High resolution urban monitoring using neural network and transform algorithms

The advent of new high spatial resolution optical satellite imagery has greatly increased our ability to monitor land cover from space. Satellite observations are carried out regularly and continuously and provide a great deal of information on land cover over large areas. High spatial resolution imagery makes it possible to overcome the “mixed-pixel” problem inherent in more moderate resolution satellite sensors. At the same time, high-resolution images present a new challenge over other satellite systems since a relatively large amount of data must be analyzed, processed, and classified in order to characterize land cover features and to produce classification maps. Actually, in spite of the great potential of remote sensing as a source of information on land cover and the long history of research devoted to the extraction of land cover information from remotely sensed imagery, many problems have been encountered, and the accuracy of land cover maps derived from remotely sensed imagery has often been viewed as too low for operational users. This study focuses on high resolution urban monitoring using Neural Network (NN) analyses for land cover classification and change detection, and Fast Fourier Transform (FFT) evaluations of wavenumber spectra to characterize the spatial scales of land cover features. The contributions of the present work include: classification and change detection for urban areas using NN algorithms and multi-temporal very high resolution multi-spectral images (QuickBird, Digital Globe Co.); development and implementation of neural networks apt to classify a variety of multi-spectral images of cities arbitrarily located in the world; use of different wavenumber spectra produced by two-dimensional FFTs to understand the origin of significant features in the images of different urban environments subject to the subsequent classification; optimization of the neural net topology to classify urban environments, to produce thematic maps, and to analyze the urbanization processes. This work can considered as a first step in demonstrating how NN and FFT algorithms can contribute to the development of Image Information Mining (IMM) in Earth Observation.