【摘要】 文章提出一种基于深度学习的遥感图像语义分割方法，用于分割不同类别的土地区域。构建以编解码结构为主干的遥感图像语义分割模型，利用通道注意力机制来强化对分割任务有效的特征，提升分类精度；提出基于残差学习框架的多尺度特征融合算法，使得低层特征的局部细节信息和高层特征的语义信息互补，实现遥感图像的精细化分割。以中国南方某地区为例，采用文中提出的语义分割模型，实现了地表覆盖分类自动化，达到90.88%的分割准确率。实验结果表明：除水体分割外，该模型对其他类别的分割均优于原始U-Net,尤其是对道路和建筑的分割，精度提升明显；具有较高的分割准确率和较好的泛化能力，能够用于土地覆盖感遥感图像分割。更多还原

【Abstract】 In this paper, a deep learning based semantic segmentation method for remote sensing image is proposed, which is used to segment different types of land areas. Firstly, a remote sensing image semantic segmentation model based on encoding and decoding structure is constructed, and the channel attention mechanism is used to strengthen the effective features of segmentation tasks, thus improving the classification accuracy. Next, a multi-scale feature fusion algorithm based on residual learning framework is proposed, which makes the local details of low-level features and semantic information of high-level features complement each other, and finally realize the fine segmentation of remote sensing images. Taking a certain area in southern China as an example, the automation of land cover classification is realized by using the proposed semantic segmentation model, and the segmentation accuracy reaches 90.88%. The experimental results show that the model is better than the original U-Net for other categories except water segmentation, especially for road and building segmentation, and the accuracy is improved significantly. The model has high segmentation accuracy and good generalization ability, which is suitable for land cover remote sensing image segmentation.更多还原