Research On Urban Water Body Information Extraction From Highresolution Remote Sensing Image Based On Multi-feature Fusion

Water is the source of our life and a vital natural resource in our daily life.Water system is one of the important elements of China’s geographical conditions.It is of great significance to master its spatial planning layout and change cycle frequency for disaster monitoring,energy reconnaissance and environmental defense.With the continuous development of earth detection technology by remote sensing satellites,it provides the main data support for water extraction and dynamic monitoring.Automatic classification and extraction of water system information from remote sensing images has become a current research hotspot.The object-oriented image analysis method provides a new development direction for the segmentation and classification of high-resolution remote sensing images.The object result after image segmentation is no longer only using the pixel of spectral information,but also can fully integrate the spectral,texture,space,and multi-feature information.Using this technology,a multi-feature fusion deterministic rule classification combined with nearest neighbor classification method is proposed for automatic drainage system extraction,and the effectiveness of the method is verified by using World View-2satellite data as experimental object.The data source and research area of this paper are World View-2 high-resolution remote sensing images of some urban areas of Zhengdong New District,Zhengzhou city.Firstly,the remote sensing images are preprocessed and filtered,and then the regional multi-scale segmentation and spectral difference segmentation are used to segment the images.After that,the major features in the images are analyzed and statistically.Finally,the method of multi-feature fusion is used to establish rules,and a classification method combining the determination rule classification with the nearest neighbor classification is proposed to classify and optimize the local objects.The object oriented image analysis method is universal in the segmentation and classification of high-resolution remote sensing images because it has the characteristics of accurate,rapid and efficient extraction of ground object information compared with the traditional manual visual interpretation method.In recent years,using this technology to extract different ground object information has become a research hotspot.The main research contents of this paper are as follows:(1)A series of pre-processing operations for the original data of high-resolution remote sensing images are studied.In this chapter,ENVI software is used as the operating platform to conduct radiometric calibration,atmospheric correction,image fusion and image clipping for the original data of images successively.In the image fusion processing,Brovey fusion,GS fusion and NNDifuse fusion were used for image fusion after radiation correction,and then local fusion effect pictures were selected for qualitative and quantitative evaluation.Finally,GS fusion was the most suitable fusion method for this study area.(2)different image filtering on the effects of the high resolution remote sensing image smoothing filtering method is through the use of 6 kinds of contrast experiment was carried out on the image,respectively,each group set a group of 4 to 6 different parameters,the selection of the best parameters of various methods,then 6 kinds of optimal parameter setting of filter method,the selection of local area for target detail comparison,Finally,it is verified that the guided filtering method can effectively remove noise and protect edges,enhance the internal information of images,and effectively improve the accuracy of subsequent image segmentation and classification.(3)Two different image segmentation methods are tested,and the pixel-based mean shift segmentation method and the region-based classification network evolutionary segmentation method are respectively used for comparison experiments.The method has the characteristics of multi-scale,choose the optimal parameters of the two methods of multi-scale Settings,by selecting multiple parts of the segmentation effect for qualitative and quantitative analysis,finally choose multi-scale segmentation based on region and,with the method of spectral differences as the segmentation results,this paper lay a foundation for the next step of image classification.(4)The statistics of spectrum,texture and thematic index in the image of the study area are analyzed.Based on matlab software as a platform,deterministic classification rules are constructed according to fuzzy classification,appropriate membership function is used,and classification optimization is carried out in combination with nearest neighbor classification in supervised classification as the experimental results of this paper.In order to compare the superiority of this method,the accuracy of this method is compared with that of the traditional method,maximum likelihood method.Through the comparative analysis of experiments,the producer accuracy,user accuracy,overall accuracy and Kappa coefficient of the method combining deterministic rules with nearest neighbor classification are 98.05%,98.58%,96.46% and 0.9553,respectively.It is obviously better than the traditional supervised classification method based on pixels.For the study of river system information extraction in the experimental area,the object-oriented multi-feature fusion method has higher accuracy than the traditional classification method.