基于FY-4A/AGRI时空特征融合的新疆地区积雪判识

高时间分辨率的积雪判识对于新疆牧区农牧业发展和雪灾预警具有重要作用,针对已有积雪产品易受复杂地形地貌,下垫面类型以及云遮蔽的影响,导致积雪判识精度降低的问题,提出一种利用深度学习方法对风云4号A星多通道辐射扫描计（AGRI）数据与地理信息数据进行多特征时序融合的积雪判识方法：以多时相FY-4A/AGRI多光谱遥感数据,以及高程、坡向、坡度和地表覆盖类型等地形地貌信息作为模型输入,以Landsat 8 OLI提取的高空间分辨率积雪覆盖图作为“真值”标签,构建并训练基于卷积神经网络的积雪判识模型,从而有效区分新疆复杂地形与下垫面地区的云、雪以及无雪地表,最终得到逐小时积雪覆盖范围产品。经数据集和2019年地面气象站实测雪盖验证,该方法精度高于国际主流MODIS逐日积雪产品MOD10A1和MYD10A1,显著降低云雪误判率。     

基于FY-MWRI的中国西部被动微波积雪判识算法北大核心CSCD

积雪是冰冻圈中分布最广泛的要素,在气候变化以及水文循环中扮演着重要角色。微波遥感因其全天时全天候工作、具有一定穿透性等优势,成为积雪监测的重要手段。利用FY-3C卫星同步观测获取的微波成像仪(MWRI)被动微波亮度温度数据、融合可见光红外扫描仪(VIRR)与中等分辨率成像光谱仪(MERSI)数据得到的积雪产品,结合MODIS地表分类数据、地表温度数据,发展了基于国产卫星数据的被动微波积雪判识算法。首先提取无云覆盖的不同地表类型被动微波数据像元样本,然后对各地表类型的微波特征进行分析,利用空间聚类的方法,得到TB19V-TB19H、TB19V-TB37V、TB22V、TB22V-TB89V、(TB22V-TB89V)—(TB19V-TB37V)这五类可以较好地区分积雪和其他类似积雪地表的指标。最后应用MODIS积雪产品为参考对该积雪判识算法进行精度评价,该算法在中国西部积雪判识总体精度为87.1%,漏判率为4.6%,误判率为23.3%;Grody算法判识总体精度为78.6%,漏判率为9.8%,误判率为30.7%,该算法判识精度高于Grody算法;通过Kappa系数分析比较,该算法积雪判识结果的Kappa系数值为47.3%,高于Grody算法判识结果的Kappa系数值39.9%,表明该算法积雪判识结果与MODIS积雪产品判识结果一致性更好。     

基于FY-MWRI的中国西部被动微波积雪判识算法

积雪是冰冻圈中分布最广泛的要素,在气候变化以及水文循环中扮演着重要角色。微波遥感因其全天时全天候工作、具有一定穿透性等优势,成为积雪监测的重要手段。利用FY-3C卫星同步观测获取的微波成像仪(MWRI)被动微波亮度温度数据、融合可见光红外扫描仪(VIRR)与中等分辨率成像光谱仪(MERSI)数据得到的积雪产品,结合MODIS地表分类数据、地表温度数据,发展了基于国产卫星数据的被动微波积雪判识算法。首先提取无云覆盖的不同地表类型被动微波数据像元样本,然后对各地表类型的微波特征进行分析,利用空间聚类的方法,得到TB19V-TB19H、TB19V-TB37V、TB22V、TB22V-TB89V、(TB22V-TB89V)—(TB19V-TB37V)这五类可以较好地区分积雪和其他类似积雪地表的指标。最后应用MODIS积雪产品为参考对该积雪判识算法进行精度评价,该算法在中国西部积雪判识总体精度为87.1%,漏判率为4.6%,误判率为23.3%;Grody算法判识总体精度为78.6%,漏判率为9.8%,误判率为30.7%,该算法判识精度高于Grody算法;通过Kappa系数分析比较,该算法积雪判识结果的Kappa系数值为47.3%,高于Grody算法判识结果的Kappa系数值39.9%,表明该算法积雪判识结果与MODIS积雪产品判识结果一致性更好。     

基于Himawari-8卫星的火点判识算法研究

利用新一代静止气象卫星Himawari-8数据，提出一种新的自适应阈值决策树低温火点判识方法。该方法基于2.3μm和0.86μm通道数据，以晴空像元和背景像元本地化判识结果为基础进行火点识别。选取山西省作为研究区域，利用2020年4月24日和2021年2月20日数据进行验证，结果表明：(1)对于森林火灾初期火势较小的火点可以尽早判识（采样点最早提前40 min）；(2)对草地、耕地上范围较小、温度较低的火点判识在时效性和准确率方面均表现良好；(3)新增的低温火点判识算法有效地解决了火点多判和漏判的矛盾问题，为尽早识别火点信息，实现有效的灾情监测提供了新思路。     

基于MODIS数据的白天多层云检测算法

分析了一种利用中分辨率成像光谱仪(MODIS)数据进行白天多层云检测算法。首先利用IMAPP软件包中的云检测算法对像元区域进行云检测,然后采用红外三通道(8.5μm、11μm和12μm)技术进行云相态判识,区分出单层水云和单层冰云,最后利用2.1μm反射率和11μm亮度温度双通道散点图,计算出多层云像元在2.1μm和11μm两个通道上的取值范围,从而识别多层云系。利用该算法对热带风暴云系进行了多层云检测试验,试验结果显示算法可简单有效地识别典型的多层云系。     

基于防突预测特征的地质异常智能判识方法

针对煤矿现有物探、钻探手段超前探测小型地质构造和煤层赋存变化等地质异常效果不好,以及防突预测数据隐含信息发掘不够、利用不足等问题,提出了根据防突预测特征与地质异常之间的相关性进行地质异常智能判识的思路;从单次防突预测事件数据分布和前后连续防突预测事件数据变化2个层面,构建了10个防突预测特征指标,形成了防突预测特征指标体系;采用关联分析方法,提出了基于防突预测特征的地质异常智能判识方法,并对特征指标二元属性转换、关联规则分析、有效规则提取、判识准则建立、地质异常可能性等级划分等关键环节进行了重点阐述;采用B/S架构,设计、开发了基于防突预测特征的地质异常智能判识系统,实现了防突预测信息在线采集、防突预测特征自动分析、地质异常超前动态判识,以及判识结果网站和移动终端等多渠道联动发布。现场试验结果表明,该系统能够自主构建地质异常判识准则,地质异常判识总准确率达到了87.63%,为煤矿超前掌握地质异常提供了有效手段,实现了防突预测数据隐含价值的拓展应用。     

FY-3/VIRR资料积雪多阈值综合判识方法研究

利用卫星遥感监测积雪分布相比地面观测具有明显优势,目前基于FY-3卫星数据在积雪监测方面的研究较少。借鉴现有积雪卫星遥感监测算法,研究出适用于FY-3/VIRR资料的积雪判识方法,利用归一化积雪指数和多波段综合阈值实现积雪判识,提取积雪信息生成区域二值化积雪分布图。通过实例分析验证算法有效可行,并与MODIS积雪产品MOD10及其L1B数据NDSI判识结果进行对比,说明算法判识结果良好。研究表明,FY-3卫星数据可作为积雪遥测的可靠资料来源,可延用于积雪监测与灾害预警业务系统中,促进国产卫星数据的应用与推广。     

风云三号卫星监测海洋浒苔、赤潮和溢油

风云卫星对海洋环境灾害（浒苔、赤潮和溢油）的定量遥感监测业务能力需要提升。利用风云三号D星中分辨率光谱成像仪MERSI-II资料以及部分高分卫星数据，探索解决大气校正和云检测两个技术难点问题的方法，分别依据浒苔、赤潮和溢油的光谱特征，建立相关的遥感监测算法，实现对海洋浒苔、赤潮和溢油的判识，并在相应个例中取得较好的监测效果。     

NOAA气象卫星云检测方法的研究

为了提高气象卫星资料的可用性，排除云干扰，本文根据云的时空分布变化特征，对可见光波段反射率和热红外波段温度进行分析和研究，提出了可见光反射率自动判云、热红外温度自动判云，可见光和热红外组合判云以及设立云区阈值判云等一系列检测方法。并对检测出的云区采用同周期相近时相的图像资料相对变化率来反演替代云区灰度值。保证了图像的连续性和客观性，取得了较好的效果。     

融合FY-3C号和FY-4A号卫星数据的积雪面积变化研究—以祁连山区为例

祁连山区积雪类型丰富、判识复杂,是中国积雪研究的典型区域。因此,精确地监测祁连山区积雪面积变化及其时空演变,对祁连山区生态环境和社会经济发展等具有重要意义。FY-3C MULSS利用多阈值积雪指数模型提供全球日积雪覆盖产品,FY-4A AGRI传感器每15~60 min提供一景覆盖全球的多光谱影像。基于FY-4A AGRI高时间分辨率的特征,构建适合于FY-4A号数据的动态多阈值多时相云隙间积雪识别方法,很大程度上减小了云对光学数据识别积雪造成的影响,并结合FY-3C MULSS积雪覆盖日产品较高空间分辨率的优势,融合得到去除云后的FY3C4积雪覆盖数据。利用Landsat 8 OLI卫星数据对融合后的积雪数据进行对比验证,结果表明融合FY-3C和FY-4A后的数据能更好地判识祁连山区的积雪覆盖情况。以MODIS MOD10A2积雪产品为真实值,随机检验了2018年3月~2019年3月融合后数据的积雪判识精度,发现无云情况下方法的总体精度可达到85.25%。进一步研究发现祁连山区积雪面积在海拔、气候和坡向等因素的影响下时空分布极不均匀,总体呈现出冬春季节大于夏秋季节,以及东部积雪面积大于西部积雪面积的特征。     

Statistical cloud detection from SEVIRI multispectral images

Cloud detection from geostationary satellite multispectral images through statistical methodologies is investigated. Discriminant analysis methods are considered to this purpose, endowed with a nonparametric density estimation and a linear transform into principal and independent components. The whole methodology is applied to the MSG-SEVIRI sensor through a set of test images covering the central and southern part of Europe. “Truth” data for the learning phase of discriminant analysis are taken from the cloud mask product MOD35 in correspondence of passages of MODIS close to the SEVIRI images. Performance of the discriminant analysis methods is estimated over sea/land, daytime/nighttime both when training and test datasets coincide and when they are different. Discriminant analysis shows very good performance in detecting clouds, especially over land. PCA and ICA are effective in improving detection.     

Cloud detection using infrared atmospheric sounding interferometer observations by logistic regression

Hyper-spectral infrared radiance data play an important role in cloud detection. To improve the cloud detection accuracy, this study proposes a novel cloud detection method based on the logistic regression model that uses the Infrared Atmospheric Sounding Interferometer (IASI) radiance data of four characteristic channels as the training features. Due to significant differences in the terrain between the land and the sea, the data from the oceans and continents are trained separately. Thereafter, the proposed scheme is verified and compared with existing methods. The results show that the accuracy of the proposed method (97% at sea and 88% on land) outperforms that of the existing Advanced Very High Resolution Radiometer (AVHRR)/IASI scheme (75% at sea and 55% on land). In addition, the proposed method uses only IASI observations as input and thus does not require the use of other auxiliary data.     

Object-based cloud and cloud shadow detection in Landsat imagery

A new method called Fmask (Function of mask) for cloud and cloud shadow detection in Landsat imagery is provided. Landsat Top of Atmosphere (TOA) reflectance and Brightness Temperature (BT) are used as inputs. Fmask first uses rules based on cloud physical properties to separate Potential Cloud Pixels (PCPs) and clear-sky pixels. Next, a normalized temperature probability, spectral variability probability, and brightness probability are combined to produce a probability mask for clouds over land and water separately. Then, the PCPs and the cloud probability mask are used together to derive the potential cloud layer. The darkening effect of the cloud shadows in the Near Infrared (NIR) Band is used to generate a potential shadow layer by applying the flood-fill transformation. Subsequently, 3D cloud objects are determined via segmentation of the potential cloud layer and assumption of a constant temperature lapse rate within each cloud object. The view angle of the satellite sensor and the illuminating angle are used to predict possible cloud shadow locations and select the one that has the maximum similarity with the potential cloud shadow mask. If the scene has snow, a snow mask is also produced. For a globally distributed set of reference data, the average Fmask overall cloud accuracy is as high as 96.4%. The goal is development of a cloud and cloud shadow detection algorithm suitable for routine usage with Landsat images.     

Discriminating sea ice from low-level water clouds in split-window,mid-wavelength IR imagery

A technique is demonstrated to enhance the contrast between sea ice and low-level water clouds. The approach uses the brightness temperature difference (BTD) feature from data collected in the split-window, mid-wavelength infrared (IR) region (i.e. two bands at 3.7 μm and 4.0 μm). These spectral data are available with Visible Infrared Imager Radiometer Suite (VIIRS) moderate-resolution bands M12 and M13, respectively. Under daytime conditions, the data collected in these bands contain energy that originates from both the sun and the Earth–atmosphere system. Due to the small wavelength difference between these, the terrestrial energy component in the bands is typically quite similar as are the surface reflectances for sea ice and ocean surfaces. Thus, the enhanced contrast between sea ice and water clouds, evident in a M12–M13 BTD image, results from differences in the solar energy, which decreases rapidly across this atmospheric window. Observed BTD values for water clouds can exceed 30K, while those for snow-ice fields are typically much smaller (e.g. 0–5K). Thus, water clouds appear bright in the image while sea ice, oceans, and most land surfaces are very dark. The enhanced contrast in the split-window, mid-wave IR BTD image makes it valuable for both image analysis and use in cloud algorithms. In addition, these images support the creation of manually generated cloud masks that have been shown useful for quantitatively evaluating the performance of automated cloud analysis algorithms and cloud forecast models. In this article, the value of 3.7 μm minus 4.0 μm BTD imagery for distinguishing between sea ice and low-level water clouds is shown using VIIRS data collected over the Beaufort Sea on 31 May 2012. Manually generated cloud masks, derived in part from these data, are then used to quantitatively evaluate the effectiveness of various cloud tests, including those used in the VIIRS cloud mask algorithm and the Moderate Resolution Imaging Spectroradiometer (MODIS) cloud mask algorithm. The results strongly suggest that split-window, mid-wavelength IR imagery provides valuable information to help differentiate between clouds and sea ice. It is concluded that collecting data in these mid-wavelength IR bands should be considered part of any future satellite sensor designed for environmental monitoring, especially over the polar regions.     

Night-time cloud detection for FY-3A/VIRR using multispectral thresholds

Night-time cloud detection using satellite data is a challenging area of research. This article presents a night-time cloud detection algorithm based on multispectral thresholds for the Visible and Infrared Radiometer (VIRR). VIRR is one of the keystone instruments on board the Chinese Feng Yun 3 (FY-3) polar-orbiting meteorological satellite. In this algorithm, three thermal infrared channels and other ancillary data are used to test for the presence of clouds according to different underlying surface types, and the four levels of possible cloud confidence are used to report whether a pixel is cloudy or clear. This algorithm strengthens the ability of identification of low cloud using the brightness temperature difference between the 3.7 and 12 μm channels. The comparisons of a new cloud mask with the official VIRR cloud mask product and with the official Moderate Resolution Imaging Spectroradiometer (MODIS) cloud mask product are shown to illustrate and validate the effect of this new algorithm. In addition, this algorithm is applied to FY-3B/VIRR data to test the validity and accuracy of cloud detection.     

Physical and statistical approaches for cloud identification using Meteosat Second Generation-Spinning Enhanced Visible and Infrared Imager Data

In this paper a cloud detection algorithm applied to the MSG-SEVIRI (Meteosat Second Generation-Spinning Enhanced Visible and Infrared Imager) data is described. In order to obtain a good performance in cloud detection, physical, statistical and temporal approaches have been used. In the statistical algorithm, the spectral and textural features of the MSG-SEVIRI images have been used as input, while, in the physical tests, a set of dynamic thresholds has been used. The physical algorithm does not use real time ancillary data— such as sea surface temperature map and NWP temperature and humidity profiles. A further test is applied to that pixels having low confidence to be clear or cloudy. This test takes advantage of the best MSG-SEVIRI temporal resolution and it applies the K-Nearest Neighbour classifier to the spectral and textural features calculated in “temporal” boxes 3 × 3 pixels, defined “temporal” because their elements belong to three subsequent MSG-SEVIRI images. The MACSP (cloud MAsk Coupling of Statistical and Physical methods) algorithm has been validated against the MODIS cloud mask and compared with CPR (Cloud Profiling Radar) and SAFNWC cloud masks. The outcomes show that the MACSP detects 91.8% of the total number of the pixels used for validation against MODIS cloud mask correctly, while the SAFNWC cloud mask detects 89.2% of them correctly. In particular, the MACSP classifies as cloudy 8.8% of the pixels classified by the MODIS cloud mask as clear, while the SAFNWC cloud mask classifies as cloudy 12.1% of them. The MACSP detects 91.2% of the cloudy CPR pixels and 90.8% of the cloud-free CPR pixels, considered for comparison, correctly. On the other hand, the SAFNWC and CPR cloud masks agree in the detection of 90.7% of the cloudy pixels and of 90.2% of the cloud-free pixels.     

A new Landsat 8 cloud discrimination algorithm using thresholding tests

ABSTRACT

In this work, we propose a Cloud Discrimination Algorithm for Landsat 8 (CDAL8) to improve a high-frequency automatic land change detection system developed at the National Institute of Advanced Industrial Science and Technology (AIST), Japan for large-scale satellite image analysis. Although the land change detection system can process several kinds of satellite remote sensing data, improvements are needed to enable practical applications using Landsat 8 data. Cloud discrimination is a necessary pre-processing step for land cover change detection. Currently, most of the prediction errors on land change detection are caused by the false cloud discrimination results as a pre-processing step. Therefore, we introduce an improved cloud discrimination algorithm (CDAL8) in this study to improve the overall performance of our land change detection system. The algorithm was developed based on a Moderate Resolution Imaging Spectroradiometer (MODIS) cloud mask algorithm and Cloud and Aerosol Unbiased Decision Intellectual Algorithm (CLAUDIA). CDAL8 is distinct in that it switches judgment tests and their thresholds using a threshold brightness temperature and uses separate features in cloud judgment and clear-sky judgment. To evaluate the accuracy of the proposed algorithm, we compared it with the Automated Cloud-Cover Assessment algorithm (ACCA) and Function of Mask (Fmask) version 3.3 using US Geological Survey Landsat 8 cloud cover assessment validation data, which contain 96 cloud masks. Our proposed cloud discrimination algorithm (CDAL8) have promising results with an accuracy of 88.1%, which was greater than that of the ACCA (82.5%) and Fmask (84.6%). Furthermore, we also confirmed that the average accuracy of CDAL8 was approximately 91.2% when low solar elevation scenes were removed.     

复杂场景下遥感船舶的检测与分割定位

遥感图像船舶识别是目标识别的一个重要领域,在海防和救援方面具有重大应用价值.但遥感图像中的船舶普遍存在云雾遮挡、陆地背景干扰和体积小等因素所造成的识别难的问题.为了能准确识别复杂场景下船舶目标,在网络的特征提取部分加入了视觉注意机制,增强网络提取船舶特征信息的能力;并采用多级特征提取和去量化操作的学习方法来解决船舶体积小的问题;采用难样本重学习的学习策略来弱化云雾遮挡和陆地背景的干扰.通过上述方法,船舶识别的综合准确率达到了92.56%,召回率达到了89.26%,与相同实验环境(PyTorch)下其他常见目标检测算法相比,精确率和召回率都有明显提升.实验结果表明,文中方法在一定程度上解决了复杂场景下船舶分割和识别难的问题.实验中所使用代码和部分结果详见https://github.com/curioyang/First_paper.     

激光雷达数据辅助的FY-3D影像阈值自适应云检测方法

国产风云系列卫星可为全球范围内大气、陆地和海洋的遥感监测提供重要数据支撑,由于光学卫星影像不可避免受到云覆盖的影响,通过云检测获取准确的云掩膜是风云系列卫星影像精细处理与应用的关键。现有的云检测方法大多采用简单高效的阈值法,然而由于传感器光谱响应以及不同场景云覆盖下垫面的辐射差异,在缺少大量真实云覆盖标记情况下,现有方法往往难以确定最优的检测阈值。鉴于此,提出了一种阈值自适应的云检测方法（TACD）,顾及传感器波段特性以及云覆盖下垫面差异,设置不同场景下的多通道阈值测试,包括反射率及反射率组合测试、亮度温度测试、亮度温度差异值测试、卷云测试等,联合具有高精度云层信息的激光雷达数据构建全球范围的云检测样本集,实现基于样本集真实云标记的迭代阈值优化,最终基于最优的阈值进行云检测。以风云三号（FY-3D）MERSI-II影像为例,联合CALIOP云层数据构建全球范围的云检测数据集,并将所提出的TACD方法云检测结果与官方云掩膜产品进行对比,结果表明该方法较官方云检测算法精度有明显提高,其中平均交并比从80.35%提升至84.09%,召回率可达92.67%,具有业务化应用的潜力。