Abstract: This paper mainly studies the wetland cover in Liao River estuary. In order to realize dynamic monitoring of large-scale wetland, the Sentinel2-L1C and AW3D30 DSM have been used as basic data and the random forest model has been used for raster classification. The results extracted by random forest model will be further processed by using digital signature of terrain and polygonal shape characteristics. The visual interpretation method has been used for verifying the accuracy of the classification, the results showed: this method has a high degree of automation, which can extract wetland coverage with less manual intervention. The extraction results achieved higher precision with the producer's and overall accuracy of 91.04% and 82.65%, respectively. The Kappa coefficient is 0.599 7, which indicated that the computer classification method this paper used and artificial visual interpretation have good consistency.

Key words: Panjin wetland, random forest, information extraction, remote sensing classification, visual interpretation