基于机器学习的遥感图像分类研究

发布时间：2018-04-24 03:08

本文选题：遥感影像 + 混合核函数　；参考：《江西理工大学》2017年硕士论文

【摘要】：随着航空航天科技和传感器科技的快速更新换代,遥感影像的数据来源变得多样化,数据集也日趋复杂,地物复杂难辨,如何准确、高效地进行遥感图像分类成为了近年来研究的重要内容。由于人工智能科技发展迅速,机器学习分类方法也逐渐成为一种有效的遥感图像分类处理方法,为了有效提高影像的分类精度,本文在机器学习的理论上构建了一种高效简单的分类模型以及一种两级分类模型。论文的主要工作与创新点如下:(1)系统地介绍了遥感技术发展的研究现状,简要阐述监督分类与非监督分类常用的一些方法以及现今前沿的分类器;扼要地总结了本文的研究内容、及其组织结构流程。(2)针对遥感影像在获取过程中易受大气吸收与散射、传感器定标、地形等因素的影响而造成图像失真的特点,本文利用二次多项式模型进行遥感影像几何校正,采取双线性内插法进行重采样等技术对影像进行校正预处理,并进行了大气校正,有效地去除了传感器等因素的畸变影响,同时也排除了大气中散射颗粒的影响,为后续分类奠定基础。(3)神经网络模型具有容错性、学习能力强等特点,但要得到较好的分类效果耗时非常大,而极限学习机分类器是一种结构简单的神经网络方法,能快速高效的对样本进行识别。本文构建了一种混合核极限学习机遥感图像分类模型,该模型利用混合核函数的全局与局部特性,结合遥感图像的邻域信息,有效地提高了分类精度。(4)基于遥感影像的光谱和空间信息提出了一种两级分类器的方法。结合光谱信息与空间结构信息,首先采用光谱角匹配方法作为前级分类器,提取影像中光谱信息特征明显且区别较大的地物;然后利用遥感数据的张量空间结构信息,选取支持张量机作为后级分类器。对选取的感兴趣区域进行分类,不仅提高了分类精度,而且分类视觉效果也有了明显改善。
[Abstract]:With the rapid upgrading of aerospace and sensor technologies, the data sources of remote sensing images become more and more diverse, the data sets become more and more complex, the complexity of ground objects is difficult to distinguish, how to be accurate, Efficient classification of remote sensing images has become an important research content in recent years. Due to the rapid development of artificial intelligence technology, machine learning classification method has gradually become an effective remote sensing image classification processing method, in order to effectively improve the image classification accuracy, In this paper, an efficient and simple classification model and a two-level classification model are constructed in theory of machine learning. The main work and innovation of this paper are as follows: (1) the research status of remote sensing technology is introduced systematically, and some common methods of supervised classification and unsupervised classification are briefly described, as well as the current frontier classifiers. This paper briefly summarizes the research contents of this paper, and its organization and structure. 2) aiming at the characteristics of image distortion caused by the influence of atmospheric absorption and scattering, sensor calibration, topography and so on, in the process of remote sensing image acquisition. In this paper, the quadratic polynomial model is used for the geometric correction of remote sensing image, the bilinear interpolation method is used to resample the image and the atmospheric correction is carried out, which effectively removes the distortion effect of the sensor and other factors. At the same time, the influence of scattering particles in the atmosphere is excluded, which lays the foundation for the following classification. The neural network model has the characteristics of fault tolerance and strong learning ability, but it takes a lot of time to obtain better classification effect. The extreme learning machine classifier is a simple neural network method, which can identify samples quickly and efficiently. In this paper, a hybrid kernel extreme learning machine remote sensing image classification model is constructed. The model combines the global and local characteristics of the hybrid kernel function and the neighborhood information of the remote sensing image. A two-level classifier based on spectral and spatial information of remote sensing image is proposed. Combining the spectral information with the spatial structure information, the spectral angle matching method is first used as the front classifier to extract the features of the spectral information in the image, and then the spatial structure information of Zhang Liang from the remote sensing data is used. Zhang Liang machine is selected as the posterior classifier. The classification of selected regions of interest not only improves the classification accuracy, but also improves the visual effect of classification.
【学位授予单位】：江西理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP751;TP181

【参考文献】