| Popis: |
Buildings extracted from remote sensing images play a crucial part in resource development and urban planning. With the development of convolutional neural networks for the past several years, the use of deep learning to automatically extract feature information has become a current research hotspot. However, due to the diverse shapes of buildings and complex backgrounds, the segmentation result map faces the problems of low accuracy and blurred edge contours. At the same time, computer hardware also affects the segmentation performance of the model. To resolve these problems, this article puts forward an algorithm to improve the U-Net model after many experiments. The results of the high-scoring remote sensing image test data set show that compared with the basic U-Net, the F1-Score, and IoU of the proposed algorithm are improved by 2.2% and 2.4%, respectively, and the training time of the model is shortened by 5 h. |
