Remote Sensing Image Feature Extraction Method Study

Remote sensing image has great importance for military reconnaissance, precision attack and civil activities. Feature extraction is critical for the automatic recognition technology of remote sensing image, so it has good application prospect to study feature extraction methods of remote sensing image. This thesis focuses the research work mainly on the feature extraction methods of the two most important features, spectrum and texture.Firstly, the thesis introduces the basic theory and algorithms of remote sensing image feature extraction methods. The commonly used remote sensing image feature extraction methods for spectrum and texture are generalized separately. Considering the character of the remote sensing image data and the limitation of traditional PCA and KPCA methods when they are used to extract the spectrum feature of remote sensing image, a combination of the FCM and KPCA methods is used for extracting the spectrum feature. Both the theory and algorithm are studied, as well as the implementation. A comparison between the results of PCA, KPCA and MFCM+KPCA methods is given, which shows that the MFCM+KPCA method can give a much better result than other methods, especially in extracting the nonlinear information of multispectral remote sensing images.As an important complementarity of spectrum feature, texture is also a basic and important feature of remote sensing image. Based on an introduction of the basic theory and quick algorithm of wavelet and wavelet packet, the method based on wavelet packet transformation is studied to extract the texture feature of the remote sensing imagery. As the energy of texture information mainly concentrates on the intermediate and high frequency sub-bands, the characteristic vector of texture feature is constructed from the intermediate and high frequency sub-bands after the wavelet packet transformation of remote sensing imagery. Moreover, a measurement of the similarity between two different textures is improved. An image searching experiment...