Fine Scale Impervious Surface Extraction Method Research Based On Coupled Spectral And Spatial Features

The percentage,components,spatial distribution of impervious surface have played an important role in the urbanization process and the environmental quality assessment,accurately estimation and extraction of impervious surface information can provide an important supporting data for building "sponge" ecological city.With the imprvovment of image processing technique and spatial resolution,it provides the opportunity for extracting fine scale impervious surface information.However,due to the complexity of urban background and the variaty of impervious surface composition,accurate impervious surface extraction from high resolution image faces a severe problem of “same object with different spectrums”?”different objects with different spectrums”.High resolution images not only can provide spectral information,but also contain rich shape,texture,context information,etc.,and the proper composite applications of these characteristics can greatly improve the impervious surface extraction accuracy.In this situation,the traditional impervious extraction methods based on single spectral information for low and medium resolution image obviously can not apply to the high resolution image.Therefore,from the viewpoint of fusing image spectral and spatial characteristics,this paper fully mine suitable spectral and spatial characteristics for impervious surface,establish a classification model with a collaboratively application of spectral and spatial features,and finally realize the accurate extraction of urban scale impervious surface map.The main study contents and innovation points of this paper are as follows:(1)A VS-scale and VS-layered remote sensing image segmentation method based on the guiding of the prior knowledge was put forward.Considering that traditional single-scale segmentation method can't meet the VS-scale demand of the complex urban objects,and produce serious under and over segmentation problem,in this paper,based on the priori knowledge,we proposed a VS-scale and VS-level segmentation method.The fusion of the pixel-level image classification result and the first large-scale image segmentation result were translated into a priori knowledge,which assissted in VS-scale image segmentation process at the second level,finally implementing the adaptive application of the optimal segmentation scale parameter for different ground objects.Compared to the experimental results with the single-scale method,the proposed method can obtain more homogeneous and complete feature patches,is advantageous to the feature extraction of different categories and subsequent identification.(2)Coupled spectral and spatial features based on distance metric learning for high resolution impervious surface classification was proposed.Considering the deficiency of traditional VS-feature fusion method using vector superposition when applied in highly spectral heterogeneous urban environment,the distance metric learning technology is introduced into the impervious surface information extraction process.By learning from the labelled samples of the current image,we obtain a new distance measurement function which can reflect the spatial structure of the samples.In this metric space,the optimal spectral and spatial feature composition and application mode were mined,and further realized the high-precision impervious surface extraction.The experimental results showed that the proposed feaure fusion method can significantly improve the classification accuracy in the urban environment and provide a solid foundation for the subsequent high-precision impervious surface extraction.(3)A distributed high-resolution impervious surface extraction method based on Hadoop was proposed.Since the variaety of the components of the impervious surface and the complexity of urban background,the algorithm complexity of the fine impervious surface extraction is also high.Traditional impervious surface method running on single machine can not satisfy the application requirement of extracting impervious surface quickly and accurately at the urban scale.In this paper,an open source cloud platform Hadoop in the computer field was introduced to the application of impervious surface information extraction from remote sensing image.And for the characteristic of the remote sensing image and the algorithm of high-resolution impervious surface extraction algorithm,we designed and implemented the parallel mechanism of urban impervious surface extraction algorithm and the automatic partition and merging strategy of the remote sensing image on the Hadoop platform.Finally,by the automatically fragmental image processing,we realized the distributed and fast high-precision extraction of impervious surface.The experimental results show that the distributed method can significantly improve the extraction rate on condition that achieving a high extraction accuracy.