高光谱图像的分类技术研究

发布时间：2018-06-14 17:07

本文选题：高光谱图像 + 模式分类　；参考：《重庆大学》2014年博士论文

【摘要】：高光谱遥感是当前遥感技术发展的一个前沿领域，它利用很多很窄的电磁波波段从感兴趣的物体获得有用信息。高光谱图像作为遥感领域的一项重大突破，在保留较高空间分辨率同时，其光谱分辨率有极大的提高，达到了纳米的数量级，可以用来探测和识别传统全色和多光谱遥感中不可探测的地物类别。与传统的多光谱遥感图像相比，高光谱遥感图像有着信息量大、光谱分辨率高等特点，这使得在描述与区分地物类别方面的能力有了大幅提高，进而为地物光谱信息的精确处理与分析提供了可能。高光谱遥感系统已在全球许多国家的先进对地观察遥感系统中占有重要的位置，己成为地球陆地、海洋、大气观察的生力军。但是由于高光谱图像具有较高的数据维数，常规的图像分类方法在处理高光谱图像时有较大的限制，如何从大量的高光谱数据中快速而准确地挖掘出所需要的信息，，实现高精度的分类，仍是一个亟待解决的问题。本文从高光谱图像数据的特点入手，在对现有算法进行分析的基础上，针对高光谱遥感图像分类算法进行深入研究。主要的研究工作如下： ①在对高光谱遥感影像进行预处理之后，对所用高光谱图像做了大气校正。几何校正选取为二次多项式模型，重采样采用的是最近邻插值法，精度方面的要求得到了充分保证，为下一步的正确分类打下了坚实的基础。 ②提出了一种基于自适应粒子群优化算法的RBF神经网络高光谱遥感图像分类方法。由于人工神经网络具有并行处理、模糊识别和非线性映射等优点，很适合高光谱图像分类，但是其参数难选。采用自适应粒子群优化算法对RBF神经网络的参数进行了优化，建立了基于粒子群优化算法的的RBF神经网络模型，分类实验结果表明了基于粒子群优化的RBF神经网络模型具有很高的分类精度。 ③提出了一种基于自适应粒子群优化算法的SVR高光谱遥感图像分类方法。首先分析了支持向量回归的核函数的构造和模型参数的优选问题。由于本文数据样本较少，模型参数优选的比较复杂，本文采用了CV估计模型推广误差，并使用自适应粒子群优化算法来优选SVR模型参数，构建了基于粒子群优化算法的SVR高光谱遥感图像分类模型，在一定程度上解决了高光谱数据标记样本不足的问题。 ④从稀疏表示的基本理论出发提出了一种基于自适应稀疏表示的高光谱分类方法。利用训练样本构建字典，聚类每一步迭代所产生的余项，将聚类中心作为新的字典原子，然后将测试样本看成冗余字典中训练样本的线性组合，令字典能够更适应于样本的稀疏表示。通过对高光谱图像的分类实验，验证了自适应稀疏表示算法的有效性。
[Abstract]:Hyperspectral remote sensing is a frontier field in the development of remote sensing technology. It uses a lot of narrow electromagnetic wave bands to obtain useful information from objects of interest. As a major breakthrough in the field of remote sensing, the spectral resolution of hyperspectral images has been greatly improved, reaching the order of magnitude of nanometer, while retaining higher spatial resolution. It can be used to detect and identify undetectable features in traditional panchromatic and multispectral remote sensing. Compared with traditional multispectral remote sensing images, hyperspectral remote sensing images have the characteristics of large amount of information and high spectral resolution. It also provides the possibility for the accurate processing and analysis of the spectral information of ground objects. Hyperspectral remote sensing system has played an important role in the advanced earth observation remote sensing system in many countries all over the world, and has become a new force in the observation of the earth's land, ocean and atmosphere. However, because of the high data dimension of hyperspectral images, the conventional image classification methods have great limitations in processing hyperspectral images. How to quickly and accurately mine the needed information from a large number of hyperspectral data. The realization of high-precision classification is still a problem to be solved. Based on the characteristics of hyperspectral image data and the analysis of existing algorithms, the classification algorithm of hyperspectral remote sensing image is studied in this paper. The main research work is as follows: 1 after preprocessing the hyperspectral remote sensing image, the atmospheric correction of the hyperspectral image is done. The geometric correction is chosen as quadratic polynomial model, and the nearest neighbor interpolation method is used for resampling. It lays a solid foundation for correct classification in the next step. 2 A RBF neural network hyperspectral remote sensing image classification method based on adaptive particle swarm optimization algorithm is proposed. Because of the advantages of parallel processing, fuzzy recognition and nonlinear mapping, artificial neural network is suitable for hyperspectral image classification, but its parameters are difficult to select. The parameters of RBF neural network are optimized by adaptive particle swarm optimization algorithm, and the RBF neural network model based on particle swarm optimization algorithm is established. The classification experiment results show that the RBF neural network model based on particle swarm optimization has high classification accuracy. 3 A SVR hyperspectral remote sensing image classification method based on adaptive particle swarm optimization algorithm is proposed. Firstly, the construction of kernel function of support vector regression and the optimization of model parameters are analyzed. Because of the small number of data samples and the complexity of the optimal selection of the model parameters, the CV estimation model is used to extend the error, and the adaptive particle swarm optimization algorithm is used to optimize the SVR model parameters. The SVR hyperspectral remote sensing image classification model based on particle swarm optimization algorithm is constructed. To some extent, the problem of insufficient samples of hyperspectral data markers is solved. 4 based on the basic theory of sparse representation, a hyperspectral classification method based on adaptive sparse representation is proposed. Using training samples to construct dictionaries, the remainder of each iteration is clustered, the cluster center is regarded as a new dictionary atom, and the test samples are regarded as linear combinations of training samples in redundant dictionaries. The dictionary is more suitable for sparse representation of samples. The effectiveness of the adaptive sparse representation algorithm is verified by the classification experiments of hyperspectral images.
【学位授予单位】：重庆大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TP751

【参考文献】