Research On Water Extraction Based Vector-Data Constraint And Object-Based Image Analysis For High Resolution Remote Sensing Imagery

With the growing serious problem on water pollution in our country,it is significant to monitor the water distribution,water pollution and its temporal changes.To monitor water pollution by remote sensing and realize the dynamic monitoring of water body and water quality,we need to combine automatic water information extraction with the water-quality quantitative inversion on high-resolution remote sensing images.In this paper,to satisfy the application demands of water monitoring by high resolution remote sensing imager,we focus on researching the automatic water information extraction on high resolution remote sensing image,and the main works are following:(1)We design a method of water extraction based on object-based image analysis for high resolution remote sensing image.Firstly,we segment images by the method of hard-boundary constraint and two-stage merging to get segment primitives.Secondly,we establish the rules of water body extraction through spectral features,shape features of segment primitives to extract water.The results show that compared with the method based on pixel-based image analysis,this method can effectively solve the problem of "salt-and-pepper noise",and improve the accuracy of the extraction of water.However,there are still problems such as small water cannot be extracted accurately,and buildings,shadow are extracted as water.(2)We proposed a new method for water information extraction based on vector data constraint and object-based image analysis.Firstly,we match vector data and RS imagery.Secondly,we segment imagery by the method of hard-boundary constraint and two-stage merging to get segment primitives.Thirdly,overlay the vector data and segment primitives to establish their spatial,attributed relationship.Finally,we establish the rules of water body extraction according to the spatial and attribute relationship,and spectral features,shape features.In the end,we extract water body information by this rules.The experiments show that method we proposed can effectively solve the problem of time phase difference and registration error between water vector data and segment primitives,it can also accurately extract small water,inhibit buildings,shadow,roads.So,this method we proposed can greatly improve water extraction accuracy.(3)We design and implement remote sensing dynamic monitoring prototype system for water information extraction.The main flow of system include image segmentation,water extraction,water quality index retrieving,and main modules include image pre-processing,water product extraction,query statistics,graphics.The system integrates multiple quantitative index and the model of water quality,achieved the automatic extraction of water and quantitative inversion of pollution index,comprehensive evaluation.In the system implementation,according to the demands for multi-temporal and large data processing of remote sensing image,we design the batch product mode for water product extraction,improved the efficiency of image processing,water thematic information extraction and quantitative indexes.It provides a solid foundation for the engineering application of software systems.