基于SVM的城镇边界提取算法研究

发布时间：2018-05-24 23:07

本文选题：SVM + 区域生长法　；参考：《江苏大学》2017年硕士论文

【摘要】：中国城镇整体进入快速城市化阶段,动态监控并准确预测城镇扩张成为目前学术界研究的热点和政府决策的重点问题。美国军事气象卫星(Defence Meteorological Satellite Program,DMSP)搭载的Operational Linescan System(OLS)传感器获取的全球夜间灯光数据是进行相关大尺度城市化研究的一种有效数据源。利用DMSP/OLS夜间灯光数据提取城镇用地信息可以为大尺度城镇用地空间格局分布研究提供科学依据。而利用有效的图像识别分类算法是提取城镇用地的关键,结合改进的支持向量机分类算法提取夜间灯光数据中的城镇用地,分析长时间序列城镇用地时空格局演变特征,并模拟城镇未来扩张规模,对制定宏观政策提供理论依据和参考借鉴。本文运用不变目标区域法对长时间序列的DMSP/OLS夜间灯光数据进行校正,提出了改进的SVM分类算法,对江苏省内具有代表性的城镇进行了边界提取,分析了城镇的空间扩展格局、总体发展趋势及其中心转移情况,并利用GM(1,1)灰色模型对南京城镇未来的发展状况进行预测。论文的主要内容及其结论如下:(1)采用不变目标区域法对夜间灯光数据进行饱和校正,能够有效缓解灯光数据的饱和效应;通过连续性校正以及像元异常波动校正,降低相同年份不同传感器影像差异,减少相邻年份数据异常波动,提高不同年份数据间的连续性和可比较性。(2)提出了基于改进的SVM分类算法,提取城镇用地,并结合Landsat 8影像人工提取城镇用地进行精度评价。对比结果表明,基于改进的SVM分类算法提取的城镇用地信息在总体精度、Kappa系数和用户精度等方面都具有明显的优越性。(3)采用扇形分析、同心圆分析和城镇用地中心平均转移等方法对南京长时间序列城镇发展情况进行分析,发现南京城镇发展情形经历了先缓慢后快速的扩张过程。南京主城区在1992年城镇区域的基础向四周扩散,沿长江及南北交通走廊发展,城镇土地密度不断增大。(4)最后选用GM(1,1)预测模型对南京城镇未来的发展情况进行了预测,GM(1,1)预测模型的检验精度检验表明模型精度有很好可行性。通过计算发现,2018年与2020年南京城镇用地数量预测规模分别为2030.88km~2和2564.26 km~2。
[Abstract]:China's cities and towns as a whole enter the stage of rapid urbanization, dynamic monitoring and accurate prediction of urban expansion has become the focus of academic research and government decision-making. The global nighttime light data obtained by the Operational Linescan system OLS sensor on the US military weather satellite Defence Meteorological Satellite Program DMSPis is an effective data source for large-scale urbanization research. Using DMSP/OLS night lighting data to extract urban land information can provide scientific basis for spatial distribution of large scale urban land use. Using effective image recognition and classification algorithm is the key to extract urban land, combined with the improved support vector machine classification algorithm to extract urban land from night lighting data, and analyze the spatial and temporal pattern evolution characteristics of long-term urban land use in a long time series. It simulates the scale of urban expansion in the future and provides theoretical basis and reference for macro-policy formulation. In this paper, the invariant target region method is used to correct the DMSP/OLS night light data of long time series, and an improved SVM classification algorithm is proposed. The boundary of the representative towns in Jiangsu Province is extracted, and the spatial expansion pattern of the towns is analyzed. The general trend of development and the situation of center transfer are discussed, and the future development of Nanjing cities and towns is forecasted by using GM1 / 1) grey model. The main contents and conclusions of this paper are as follows: (1) the saturation effect of night light data can be effectively alleviated by using the invariant target region method, and the saturation effect of light data can be effectively alleviated by means of continuity correction and pixel abnormal fluctuation correction. To reduce the difference of different sensor images in the same year, to reduce the abnormal fluctuation of data in adjacent years, and to improve the continuity and comparability among the data of different years, a modified SVM classification algorithm is proposed to extract urban land. Combined with Landsat 8 image artificial extraction of urban land for accuracy evaluation. The comparison results show that the urban land information extracted based on the improved SVM classification algorithm has obvious advantages in terms of overall accuracy and user accuracy. Based on the analysis of concentric circle analysis and the average transfer of urban land center, it is found that the development of Nanjing cities and towns experienced a slow and rapid expansion process in the long time series of cities and towns in Nanjing. The foundation of the main urban area of Nanjing spread around in 1992 and developed along the Yangtze River and the north-south transportation corridor. Finally, the future development of cities and towns in Nanjing is forecasted by using the GM1) forecasting model. The accuracy test of the model shows that the model is feasible. It is found by calculation that the predicted scale of urban land use in Nanjing in 2018 and 2020 is 2030.88km~2 and 2564.26 km / m respectively.
【学位授予单位】：江苏大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP751

【参考文献】