Extraction Of Urban Residential Area Based On ZY-3 Remote Sensing Images

As a key component of urban structure,residential area has an important guiding role in urban planning and construction.As the resolution of remote sensing images gradually increases,the feature information contained in the images becomes more and more abundant,which makes the traditional supervised classification algorithms based on manually designed features face challenges in dealing with complex feature information.Deep learning,as a method with powerful learning ability,is gradually becoming a way to solve this problem.How to use deep learning methods for accurate extraction of urban residential area has become a hot issue in current research.This study provides a new solution for urban residential area extraction from high-resolution remote sensing images,which has significant research significance and application value.The main research contents include the following aspects:(1)Based on a deep learning model,an urban residential area extraction method was constructed to take into account the unbalanced samples,in view of the high complexity of urban residential area features and the unbalanced number of urban residential and non-residential area samples.The method uses self-attentive mechanism and Focal Loss function to build a network model,and combines two post-processing modules of edge regularization and elimination of extraction results and neighborhood gaps to realize urban residential area extraction.The performance of the method is tested in Beijing with the F1 value of 89.66%,the Io U of 81.26%,the precision of89.05%,and the recall of 90.29%.(2)The optimization method combining multiple sources of data was proposed to address the phenomenon of omission in the extraction of urban residential area by deep learning methods.The method introduced urban building data and urban residential area POI data,and combines with KNN algorithm to realize the complement of the extraction results of the deep learning method.Taking Beijing as an example to test the accuracy of the method,the F1 value is 90.87%,the Io U is 82.56%,the precision is91.01%,and the recall is 89.83%,and the indexes are improved compared with the above results.(3)Thirty-six key cities in China were selected for urban residential area extraction using the method proposed in this study,and the extraction results were evaluated for accuracy as well as landscape pattern index analysis and correlation analysis.The results show that the method has high extraction accuracy and good adaptability,and the analysis of the residential area layout of each city combined with the economic model leads to the following conclusions:(1)the proportion of residential area in economically developed cities is relatively low,which indicates that these cities have higher efficiency in land use and reflects the tightness of land resources in large cities;(2)there is a strong correlation between the residential area and the local population and GDP,which indicates that the residential area is basically related to the population.There is a strong correlation between the area of residential land and the number of local population and GDP,indicating that the area of residential land basically matches the number of population;(3)There is a dissonance between the area of residential land and the population and the amount of investment in real estate development in some cities,which implies that there may be a problem of excessive real estate development in these cities.