Extraction Method Of Rive Width For Small Mountain Rivers Based On GF-1 Image

River width is the basic parameter of various hydrological application models,and the traditional acquisition method is mainly manual field measurement.For areas with bad natural conditions,complex terrain and large areas,artificial field measurement is dangerous,time-consuming and labor-intensive,and river width data cannot be obtained quickly and in a timely manner.Remote sensing technology has many advantages,such as large range of data,fast speed,short period,few limited conditions and so on.Through image classification and extraction technology,river width data can be obtained efficiently and quickly,and it has been widely used in hydrology and water resources,river geomorphology,ecology and other related research.The spatial resolution of panchromatic images and multispectral images provided by the China's GF-1 satellite are 2 m and 8 m respectively.The civil high-resolution optical service constellation composed of four stars with high resolution can provide a large number of high-precision images and can solve the problems of low source and poor quality of high-resolution remote sensing images in mountainous areas.In this paper,the high-resolution remote sensing image is used as the data source,and the decision tree method of object-oriented classification is used to extract the width of small rivers in the mountainous areas of the Jimai River Basin and the Darilu River Basin.For the mountain rivers,due to the complex terrain,narrow river channel,and easy to be affected by the shadow of the mountain,the extraction is difficult and the precision is low.Firstly,the multi-scale segmentation is improved by the method of blending edge features and regional features,which is based on the Canny operator.The image data layer generated by edge segmentation is added to the multi-scale segmentation,and the optimal segmentation scale is determined by segmenting the optimal segmentation scale by calculating the improved absolute mean difference ratio(MRMAS).Secondly,the optimal classification feature is obtained using the Separation Threshold Algorithm(SEa TH).Finally,the classification decision tree constructed by CART algorithm extracts the water information of small rivers in mountainous areas,and combines the river network extracted by DEM to remove the background noise to obtain the complete river boundary,and realize the automatic acquisition of river width in Matlab.The classification accuracy of small rivers in the Jimai River Basin and the Dariluo River Basin extracted by the decision tree method of object-oriented classification was evaluated.The overall accuracy is above 90%and the kappa coefficient is above 85%.The results show that the decision tree method of object-oriented classification can obtain more accurate river water information.The data of river width extracted automatically from the two basins are verified by linear regression.The linear regression coefficient of river width is 0.981 and R~2is 0.823 when the river width is more than 6 m in the Jimai River Basin.The linear regression coefficient of river width over 6 m in the Dariluo River Basin is 0.972 and R~2is 0.802.The results show that the method used in this paper can extract the width data of small rivers over 6m,and provide some scientific reference for water resources planning and management,river geomorphology research,water conservancy project construction and so on.