An Exploration of Frequency-Based Multi-Scale Multi-Decomposition Techniques for Effective Urban Mapping

尽管地理信息科学取得了重大进展，但将高分辨率数字图像准确地分类为城市土地类别仍然是一个挑战。空间要素的高频出现是城市制图的主要限制。传统的分类器使用基于单像素值的光谱信息，而没有考虑图像的空间属性。当前的地理空间方法也是无效的，因为它们不能在多个尺度上提取不同方向的不同空间特征。小波理论可以为描述复杂的特征提供很多机会，因为它在每个尺度上提供了四个不同的空间信息。研究人员将开发一个基于小波的框架和运算算法，通过在多个尺度上组合不同方向的不同空间信息来准确识别城市类别。为了进行比较，本项目还将开发和评估其他地理空间技术--分形等值线、分形方差函数、空隙、空间自相关和G指数。这项研究将对地理现象的尺度相关性、数据挖掘、模式识别和要素的自相似性产生影响。该项目的结果和发现将为建模和理解城市系统提供新的机会。这项研究还将揭示不同地理空间技术的优势和劣势，并有助于改进自动化分析程序。

Despite significant advances in geographic information science, accurately classifying high-resolution digital images into urban land categories remains a challenge. The high frequency appearance of spatial features is the major limitation in urban mapping. The conventional classifiers employ spectral information based on single pixel values without considering the images' spatial properties. Current geospatial approaches are also ineffective, since they are incapable of extracting different spatial features in different directions at multiple scales. Wavelet theory could open up a lot of opportunities for characterizing complex features, since it provides four different pieces of spatial information at each scale. The investigator will develop a wavelet based framework and operational algorithms to identify urban classes accurately by combining different spatial information in different direction at multiple scales. For comparison purposes, this project will also develop and evaluate other geospatial techniques - fractal isarithm, fractal variogram, lacunarity, spatial autocorrelation, and G index.This research will have implications for scale dependency of geographic phenomena, data mining, pattern recognition, and self-similarity of features. Results and findings from the project will provide new opportunities for modeling and understanding urban systems. This study will also reveal the advantages and weaknesses of different geospatial techniques, and help improve automated analysis procedures.