基于高分辨率遥感的水源地风险源提取技术研究

Research on Risk Source Extraction of Water Source Areas Based on High Resolution Remote Sensing

风险源提取是实现饮用水源地遥感监测的重要技术环节。基于高分遥感的风险源提取的技术方法体系,研发了结合面向对象和深度学习技术的风险源提取方法并进行了软件实现。以图像分割为基础通过面向对象深度学习分类提取大尺度自然分险源,再利用语义分割提取各类人工分险源,实现了不同分险源的分级提取。依托相应软件系统,以高分二号影像为主要数据源,以南京市夹江水源地为示范区开展了风险源提取试验。结果表明系统实现了包括水源地水体分布,以及水体周边建筑物、道路、农、林等多类人工、自然风险源目标的准确提取。

Risk source extraction is an important technical step for remote sensing monitoring of drinking water sources.The research group builds up a technical framework for risk source extraction from high-spatial-resolution images by combining object-based image analysis(OBIA)and deep learning techniques.On the basis of image segmentation,the largescale natural risk sources are classified by object-based deep learning classification,and various artificial risk sources are extracted by semantic segmentation,so as to realize the hierarchical extraction of different risk sources.Experiments of risk source extraction at the Jiajiang water source area of Nanjing city were carried out based on GF-2 imagery.The software system accurately extracts the distribution of water bodies,and different kinds of artificial and natural risk sources including buildings,roads,agriculture and forest lands around water bodies.