Remote sensing extraction of river ice in the Yellow River based on NDSI and deep learning
The special geographical location and environmental factors in the Yellow River Basin lead to its complex transit characteristics.Satellite remote sensing can realize the rapid extraction of the Yellow River ice in a large range.Currently,there are two commonly used methods:remote sensing index and deep learning.In order to verify and compare the effectiveness of different methods for remote sensing monitoring and identification of the Yellow River ice,Sentinel-2 remote sensing data is used.Using normalized snow index and its improved form and U2-Net,the ice in the Ning-Meng section of the Yellow River in 2023 was extracted by remote sensing.Sentinel-2 image was extracted on February 20,2023,and GF data from February 19-21 was used as verification.The clas-sification accuracy of NDSI,MNDSI and U2-Net methods were 83.42％,87.98％and 92.01％,respectively,and the Kappa coefficient was 0.88,0.90 and 0.97,respectively.It can be seen from the classification results that the three methods have a good extraction effect on river ice.However,NDSI has poor recognition effect on other land classes such as runling,gully clearing,etc.MNDSI can distinguish different land classes from river ice,but the extraction boundary is more chaotic.Generally,U2-Net can better distinguish river ice from other land classes.
万方数据
维普
