| 英文摘要 |
The Waisanding sandbar has suffered from continuous beach erosion and is gradually disappearing. It is necessary and urgent to solve this problem under the homeland protection policy. We have provided a satellite-image-based solution for this problem. We found that the northern end of the Waisanding sandbar has been reduced by 2.1km in the past 15 years, and the southern end has been significantly shortened and rotated counterclockwise. Since 2011, the southern tip has been submerged in the water, and gradually disappeared at the average tide level. Three kinds of convolutional neural network were established to identify the sea and land near the Waisanding sandbar. VGG16UNET with an identification accuracy of higher than 90% was examined to be superior to the other two Waisanding sandbar models. VGG16UNET can effectively distinguish the sea and land areas covered by thin clouds and is applicable to the land areas that are not in the training group. |