入的研究,这严重制约了国产高光谱遥感数据的高质量应用。针对此问题,综合采用6S模型和FLAASH模块,选取了三个实验区的三种典型地物及外业光谱数据,采用三种定量化指标进行大气校正,得出了以下结论:三种地物大气校正反射率与实测反射率曲线特征基本一致,黑土地的大气校正反射率光谱最优,水体由于反射率数值较低,大气校正反射率光谱稍差;可见光近红外波段大气校正效果优于短波红外波段;6S模型大气校正结果略优于FLAASH模块,更适用于GF-5卫星高光谱影像。     

一种单时相高分辨率遥感影像时空融合算法

受制于传感器本身材料及卫星轨道参数,空间分辨率和时间分辨率是卫星遥感传感器固有的性能指标且难以兼备,使得高空间分辨率卫星的多时相数据合成问题至今仍是制约其广泛应用的关键问题之一。由于可有效综合空间-光谱-时间维的影像信息,多源遥感影像时空融合技术在近十年间得到迅速发展并已成为解决多时相数据合成问题的有力手段,其中基于学习的时空融合策略在合成精度上具有显著优势且应用潜力较高,但因其对字典训练过程的依赖程度较高而在融合过程中存在一定的不确定性。为提高基于学习的时空融合策略的预测精度、运算效率及鲁棒性,通过综合基于辐射归化的大气校正方法、基于误差约束的数据标准化转换机制、自适应多层递进融合策略以及高效的稀疏求解函数库,设计了一种适用于单时相高分辨率遥感影像的时空融合框架,并以国产高分二号卫星与Landsat-8卫星遥感影像为实验数据对该方法进行充分的对比性分析。实验结果表明,该融合框架不仅提升了运算效率,还在影像保真度、纹理特征描述以及光谱一致性等方面比当前的单数据对融合方法具有更好的重构质量。     

Blending MODIS and Landsat images for urban flood mapping

Satellite images provide important data sources for monitoring flood disasters. However, the trade-off between spatial and temporal resolutions of current satellite sensors limits their uses in urban flooding studies. This study applied and compared two data fusion models, the Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM) and Enhanced Spatial and Temporal Adaptive Reflectance Fusion Model (ESTARFM), in generating synthetic flooding images with improved temporal and spatial resolution for flood mapping. The synthetic images are produced in two scenarios: (1) for real-time prediction based on Landsat and MODIS images acquired before the investigated flooding; and (2) for post-disaster prediction based on images acquired after the flooding. The 2005 Hurricane Katrina in New Orleans was selected as a case study. The result shows that the Landsat-like images generated can be successfully applied in flood mapping. Particularly, ESTARFM surpasses STARFM in predicting surface reflectance in both real-time and post-flooding predictions. However, the flood mapping results from the Landsat-like images produced by both STARFM and ESTARFM are similar with overall accuracy around 0.9. Only for the flooding maps of real-time predictions does ESTARFM get a slightly higher overall accuracy than STARFM, indicating that the lower quality of the Landsat-like image generated by STARFM may not affect flood mapping accuracy, due to the marked contrast between land and water. This study suggests great potential of both STARFM and ESTARFM in urban flooding research. Blending multi-sources images could also support other disaster studies that require remotely sensed data with both high spatial and temporal resolution.     

A spatio-temporal fusion method for remote sensing data Using a linear injection model and local neighbourhood information

This paper presents a spatio-temporal fusion method for remote sensing images by using a linear injection model and local neighbourhood information. In this method, the linear injection model is first introduced to generate an initial fused image, the spatial details are extracted from the fine-resolution image at the base date, and are weighted by a proper injection gains. Then, the spatial details and the relative spectral information from the coarse-resolution images are blended to generate the fusion result. To further enhance its robustness to the noise, the local neighbourhood information, derived from the fine-resolution image and the fused result simultaneously, is introduced to refine the initial fused image to obtain a more accurate prediction result. The algorithm can effectively capture phenology change or land-cover-type change with minimum input data. Simulated data and two types of real satellite images with seasonal changes and land-cover-type changes are employed to test the performance of the proposed method. Compared with a spatial and temporal adaptive reflectance fusion model (STARFM) and a flexible spatio-temporal fusion algorithm (FSDAF), results show that the proposed approach improves the accuracy of fused images in phenology change area and effectively captures land-cover-type reflectance changes.     

线性模型的遥感图像时空融合

目的 时空融合是解决当前传感器无法兼顾遥感图像的空间分辨率和时间分辨率的有效方法。在只有一对精细-粗略图像作为先验的条件下，当前的时空融合算法在预测地物变化时并不能取得令人满意的结果。针对这个问题，本文提出一种基于线性模型的遥感图像时空融合算法。方法 使用线性关系表示图像间的时间模型，并假设时间模型与传感器无关。通过分析图像时间变化的客观规律，对模型进行全局和局部约束。此外引入一种多时相的相似像素搜寻策略，更灵活地选取相似像素，消除了传统算法存在的模块效应。结果 在两个数据集上与STARFM（spatial and temporal adaptive reflectance fusion model）算法和FSDAF（flexible spatiotemporal data fusion）算法进行比较，实验结果表明，在主要发生物候变化的第1个数据集，本文方法的相关系数CC（correlation coefficient）分别提升了0.25%和0.28%，峰值信噪比PSNR（peak signal-to-noise ratio）分别提升了0.153 1 dB和1.379 dB，均方根误差RMSE（root mean squared error）分别降低了0.05%和0.69%，结构相似性SSIM（structural similarity）分别提升了0.79%和2.3%。在发生剧烈地物变化的第2个数据集，本文方法的相关系数分别提升了6.64%和3.26%，峰值信噪比分别提升了2.086 0 dB和2.510 7 dB，均方根误差分别降低了1.45%和2.08%，结构相似性分别提升了11.76%和11.2%。结论 本文方法根据时间变化的特点，对时间模型进行优化，同时采用更加灵活的相似像素搜寻策略，收到了很好的效果，提升了融合结果的准确性。     

Spatio-temporal reflectance fusion via unmixing: accounting for both phenological and land-cover changes

Owing to technical limitations the acquisition of fine spatial resolution images (e.g. Landsat data) with frequent (e.g. daily) coverage remains a challenge. One approach is to generate frequent Landsat surface reflectances through blending with coarse spatial resolution images (e.g. Moderate Resolution Imaging Spectroradiometer, MODIS). Existing implementations for data blending, such as the Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM) and Enhanced STARFM (ESTARFM), have their shortcomings, particularly in predicting the surface reflectance characterized by land-cover-type changes. This article proposes a novel blending model, namely the Unmixing-based Spatio-Temporal Reflectance Fusion Model (U-STFM), to estimate the reflectance change trend without reference to the change type, i.e. phenological change (e.g. seasonal change in vegetation) or land-cover change (e.g. conversion of a vegetated area to a built-up area). It is based on homogeneous change regions (HCRs) that are delineated by segmenting the Landsat reflectance difference images. The proposed model was tested on both simulated and actual data sets featuring phenological and land-cover changes. It proved more capable of capturing both types of change compared to STARFM and ESTARFM. The improvement was particularly observed on those areas characterized by land-cover-type changes. This improved fusion algorithm will thereby open new avenues for the application of spatio-temporal reflectance fusion.     

An enhanced spatial and temporal adaptive reflectance fusion model for complex heterogeneous regions

Due to technical and budget limitations, remote sensing instruments trade spatial resolution and swath width. As a result not one sensor provides both high spatial resolution and high temporal resolution. However, the ability to monitor seasonal landscape changes at fine resolution is urgently needed for global change science. One approach is to “blend” the radiometry from daily, global data (e.g. MODIS, MERIS, SPOT-Vegetation) with data from high-resolution sensors with less frequent coverage (e.g. Landsat, CBERS, ResourceSat). Unfortunately, existing algorithms for blending multi-source data have some shortcomings, particularly in accurately predicting the surface reflectance of heterogeneous landscapes. This study has developed an enhanced spatial and temporal adaptive reflectance fusion model (ESTARFM) based on the existing STARFM algorithm, and has tested it with both simulated and actual satellite data. Results show that ESTARFM improves the accuracy of predicted fine-resolution reflectance, especially for heterogeneous landscapes, and preserves spatial details. Taking the NIR band as an example, for homogeneous regions the prediction of the ESTARFM is slightly better than the STARFM (average absolute difference [AAD] 0.0106 vs. 0.0129 reflectance units). But for a complex, heterogeneous landscape, the prediction accuracy of ESTARFM is improved even more compared with STARFM (AAD 0.0135 vs. 0.0194). This improved fusion algorithm will support new investigations into how global landscapes are changing across both seasonal and interannual timescales.     

Time and Space Fusion Model Comparison of Temperature Vegetation Drought Index

Drought is the first disaster affecting agricultural production. The annual precipitation in Xinjiang of China is scarce and the climate is dry. This is one of the major obstacles to the agricultural transformation and rural revitalization in Xinjiang. Therefore, timely and accurate monitoring of agricultural drought in Xinjiang is of great significance for safeguarding agricultural production. Yanqi Basin in Xinjiang was took as an example. Landsat8 and MODIS data were used. The Spatio Temporal Adaptive Reflectivity Fusion Model (STARFM), the Enhanced STARFM (Enhanced STARFM, ESTARFM) Model and Flexible Spatio Temporal Data Fusion (FSDAF) model were used to construct the Temperature Vegetation Dryness Index (TVDI). At the same time, the Relative Soil Moisture (RSM) was used to verify the TVDI inversion results. The results show that coefficient of determination (R²) and root mean square error (RMSE) of the drought factors(NDVI and surface temperature) simulated by the ESTARFM model were better than that by the other two models. And the R² and RMSE of NDVI simulated by the ESTARFM model reached 0.924 and 0.076. In addition, the R² and RMSE of surface temperature simulated by the ESTARFM model reached 0.877 and 2.799. Comparing with TVDI of the real Landsat8 data inversion and RSM data, it was found that the TVDI simulated by the ESTARFM model is better than the other two models, with 0.873 of R² and 0.248 of RMSE. The ESTARFM model can more accurately simulate the TVDI distribution of the Landsat8 images in the same period, so as to monitor the drought degree of the farmland in Xinjiang.     

温度植被干旱指数时空融合模型对比

为实时准确地对新疆农业干旱程度进行反演监测,以新疆焉耆盆地为例,通过运用时空自适应反射率融合模型（Spatio Temporal Adaptive Reflectivity Fusion Model,STARFM）、增强型STARFM（Enhanced STARFM,ESTARFM）模型及灵活的时空数据融合模型（Flexible Spatio Temporal Data Fusion,FSDAF）这3种常见的模型对Landsat 8和MODIS数据进行融合,构建了温度植被干旱指数（Temperature Vegetation Dryness Index,TVDI）,并采用土壤相对湿度（Relative Soil Moisture,RSM）数据对TVDI反演结果进行了验证。结果表明：①3种数据融合模型所模拟预测的干旱因子（归一化植被指数和地表温度）与真实Landsat 8数据所反演的干旱因子相比,ESTARFM模型模拟预测的干旱因子判定系数（R²）和均方根误差（RMSE）均优于其他两种模型,归一化植被指数（NDVI）的R²和RMSE分别达到了0.924和0.076,地表温度（LST）的R²和RMSE分别达到了0.877和2.799;②3种数据融合模型模拟预测的TVDI通过与真实Landsat 8数据反演的TVDI及RSM数据进行对比验证,发现ESTARFM模型模拟预测的TVDI与上述两种数据之间的R²也均优于其他两种模型,分别达到了0.873和0.248。ESTARFM模型在一定程度上更能准确地模拟预测同时期Landsat 8影像的TVDI分布状况。     

基于DSmT理论的多视角融合目标检测识别

现有无人车在目标检测中大多依靠单一检测视角进行多传感器数据融合,受传感器检测范围的局限,难以大幅提高准确率,且对融合过程中的类别判定的高冲突情况处理较少．针对以上问题,本文基于多假设思想提出了多视角检测结果的聚类合并方法,并基于DSmT（Dezert-Samarandache theory）和时序信息,改进了冲突分配准则,降低了目标检测的漏检率与误检率．首先利用图像检测算法检测图像中的有效目标,将激光雷达的目标检测结果投影在图像平面上,通过交并比关系构建2种传感器检测结果之间的关联概率矩阵,基于多假设思想实现聚类合并,获取单帧融合检测结果．针对融合过程中可能出现的类别判定冲突情况,利用DSmT融合识别置信度,并结合时序信息对冲突重新分配,获取目标类别的准确识别结果．最后,通过实车实验对算法的有效性进行了验证．