Residential Area Recognition Using Texture Filtering from Hyper-spectral Remote Sensing Imagery

Hyper-spectral remote sensing imagery provides a large amount of spectral and structure information.However,these availabilities challenge the traditional spectral segmentation methods which may cause salt and pepper effect and low information extraction accuracy.In order to overcome this disadvantage,texture information is proposed into feature space.A 3D-Gabor filter is used to represent the spectral/spatial properties of hyper-spectral data.Thus multi-scale,multi-oriented texture features are extracted.And feature energy from 3D feature points is projected into subspace with PCA which can represent input data with lower dimensional feature vectors.Then the image segmentation is constructed by k-means clustering.Following these steps,the initial residential areas can be obtained,but with many deficiencies including the existence of holes and useless patches.To resolve these problems,a morphological space based method is used to dissolve these residential patches.The experiment on PHI-3 data demonstrates the utility of the algorithm for residential areas recognition.