地基云图观测、分析与识别方法研究
王钰
2019
页数178
学位类型博士
中文摘要


云是大气热力、动力和水汽循环过程的外部表现,在气候和地球的能量辐射收支平衡中发挥着相当重要的作用。如何准确、及时地获取云的信息,对于大气科学、大气环境监测、气候分析研究、气象预报、人工影响天气以及国民经济和军事等诸多领域都十分重要。地基云图观测是近年来迅速发展的一种云观测方式,它对云进行自下而上的观测,观测数据质量高,局部时空分辨率高,能有效弥补卫星云观测的不足,提高云观测的准确性和实效性。因此,对地基云图观测的研究一方面可以为地基云自动化观测的业务应用提供基础技术支撑,为卫星云图提供验证和融合的数据基础,提高云观测资料的全面性和数据质量;另一方面也为模式识别和人工智能研究中自然目标客观场景的图像理解开辟了一个新的研究领域,具有重要的理论意义和应用价值。

云的特性十分复杂,目前国内外对云观测的研究主要都是围绕云状(云类)、云量和云高展开。因此,本文采用模式识别和机器学习的手段,同时结合地基云图的特点,开展地基云图的云状识别(创新点1-3)、云量预测(创新点4)、以及云高融合(创新点5)的关键技术和方法研究。论文的主要工作和创新点归纳如下:

1. 提出了一种基于稳定局部二值模式(Local Binary Pattern,LBP)的地基可见光云图的云状识别方法

针对传统一致LBP描述子不能捕获地基可见光云纹理图像的全部主要模式,导致低的云状识别性能,充分考虑云图的特点:同一张云图中不同模式的出现频率以及不同云图中相同模式的出现频率变化都非常大,跨越几个数量级,提出了一种基于所有旋转不变LBP模式出现频率排序平均的稳定LBP特征抽取方法。该方法获取了地基云图的最频繁出现的LBP模式,使提出的特征更具分类判别性,同时对噪声和LBP模式出现频率的变化具有更好的鲁棒性。实验结果进一步验证了提出方法的性能统计意义上显著优于其它常用的五种LBP类方法的性能,与卷积神经网络(Convolutional Neural Network,CNN)方法的性能可比,但具有更小的计算开销。

2. 提出了一种地基红外和可见光云图多源融合的云图特征学习和云状识别方法

地基全天空可见光和红外云图是目前地基云图观测的两种主要方式,它们可以获取云的不同信息。然而,现有文献中对于可见光和红外云图的研究往往是独立进行,没有充分利用和整合不同云图中的有用信息。如果能融合这两种云图观测的信息互补,则能进一步提高地基云图云状识别的性能。因此,为进行红外和可见光云图的融合,首先,收集了一个具有相同时空分辨率的地基红外和可见光云图数据集。接着,通过考虑分别从红外和可见光云图抽取的LBP类模式的联合分布,提出了一种基于直方图融合特征的红外和可见光云图两观测联合编码策略,应用于收集的云图数据集的云状分类。这种编码策略能有效地捕获两种云图观测之间的相关性,获取更多具有类别判别性的信息。实验结果表明提出的多源数据融合方法的性能显著优于单一数据源的方法的性能。

3.  构造了一种基于正则化KL(Kullback-Leibler)距离的云图分辨率选择准则

在地基云图观测中,为了获取更多的云信息,常常需要拍摄高分辨率的云图像。但在图像的处理与分析中,高的分辨率就意味着高的计算开销,甚至不可接受。一种简单且广泛使用的方法是合适地放缩原始图像到一个低的分辨率图像上,然后在这个低分辨率图像上进行图像分析。这样,一个随之而来的问题即是:原始图像应该被压缩到什么样的程度,在这个压缩过程中是否引起了有用信息的丢失?我们的实验给出了答案,信息的损失是不可避免的,但是这个问题往往被忽略,原始图像被任意地压缩,失去了原来拍摄高分辨率图像的初衷。为此,构造了一种基于正则化KL距离的图像分辨率选择准则,为云图最优分辨率的选择提供指导和辅助。在三个公开的地基可见光云图的云状识别任务上验证了提出的分辨率选择准则的可行性和可靠性。

4. 提出了一种基于ARIMA(Autoregressive Integrated Moving Average)时间序列模型的短期云量预测方法

上述的云图分析中都是以独立的云图为研究对象的,但实际上地基云图是以一定的频率被连续拍摄获取的,因此基于云图得到的云量参数往往是一个云量时间序列,相邻的云量序列之间是时间相关的。然而,传统的分析中却忽略了这种相关性信息,为此,提出了一种基于ARIMA时间序列统计模型的短期云量预测方法,它可以有效地捕获云量序列之间的相关性信息,显著提高云量预测性能。在我们自己收集的包含8519张图像的数据集上验证了提出方法的优越性。

5. 提出了一种卫星、地基云雷达、和探空多源数据反演的云顶高融合方法

云顶高一般可以通过三种云观测方式获取,包括天基卫星、地基云雷达、以及空基探空气球,它们具有各自不同的优缺点。传统的云顶高研究往往是只进行不同观测方式的云顶高对照和一致性分析,没有考虑如何通过多种观测度量的融合来改进云顶高观测的可靠性,但是它对于实际的气象预报和自然灾害的预防具有重要的意义。针对这个问题,提出了一种基于贝叶斯决策理论的面向卫星、地基云雷达、和探空多源数据融合的云顶高计算方法。实验结果表明提出的融合策略显著改进了单一观测结果的精度和相关性。

英文摘要

Clouds are the external reflection of atmospheric dynamics, heat, and hydrological cycle. They play an extremely important role on climate and the Earth's energy budget balance. How to accurately and timely acquire the cloud information plays an important role in atmospheric science, atmospheric environment monitoring, climate research, weather forecast, weather modification, and national economy and military fields. The automatic ground-based cloud image observation is a recently developed cloud observation technology, which observes clouds from top to bottom. The obtained data have high quality and temporal-spatial resolution. Ground-based cloud observation can make up the disadvantages of satellite observation, and improve the accuracy and timeliness of cloud observation. Thus, this study may provide technical support for the automated ground-based cloud observation, provide data support for satellite remote sensing data verification and fusion, and improve the comprehensiveness and data quality of cloud observation data. Furthermore, it may develop a new research field for image understanding on objective scenes with natural objects. It has important theoretical significance and application value.

 

The characteristic of clouds is very complicated. Currently, the cloud observation mainly focuses on the study of cloud type, cloud coverage, and cloud height. Thus, this study performs the cloud type recognition, cloud coverage prediction, and cloud height integration by combining the pattern recognition and machine learning techniques and the characteristics of ground-based cloud images. The main contributions are summarized as follows:

 

1. A Ground-based Cloud Classification Method by Learning Stable Local Binary Patterns (LBPs)

 

The conventional uniform LBP approach cannot capture all the dominant patterns in ground-based visible cloud texture images, thereby resulting in low classification performance. By considering the characteristic of cloud images: not only for different patterns in a cloud image, but also for a pattern in different cloud images, the changes of the occurrence frequencies of the patterns are significant in orders of magnitude, a robust feature extraction method by learning stable LBPs is proposed based on the averaged ranks of the occurrence frequencies of all rotation invariant patterns defined in the LBPs of cloud images. The proposed method captures the most frequently occurred LBP patterns of ground-based cloud images, which makes the captured features more discriminative. The proposed descriptor is more robust to noise and the changes of the occurrence frequencies of LBP patterns. Experimental results demonstrate that the proposed method achieves significantly higher classification accuracy than the uniform LBP, local texture patterns (LTP), dominant LBP (DLBP), completed LBP (CLTP) and salient LBP (SaLBP) methods. And the performance of the proposed method is comparable with that of the popular deep convolutional neural network (DCNN) method, but with less computation complexity.

 

2. A Feature Learning and Cloud Type Recognition Method Based on the Multiple Source Data Fusion of Ground-based Infrared and Visible Light Images

 

Currently, ground-based cloud image observations generally include two ways of infrared and visible light. These two kinds of cloud image observations attain different useful information of clouds. However, they are only independently analyzed and simply compared in the current study. The useful information is not fully utilized and integrated. The classification performance could be improved if taking full advantage of the complementary information of these two observations. Thus, first, a database containing these two kinds of cloud images with same temporal resolution is released in this study. Then, a two-observation joint encoding strategy of LBP type features is proposed to implement cloud classification by encoding the joint distribution of LBP type patterns in different observations, which captures the correlation between two observations and more discriminative information. Experimental results based on this database show the significant superiority of the proposed method compared to the results based on the single observation.

 

3. A Selection Criterion for Resolution of Ground-based Cloud Images Based on the Regularized KL (Kullback-Leibler) Divergence

 

In ground-based cloud image observation, images with the highest possible resolution are captured to obtain sufficient information about clouds. However, when image processing and analysis is performed on the basis of the original images, a high-resolution probably means a high (or even more, unacceptable) computation cost. In practical application, a simple and commonly adopted method is to appropriately resize the original image to a version with a decreased resolution. An inevitable problem is what degree the original image should be resized and whether useful information is lost in this resizing operation. This study demonstrates that information loss is inevitable. However, this problem has been always neglected in previous literature and the original image is arbitrarily resized without any criterion. Thus, a selection criterion for the optimal resolution of ground-based cloud images based on the regularized KL divergence is constructed, which can provide a helpful guidance for the selection of the optimal resolution. Furthermore, experiments based on three ground-based remote sensing cloud image data sets with different original resolutions validate this criterion.

 

4. A Short-term Cloud Coverage Prediction Method Using the ARIMA (Autoregressive Integrated Moving Average) Time Series Model

 

The above-mentioned cloud image analysis is performed based on the single cloud image, however, the cloud-measuring devices on the ground actually take one image of the clouds every few minutes and collect a series of cloud images. The cloud coverages computed from the ground-based cloud images are a time series, and thus the neighbouring cloud coverages over continue time are correlated. However, in traditional prediction techniques, the correlation is always neglected. Thus, an ARIMA time series statistical model is used to predict the short-term cloud coverage, which can effectively capture the correlation between cloud coverages and improve the prediction performance. Experimental results on a collected time series database of cloud coverage show the superiority of the proposed technique.

 

5. An Integration Technique of Cloud Top Heights Retrieved from Satellite, Radiosonde, and Ground-based Cloud Radar Observations

 

Cloud top height is typically obtained via three observations, namely, satellite, radiosonde and ground-based radar, with their corresponding strengths and weaknesses. Traditionally, many studies have focused on independent comparison and consistency analysis of CTHs retrieved from different observations. The researches on how to improve the reliability of the CTH observation by integrating multiple cloud measurements are rare in the literature. However, it may be of great significance for practical meteorological forecast and disaster prevention. Thus, in this study, an integration technique of different CTHs retrieved from meteorological satellite, radiosonde and ground-based cloud radar observations is provided by using Bayesian decision theory. Experimental results show that the integration observations improve the accuracy and correlation of single observations.

关键词地基可见光云图 地基红外云图 云状识别 云量预测 云顶高融合 分辨率选择
语种中文
七大方向——子方向分类文字识别与文档分析
文献类型学位论文
条目标识符http://ir.ia.ac.cn/handle/173211/23873
专题复杂系统管理与控制国家重点实验室_影像分析与机器视觉
推荐引用方式
GB/T 7714
王钰. 地基云图观测、分析与识别方法研究[D]. 中科院自动化所. 中科院自动化所,2019.
条目包含的文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
Thesis.pdf(9472KB)学位论文 开放获取CC BY-NC-SA
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[王钰]的文章
百度学术
百度学术中相似的文章
[王钰]的文章
必应学术
必应学术中相似的文章
[王钰]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。