Research On Change Detection In Remote Sensing Images Based On 2D Gabor Wavelets

The progress in the domain of remote sensing (RS) enables more and more access to remote sensing data acquired by space-born and air-born platforms. With the help of computer science technology and spatial information technology, these remote sensing data find wide applications such as meteorological observation, detection of marine and terrestrial resources, and environmental protection. There are quite a large number of RS images acquired over the same target area, while at different time, which exhibits the challenges to extract useful information and also the opportunity of deep understanding of evolution in these areas over time. Consequently, change detection has received many attentions. Based on two or more RS images and by using appropriate algorithms, change detection can extract the difference among these images and offer the information about the change in the target area, which is therefore widely used in various tasks varying from resource detection,military reconnaissance to disaster management.Most of the change detection techniques directly utilize the pixel values to obtain useful information. However, the pixel value is not always effective for the extraction of difference.In addition, it generally requires specialized knowledge in performing change detection,which could inhibits its applications for a person without knowing too much about the details of change detection method. It is important to develop effective and efficient change detection methods, and offer user-friendly interface for its practical applications. As such, in this thesis, we propose a change detection method based on multi-channel Gabor features.We incorporate this method with ArcGIS, contributing to a user-friendly interface and providing intuitive demonstration of the detected changes for users. Specifically, the contributions in this thesis are presented as follows.1. In literature, two-dimensional (2D) Gabor wavelet transformation has demonstrated a good performance in texture analysis and target recognition. The features extracted by 2D Gabor wavelet transformation can significantly reduce the negative effective of the scale and the angle in the RS images. A change detection method is proposed based on 2D Gabor wavelets transformation. Firstly, by using the 2D Gabor wavelet transformation, we represent each pixel in RS images (for change detection) with the eigenvalues, which are obtained according to multiple scale and multiple directions. On this basis, the resulting feature for each pixel in the difference map can be obtained. Finally, we can obtain the change class and non-changed class according to the difference map by utilizing K-means algorithm and semi-negative matrix factorization. The optical RS images and synthetic aperture radar (SAR) images are used to test the proposed method. The experimental results demonstrated the effectiveness of our method.2. Inspired by the concept of "digital earth", we intend to demonstrate the evolution of the Earth and human activities in a digital manner. In this thesis,we develop a system to demonstrate the results of the proposed method by employing ArcGIS Engine/IDL integration development. In addition to the demonstration of the results of the proposed method,this system can provide satellite images for the observation of Earth surface,three-dimensional terrain of an area, and online weather service.