辅以纹理的遥感图像分类研究与应用

发布时间：2018-10-09 08:16

【摘要】：遥感图像分类是人类获取有效信息的重要手段,如何改善分类精度是遥感研究的重要内容。传统的图像分类大多基于光谱特征,而对其空间结构特征挖掘不足,造成分类结果不够理想。图像的纹理作为最具代表性的空间结构特征,在改善图像分类精度方面有其独特的优势和巨大的应用价值。纹理分析方法众多,传统的统计分析方法成熟稳定,有着广泛的应用基础,新兴的模型分析方法发展迅速,特别是分形模型在纹理分析中的应用受到极大的关注。本文选取福州市区中南部城乡交界处的一块矩形区域作为典型试验区,利用传统的灰度共生矩阵模型提取典型试验区图像的4个纹理特征,利用新兴的分形模型提取典型试验区图像的1个纹理特征。在此基础上,将1个分形纹理、4个灰度共生矩阵纹理与8个光谱特征等13个特征变量组合,构建多源特征数据库,开展辅以纹理的典型试验区图像监督分类和非监督分类实验。实验表明,在图像分类中加入纹理特征能够弥补光谱特征的不足,有效的改善图像的分类精度；通常图像的分类精度会随着多源特征变量的增多而进一步提高；不同的纹理特征对分类精度的影响程度不同,与传统的灰度共生矩阵纹理相比,新兴的分形纹理在改进分类精度上效率更高,效果更好。最后,将研究成果用于福州市区四个时相的遥感图像分类,验证了辅以纹理的遥感图像分类的普适性,通过统计分类结果,分析了近14年福州市区土地利用／覆被变化的情况和原因。
[Abstract]:Remote sensing image classification is an important means for human to obtain effective information. How to improve the classification accuracy is an important content of remote sensing research. The traditional image classification is mostly based on spectral features, but the spatial structure features are not well mined, so the classification results are not satisfactory. As the most representative spatial structure feature, image texture has its unique advantages and great application value in improving image classification accuracy. There are many texture analysis methods, traditional statistical analysis methods are mature and stable, and have a wide application foundation. The new model analysis methods have developed rapidly, especially the application of fractal model in texture analysis has received great attention. In this paper, a rectangular area at the junction of urban and rural areas in the central and southern part of Fuzhou is selected as the typical experimental area, and the four texture features of the typical experimental area image are extracted by using the traditional gray level co-occurrence matrix model. A new fractal model is used to extract a texture feature from a typical experimental area image. On this basis, 13 feature variables, such as 1 fractal texture, 4 gray-scale co-occurrence matrix texture and 8 spectral features, are combined to construct a multi-source feature database and to carry out image supervised classification and unsupervised classification experiments in a typical experimental area supplemented by texture. Experiments show that adding texture features to image classification can make up for the deficiency of spectral features, and improve the classification accuracy of images effectively, usually the classification accuracy of images will be further improved with the increase of multi-source feature variables. The effect of different texture features on classification accuracy is different. Compared with the traditional gray-level co-occurrence matrix texture, the new fractal texture is more efficient and better in improving classification accuracy. Finally, the research results are applied to the classification of four temporal remote sensing images in Fuzhou urban area, and the universality of remote sensing image classification with texture is verified. This paper analyzes the situation and causes of land use / cover change in Fuzhou urban area in recent 14 years.
【学位授予单位】：福建师范大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP751

【参考文献】