Abstract: In this paper, VGG16_BN Deep Learning network algorithm is used to classify the different ground objects over the coastal wetland in Yancheng of Jiangsu province, based on the high spatial resolution remote sensing images captured by GF-2 satellite. The results of this classification algorithm are compared with those of three other algorithms, including VGG16, Support Vector Machine (SVM) and Back Propagation (BP) neural network models. The overall classification accuracy and Kappa coefficient obtained from the VGG16_BN Deep Learning network algorithm are respectively 99.32% and 0.99, which are significantly higher than other algorithms. In addition, based on local visualization analysis, VGG16_BN network algorithm can effectively maintain the regional consistency of homogeneous features over a large area and avoid the salt-and-pepper noise. The continuity and boundary of line feature extraction such as road and river are relatively complete and clear.