首页 > , Vol. , Issue () : -
China has been experiencing rapid urbanization at an unprecedented rate, with the urban internal spatial structure changing significantly. As an inevitable byproduct, the urban village (UV), termed for the informal living space with substandard living conditions, has occurred in many newly and quickly industrialized regions and/or cities. Although UV provides plenty of living space for floating population, its poor living environment makes a great negative impact on urban landscape and public health. So it is of significant importance to obtain the spatial distribution and environmental quality information of UV in time and accurately for optimizing the urban space and improving the human settlements. Not only high-resolution remote sensing images (RSI) but also street view images (SVI) have been employed to quickly extract UV information. However, the application of combining RSI and SVI in retrieving UV information was few focused. In this study, we took the Yuexiu district in Guangzhou city as the study area, and then proposed a method of UV identification based on GF-2 high-resolution RSI and SVI released by Baidu Company. Firstly, the street space quality information was derived from SVIs using support vector machine and random forest. Then based on the pre-extraction results of GF-2 image, a multi-scale segmentation was performed based on object-based image analysis, including building instance level and block level. And 23 features were obtained including spectrum, shape, texture, building structure and scene from RSI, five indicators were obtained to measure street space quality based on SVI. Finally, random forest algorithm was applied to combine the features of the two images to identify the UV. The experimental results demonstrated that the UV recognition based on RSI has an overall accuracy of 94.5% and a Kappa coefficient of 0.58, and the overall accuracy and Kappa coefficient of UV identification based on SVI were 84.7% and 0.31, respectively. An overall accuracy of 96.1% and a Kappa coefficient of 0.67 were achieved in the fusion model of two kinds of images, which had the best performance on UV recognition. Compared with the model based on high-resolution RSI or SVI, the user accuracy was improved by 15.5% and 47.4% respectively, while the producer accuracy only showed slight improvement. Street space quality features, textural feature, structural feature and shape feature played import role in UV recognition based on fusion model of RSI and SVI. The five indicators that measure street space quality based on SVI contributed 31.6% of the feature importance in the fusion model. The information provided by RSI from the bird view and the SVI from the human perspective could complement each other, which could create a more outstanding feature space and reduce the misclassification phenomenon of UV. The key of this method was to integrate the information provided by SVI into the process of UV extraction based on high-resolution RSI, so as to obtain a more stable and reliable classification result of UV in Yuexiu district. Multi-source data fusion is an important method to improve the ability of RSI, and other data would be collected to further enrich the existing coupling methods and technical system. This paper revealed that the fusion of high-resolution RSI and SVI in the feature level could improve the recognition accuracy of UV, and the extracted UV distribution data can be used in urban planning and other studies related to urban development. The information in SVI could be integrated into high-resolution RSI and other data source to assist the identification of informal living space such as UV. It is feasible to retrieve more accurate UV information through combining RSI and SVI.