基于U-net模型的遥感影像建筑物提取

Building extraction of remote sensing images based on U-net model

使用基于全卷积神经网络的U-net模型提取遥感影像中的建筑物,采用公开的Massachusetts建筑物数据集进行模型的训练,并通过迁移学习的思想对网络的权重进行微调,以便快速高效地训练模型,从而输出更高的精度。实验结果显示,U-net模型在准确度、召回率和F1值三项精度指标中分别达到0.95852、0.88109和0.82123,与传统方法进行对比,建筑物的轮廓更加完整,准确度提高25%以上,召回率和F1值均提高2倍以上,适用于提取场景中的建筑物区域。

The U-net model based on the fully convolutional neural network is applied to the extraction of buildings within remote sensing images.Adopting the public Massachusetts building data set to train the model,the weights are fine-tuned through the idea of transfer learning in order to make the network model training get faster and higher output accuracy.The experimental results show that the U-net model achieves 0.95852,0.88109 and 0.82123 in the three accuracy indicators of accuracy,recall and F1 score,respectively.Compared with the traditional methods,the outline of the building is more complete,and the accuracy is improved by more than 25%.The recall rate and F1 score are both increased by more than 2 times,indicating this model is suitable for extracting the building area in the scene.