基于主动在线极限学习机的卫星云量计算

发布时间：2018-09-14 09:23

【摘要】：云量不仅是影响地气系统辐射收支平衡的重要参数,同时也是研究大气环流及气候变化的重要指标。云量计算又与云检测息息相关,卫星云图分类方法的检测精度直接影响着云量计算的准确率。在卫星云图云检测处理的实际应用中,扩大训练集是提升分类精度途经之一。然而,大量的已标记数据集需要耗费大量的人力和物力成本。在遥感领域,现代高分辨率传感器技术的飞速发展,使得收集未标记数据变得更加容易和经济。因此,通过少量已标记训练样本和大量未标记样本提高算法的检测性能就显得很有意义。本文基于机器学习理论,将主动学习与极限学习机相结合,充分挖掘卫星云图分类中大量样本的有用信息,用以少量已标记样本,快速提高分类器的性能,提高检测的精度,减少人工标记成本。论文完成的主要工作如下:(1)研究极限学习机的样本不确定性评估策略,用于主动在线极限学习机,并通过与极限学习机、主动支持向量机以及主动极限学习机在4种不同公共数据下的性能表现,证明了所提出的主动在线极限学习机的有效性。(2)运用主动在线极限学习机进行云检测,对原始卫星云图进行样本提取、预处理后,已极限学习机作为基本分类器,采用非确定性抽样提取信息丰富的样本,进行主动在线学习,实现薄云、厚云、晴空以及薄云和厚云交界的检测。在不降低分类器性能的前提下,减少样本人工标注成本,缩减分类器训练时间。通过与阈值法、主动支持向量机、ELM实验比较,验证本文提出的方法在处理卫星云图数据时的有效性。(3)将检测后的卫星云图,利用“空间相关法”在云检测的基础上进行云量计算,并与4种不同的算法进行对比实验,最后通过与专家标定的标准数据库进行对比分析,改进并完善卫星云图云量计算模型。
[Abstract]:Cloud cover is not only an important parameter to influence the balance of radiation budget in terrestrial atmosphere system, but also an important index to study atmospheric circulation and climate change. Cloud calculation is closely related to cloud detection. The accuracy of satellite cloud image classification directly affects the accuracy of cloud calculation. In the practical application of cloud detection and processing of satellite cloud image, expanding the training set is one of the ways to improve the classification accuracy. However, a large number of marked data sets require a lot of human and material costs. In the field of remote sensing, the rapid development of modern high-resolution sensor technology makes it easier and more economical to collect unlabeled data. Therefore, it is significant to improve the detection performance of the algorithm through a small number of labeled training samples and a large number of unlabeled samples. Based on the theory of machine learning, this paper combines active learning with extreme learning machine to fully mine the useful information of a large number of samples in satellite cloud image classification, so as to quickly improve the performance of classifier and improve the accuracy of detection by using a small number of labeled samples. Reduce the cost of manual marking. The main work of this paper is as follows: (1) the sample uncertainty evaluation strategy of LLM is studied, which is used for active online LLM, and through LLM, The performance of active support vector machine and active extreme learning machine under four kinds of common data proves the effectiveness of the proposed active online extreme learning machine. (2) Cloud detection is carried out by using active online extreme learning machine. After preprocessing, the extreme learning machine is used as the basic classifier, the samples with abundant information are extracted by non-deterministic sampling, and active online learning is carried out to realize the thin cloud and thick cloud. Clear skies and the detection of the boundary between thin clouds and thick clouds. Without reducing the performance of classifier, the cost of manual labeling of samples is reduced, and the training time of classifier is reduced. Comparing with the threshold method and the ELM experiment of active support vector machine, the validity of the proposed method in the processing of satellite cloud image data is verified. (3) the detected satellite cloud image will be obtained. The spatial correlation method is used to calculate cloud amount on the basis of cloud detection, and compared with four different algorithms. Finally, it is compared and analyzed with the standard database calibrated by experts. The cloud volume calculation model of satellite cloud image is improved and improved.
【学位授予单位】：南京信息工程大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P412.27;TP18

【参考文献】