Research On Residential Area Extraction From High Resolution Remotely Sensed Imagery

Abstract:Residential area is the centre site of human beings’daily life and activity. Residential area is composed by dense buildings, vegetation inside and traffic route around in high resolution remotely sensed image. Extracting the residential areas accurately and fast have become crucial for several domains, for example, urbanization construction, digital city, urban planning, land use and geographic information system updating. From this standpoint, this dissertation is trying to propose several algorithms for residential area extraction from high resolution remotely sensed images. Concretely, main contents of this dissertation include the following three parts:(1) Residential area extraction using edge density features from high-resolution remote sensed imageryUsing the edge density features difference between the residential and non-residential areas. In this method, we first smooth the original images by mean shift filter; and then implement the edge detection followed by fitting them as several straight lines, afterwards, based on the edge distribution in the image, we form the spatial voting matrix and extract the residential area by threshold segmentation algorithm. The proposed method is complete automatic which avoid the artificial influence, it increased the accuracy and efficiency of residential extraction.(2) Automatic urban area extraction using gabor filter from high resolution remote sensed imageryResidential area not only have rich edge features but also dense corner features, the corners have obvious gray level gradient change and curvature change, therefore, the Gabor filter take great response at the location of corner. We propose another residential extraction algorithm based on the Gabor feature from high resolution remotely sensed image. In the algorithm, we first obtain the filtering response images by Gabor filters group at various central frequencies and orientations, and then detect the gabor features followed by feature optimization, finally, we realize the residential area extraction by forming the spatial voting matrix with the Gaussian function. This method is unsupervised, compared with the method upon, it has better results.(3) Town and country residential area supervised classification from high resolution remote sensing imagery.The two proposed residential area extraction method in front have no distinguish between the two kinds of residential areas. So we can’t get the improvement information of the urbanization. Both the edge density features and the Gabor features have different distribution information in the urban and rural residential areas. So we could use this message to distinguish them, we first design five kinds of rule to reflect their difference, and then create large numbers of training samples to learn the five rules, finally, use the test samples to statistic the accuracy of the rules. The developed method was just a primary explore, it still have many shortcomings to improve, but it still have originality.