基于纹理方向整体分布特征的台风自动识别方法

为实现卫星云图上台风的自动识别，提出了一种基于纹理方向整体分布特征的台风云系图象自动识别方法，通过引入矢量矩的概念来表现图象纹理整体分布规律，该识别方法采用全局搜索方式，将一窗口在整幅图象上滑动，首先计算出窗口图象内各点的纹理方向，进而得出窗口图像的矢量矩，将矢量矩与阈值比较来判整幅图象是否为台风云系图像，实验结果表明，该方法能够识别不同类型和不同发展阶段的台风云系图象，能够很好地将台风云系与其他干扰云系区分开，具有较广泛的适应性和较高的识别率。     

台风卫星云图分割方法研究

本文针对台风卫星云图的具体特点,采用一种利用迭代模型并结合台风云系面积特征的分割方法,对台风卫星云图进行分割,取得了满意的效果。     

台风卫星云图分割方法研究

本文针对台风卫星云图的具体特点,采用一种利用迭代模型并结合台风云系面积特征的分割方法,对台风卫星云图进行分割,取得了满意的效果。     

基于卫星云图的台风云系特征提取算法研究

利用静止卫星观测数据,进行了滤波、去噪等预处理,及综合优化的云检测处理;在此基础上,利用改进的正交匹配小波算法和优化的遗传-模糊聚类算法,实现了西北太平洋地区台风云系特征的提取。这两种算法提取出的台风云系特征参数简单,用于热带气旋(Tropical Cyclone,TC)的监测预警中可提高效率,适用于台风预报预警监测的业务化应用。     

基于纹理和区域特征的台风卫星云图分割方法

在利用GMS红外卫星云图进行无眼台风自动定位方法的研究中,台风云系的分割是处理中关键的一步,文章提出了一种基于纹理和区域特征的台风云系分割方法。首先利用图像的分形维数和灰度特征对台风云系中的密蔽云区进行有效的识别后,然后启动基于区域约束的区域生长计算得到台风云系。     

基于FY-2C气象卫星云图的台风分割方法的研究

在利用FY-2C气象卫星云图进行台风中心自动定位与跟踪的研究中,台风云系的分割是处理中关键的一步。结合台风主体云系具有灰度值较高、面积较大、活动范围有限和像素集中的特点,综合采用阈值法、数学形态学法和数理统计等方法,对FY-2C气象卫星云图中的红外一通道兰勃托原始投影云图中的台风分割进行了研究,取得了满意的效果。     

台风中心的旋转定位

台风在大气中绕着自己的中心急速旋转、同时又向前移动的空气涡旋。由于台风中心集中了台风绝大部分的能量和含水量,也是台风破坏力集中的地方,因为台风中心的定位对于台风路径预测和灾害预报来说都是非常重要的。传统的台风中心定位方式精度底,误差大,难以满足实际需要。为了准确地进行台风中心的定位,根据台风卫星云图的数字化特征和台风运动规律,提出了旋转定位方式。它的理论基础是,根据运动学和台风学原理,台风虽然是非刚性物体,但是其中心部分可相对看成是刚体的运动,因为热带气旋中心具有转动矢零的特点。这样,在卫星云图序列动画中,找出前后两次相应特征点的轨迹,即可算出其中的转动原点,也就是转动矢量为零的点。由于该方法结合了云图特性和运动因素,因而提高了精度和合理性。由此可见,对于台风这样的运动物体,结合其运动特征的中心定位法是台风中心定位的必然方向。     

面向卫星云图及深度学习的台风等级分类

台风是最严重的自然灾害之一,做好台风等级分类预测是防灾减灾的关键。针对现有卫星云图特征提取复杂、识别率低等问题,基于卷积神经网络框架,以北太平洋1978—2016年近1 000多个台风过程的卫星云图作为样本,提出改进深度学习模型Typhoon-CNNs。该框架采用循环卷积策略增强模型表征力,使用十折交叉验证引入信息熵、Dropout置零率以优化模型灵敏度及防止过拟合。通过800样本测试集对模型Typhoon-CNNs进行验证,实验结果表明,本文算法的分类精度达到92.5%,台风和超强台风2个等级的预测正确率达到99%,优于传统分类方法。最后对模型提取的特征图进一步分析,模型能够准确识别台风眼和螺旋云带,从而证明Typhoon-CNNs对台风等级分类的可行性。     

卫星云图的多通道FCM分割算法

卫星云图是对地球大气进行遥感测量的重要资料,云团的识别、跟踪及预测都需要对卫星云图进行图像分割和边缘提取。本文通过改进模糊C均值(FCM)算法对云图进行分割,改进后的算法显示出了效率高、分辨准确的特点。实验采用了2011年6月的RGB多通道卫星云图,通过对云图分层聚类,提取多通道聚类的相交区域等过程,实现了云团快速准确的分割。     

卫星云图有眼台风中心定位算法研究

本文分析和研究了单幅红外卫星云图台风定位问题,初步研究了有眼台风的模式识别,提出有眼区台风的中心定位算法.根据气象领域知识,眼区处在台风主体云系最大内切圆圆心附近,算法先提取出台风主体云系,应用数学形态学求取台风最大内切圆中心和半径,在此范围内寻找台风眼区.为了排除云洞和云缝的干扰,提出使用距离和灰度信息组成判别式求取台风眼区.实验证明具有较高的定位精度,能适合气象预报业务化的要求.     

Assessing tropical cyclone impacts using object-based moderate spatial resolution image analysis: a case study in Bangladesh

The environmental and societal impacts of tropical cyclones could be reduced using a range of management initiatives. Remote sensing can be a cost effective, accurate, and potential tool for mapping the multiple impacts caused by tropical cyclones using high-to-moderate spatial resolution (5–30 m) satellite imagery to provide data on the following essential parameters – evacuation, relief, and management of natural resources. This study developed and evaluated an approach for assessing the impacts of tropical cyclones through object-based image analysis and moderate spatial resolution imagery. Pre- and post-cyclone maps of artificial and natural features are required for assessing the overall impacts in the landscape that could be acquired by mapping specific land cover types. We used the object-based approach to map land-cover types in pre- and post-cyclone Satellite Pour l’Observation de la Terre (SPOT) 5 image data and the post-classification comparison technique to identify changes in the particular features in the landscape. Cyclone Sidr (2007) was used to test the applicability of this approach in Sarankhola Upazila in Bangladesh. The object-based approach provided accurate results for classifying features from pre- and post-cyclone satellite images with an overall accuracy of 95.43% and 93.27%, respectively. Mapped changes identified the extent, type, and form of cyclone induced impacts. Our results indicate that 63.15% of the study area was significantly affected by cyclone Sidr. The majority of mapped damage was found in vegetation, cropped lands, settlements, and infrastructure. The damage results were verified through the high spatial resolution satellite imagery, reports and pictures that were taken after the cyclone. The methods developed may be used in future to assess the multiple impacts caused by tropical cyclones in Bangladesh and other similar environments for the purposes of tropical cyclone disaster management.     

A tropical cyclone application for virtual globes

Within the past ten years, a wide variety of publicly available environmental satellite-based data have become available to users and gained popular exposure in meteorological applications. For example, the Naval Research Laboratory (NRL) has maintained a well accepted web-based tropical cyclone (TC) website (NRL TC-Web) with a diverse selection of environmental satellite imagery and products covering worldwide tropical cyclones extending back to 1997. The rapid development of virtual globe technologies provides for an effective framework to efficiently demonstrate meteorological and oceanographic concepts to not only specialized weather forecasters but also to students and the general public. With their emphasis upon geolocated data, virtual globes represent the next evolution beyond the traditional web browser by allowing one to define how, where, and when various data are displayed and dynamically updated.In this article, we describe a virtual globe implementation of the NRL TC-Web satellite data processing system. The resulting NRL Tropical Cyclones on Earth (TC-Earth) application is designed to exploit the capabilities of virtual globe technology to facilitate the display, animation, and layering of multiple environmental satellite imaging and sounding sensors for effective visualization of tropical cyclone evolution. As with the NRL TC-Web, the TC-Earth application is a dynamic, realtime application, driven by the locations of active and historical tropical cyclones. TC-Earth has a simple interface that is designed around a series of placemarks that follow the storm track history. The position coordinates along the storm track are used to map-register imagery and subset other types of information, allowing the user a wide range of freedom to choose data types, overlay combinations, and animations with a minimum number of clicks. TC-Earth enables the user to quickly select and navigate to the storm of interest from the multiple TCs active at anytime around the world or to peruse data from archived storms.     

Study of upper ocean parameters during passage of tropical cyclones over Indian seas

The primary energy supply for tropical cyclones is the upward latent heat fluxes that are directly related to Sea Surface Temperatures (SST). The strong winds induce a negative SST anomaly in the tropical cyclone wake. This is usually referred to as the ‘cold wake’. Many studies have suggested that the cold wake results in a significant reduction of upward latent heat fluxes that supply energy to the tropical cyclone, and hence provides a negative feedback on its intensity. The cold-wake feedback on the intensity of tropical cyclones is a strong motivation to understand the oceanic response to tropical cyclones. A recent study of re-examining the mechanisms controlling the cold wake has shown the importance of vertical Ekman pumping. Under the core of tropical cyclones (typically over a disk of 100 km radius), vertical pumping is responsible for a cooling of the entire water column, while surface heat fluxes and vertical mixing both contribute to the surface cooling during the cyclone passage. Farther away from the cyclone core, vertical mixing generally overwhelms the effect of vertical pumping, and also the effect of heat fluxes in the case of strong tropical cyclones. The study examines the Ekman pumping during the passage of few cyclonic storms in the Arabian Sea (AS) and Bay of Bengal (BOB) during last decade using different datasets such as Satellite (QuikScat) and reanalysis (ERA-Interim) data and calls up the value for satellite deliverables. Surface currents and subsurface temperature structure is examined further and for different Ekman pumping velocity (EPV), variable subsurface response is illustrated. The changes in ocean surface temperature, sea level and heat content obtained from the remote sensing data products, in the wake of cyclone are also reported and thus the usefulness of satellite products to study the coupled tropical system is demonstrated.     

Cloud/snow recognition for multispectral satellite imagery based on a multidimensional deep residual network

Cloud/snow recognition technology for multispectral satellite imagery plays an important role in resource investigation, natural disasters, and environmental pollution. Traditional feature based classification methods cannot make full use of the effective features and multispectral optical parameters of satellite imagery; the precision of cloud/snow recognition is not good enough. Although deep convolution neural network (CNN) can extract features effectively, it faces training gradient diffusion and model degradation, which lead to a low accuracy in classification. In order to solve this problem, an improved deep residual network with multidimensional input is proposed for the cloud/snow recognition. The multidimensional deep residual network (M-ResNet) can effectively extract the image features and spectral information of satellite imagery. The multispectral satellite imagery is divided into cloud/snow-free, cloud only, snow only and cloud/snow mixed using the proposed method. The experimental results of HuanJing-1A/1B (HJ-1A/1B) satellite imagery in China show that the M-ResNet performs a good distinction for the four kinds of images. The accuracy of the classification is higher than support vector machine (SVM), random forest, convolution neural networks, and multi-grained cascaded forest (GcForest).     

Mining geophysical parameters through decision-tree analysis to determine correlation with tropical cyclone development

Correlations between geophysical parameters and tropical cyclones are essential in understanding and predicting the formation of tropical cyclones. Previous studies show that sea surface temperature and vertical wind shear significantly influence the formation and frequent changes of tropical cyclones. This paper presents the utilization of a new approach, data mining, to discover the collective contributions to tropical cyclones from sea surface temperature, atmospheric water vapor, vertical wind shear, and zonal stretching deformation. A decision tree using the C4.5 algorithm was generated to illustrate the influence of geophysical parameters on the formation of tropical cyclone in weighted correlations. From the decision tree, we also induced decision rules to reveal the quantitative regularities and co-effects of [sea surface temperature, vertical wind shear], [atmospheric water vapor, vertical wind shear], [sea surface temperature, atmospheric water vapor, zonal stretching deformation], [sea surface temperature, vertical wind shear, atmospheric water vapor, zonal stretching deformation], and other combinations to tropical cyclone formation. The research improved previous findings in (1) preparing more precise criteria for future tropical cyclone prediction, and (2) applying data mining algorithms in studying tropical cyclones.     

基于稠密残差网络的多序列卫星图像去云

遥感影像中最常见的问题是云层污染,它会导致图像信息缺失,降低遥感数据的可用性。针对该问题,提出了一种基于稠密残差网络的多序列卫星图像去云算法。首先,该网络使用多序列的有云卫星图像作为输入,能为网络提供更多的时序特征信息,提升去云效果;其次,在网络中段使用稠密残差层,以保证卷积层之间最大程度地传递和使用特征信息,让生成的修复图像整体结构合理、边缘细节更加清晰;最后,使用像素上采样来增强空间信息的利用,提升修复效果。该方法在欧洲"哨兵-2"遥感卫星图像数据集上进行验证,峰值信噪比和结构相似度指标为27.59和0.854 0,两项指标均超过了该数据集的原处理方法STGAN,提升了遥感图像去云的效果。     

基于知识库的面向对象卫星云图分类

为了提高卫星云图分类精度和实时识别云类,基于云类知识库采用面向对象的分类方法对卫星云图进行分类。首先对2011年7~8月的FY\|3A/VIRR卫星云图进行预处理,从中裁截500个云样本,随机选取42%云样本作为训练样本,提取训练样本的光谱和纹理特征,基于ReliefF方法进行特征选择,采用反向传播神经网络进行训练构造分类器,利用剩余58%云样本进行网络测试,至此云类知识库构建完毕。然后对待解译的云图进行JSEG分割获取云对象,基于云类知识库已训练好的分类器实现面向对象的云图分类。试验结果表明:所设计的云图分类算法有效,分类结果与云分类产品数据基本达到一致。     

基于多类运动想象任务的EEG信号分类研究

针对多类运动想象脑电信号个体差异性强和分类正确率比较低的问题,提出了一种时-空-频域相结合的脑电信号分析方法:首先利用小波包对EEG原始信号进行分解,根据EEG信号的频域分布提取出运动想象脑电节律,通过“一对多”共空间模式(CSP)算法对不同运动想象任务的脑电节律进行空间滤波提取特征;然后将特征向量输入到“一对多”模式下的支持向量机(SVM)中,并利用判断决策函数值的方法对SVM的输出结果进行融合;最后通过引入时间窗对脑电信号进行时域滤波,消除运动想象开始和结束时脑电的波动,进一步提高信号信噪比和算法的分类效果;实验结果显示:在时间窗为2 s时,平均最大Kappa系数达到了0.72,比脑机接口竞赛第一名提高了0.15,验证了该算法能够有效减小脑电信号个体差异性影响,提高多类识别正确率。     

基于密集连接空洞卷积神经网络的青藏地区云雪图像分类

为了提高高纬度地区云雪卫星图像的识别准确率,提出了密集连接空洞卷积神经网络与空洞卷积相结合的方法进行云雪卫星图像识别研究。该方法首先采用常规卷积层对图像进行处理得到特征图,然后采用多个密集块和过渡层对特征图进行处理。其中,密集块中采用跨层连接的方式实现了网络中所用层的特征传递,使得大量云雪特征得到重用,同时减轻了训练过程中的梯度消失问题。密集块中的卷积核采用空洞卷积,在减少参数量的同时扩大局部感受野,对云雪的光谱信息进行特征提取。最后,该方法采用平均全局池化层与全连接层得到云雪图像的预测结果。实验结果表明,与其他机器学习方法相比,该方法能够提高卫星云雪图像的识别准确率,具有良好的泛化能力。     

卫星云图中热带风暴云区的三维绘制技术研究

实现热带风暴场景的模拟对于灾害预防、评估等工作具有十分重要的意义。改进了一种基于图像色彩和灰度值的算法,能够较好地从卫星云图中分割提取出热带风暴的云层区域,进而利用一定规则生成了云区体数据。针对生成体数据的特点,采用了一种改进的快速光线投射算法来完成热带风暴云区的三维绘制,最终实现了对于云区的任意角度实时观察和动态绘制。