A Method For Extracting Built-up Areas From High-resolution Remote Sensing Images Inspired By Visual Cognition

With the acceleration of China’s urbanization process,construction land is constantly expanding.The geospatial information such as the location,scope and distribution of built-up areas has important application value for land use planning,urbanization monitoring,disaster assessment,and updating of basic geographic databases.The current rapid development of high-resolution ground observation technology provides an effective data source for fine monitoring of built-up areas.The automatic acquisition of built-up areas information from satellite images has attracted more and more attention from domestic and foreign scholars.It can effectively solve the traditional time-consuming and labor-intensive problem of manual interpretation.However,the built-up areas in high-resolution remote sensing images have the characteristics of spectral heterogeneity,complex texture,and wide coverage,which brings many challenges to the automatic extraction of built-up areas.Inspired by the cognitive mechanism of the human visual system,this paper recognizes the built-up areas in high-resolution images from two different perspectives: visual attention and perceptual organization,and builds a related built-up area extraction model to achieve automatic extraction of built-up areas.The main research work of this article is as follows:(1)Inspired by the mechanism of visual attention,the visual saliency model of built-up areas in high-resolution satellite images was studied.The multi-scale wavelet transform was used to measure the texture saliency of built-up areas,a saliency map of built-areas is generated by fusing different levels of high-frequency information,and the extraction of built-up areas was further completed through adaptive thresholding.(2)Inspired by the mechanism of visual perception,the perceptual organization model of built-areas in high-resolution satellite images was studied.First,the input images are segmented by superpixels,and the spatial characteristics of superpixels are modeled by using the spatial variation function.Then,the spatial relationship between superpixels is described according to the Gestalt rules of perceptual organization,which are incorporated into saliency modeling of superpixels.Finally,the extraction of built-up areas is completed by saliency map thresholding and post-processing.In this paper,ZY-3,QuickBird,WorldView-2 and Google Earth high-resolution satellite images were experimentally evaluated.The performance of the experimental results was analyzed through the precision rate,recall rate and F-measure,and good experimental results were obtained.Further,the performance of the method in this paper is compared with the classic Pan Tex algorithm.Compared with the Pan Tex algorithm,our approach shows great advantages.