利用多通道循环注意力机制网络提取遥感图像建筑物

Building Extraction from Remote Sensing Images Using Multi-channel Recurrent Attention Mechanism Network

从高分辨率遥感图像建筑物中获取建筑物区域对于城市规划和地理测绘具有重要意义。近年来卷积神经网络在建筑物提取领域取得了较好效果;但现有方法存在特征提取不充分和忽略不同特征之间相关性等问题。为解决高分辨率遥感图像的建筑物提取问题,提出了一种新颖的多通道循环注意力机制网络。利用多尺度通道注意力机制扩大卷积核感受野,提取丰富的建筑物区域特征信息;通过空间金字塔循环模块建立不同特征之间的相互关系,提升模型的非线性学习能力;采用多通道特征融合模块融合不同特征信息,并获取建筑物区域提取结果。实验结果表明,所构造模型在高分辨率建筑物提取数据集上获取了较高精度,证明了该方法的有效性和实用性。

Obtaining building region from high-resolution remote sensing images is of great significance for urban planning and geographical surveying and mapping.In recent years,convolutional neural networks have achieved better results in the field of building extraction.However,the existing methods have the problems of insufficient feature extraction and ignoring the correlation between different features.To solve the problems of building extraction in high-resolution remote sensing images,we proposed a novel multi-channel recurrent attention mechanism network.We used the multi-scale channel attention mechanism to expand the convolution kernel receptive field,and extracted rich building region feature information.We used the spatial pyramid recurrent module to establish the relationship between different features,and improved the nonlinear learning ability of model.We used the multi-channel feature fusion module to fuse different feature information,and obtained the building area extraction results.The experimental result shows that the constructed model achieves high accuracy on the high-resolution building extraction dataset,which can prove the effectiveness and practicability of the proposed method.