精准高效地从高分辨率遥感影像中提取建筑物信息对国土规划和地图制图意义重大，近年来基于卷积神经网络进行建筑物信息提取已经取得了很大的进展，然而在处理高分辨率遥感影像时仍存在影像的高级语义特征利用不够充分，难以获得细节丰富高精度分割影像的问题。文章针对以上问题提出了一种用于建筑物自动提取的深度学习网络结构空洞空间与通道感知网络(Atrous Space and Channel Perception Network，ASCP-Net)。该模型将空洞空间金子塔池化(Atrous Spatial Pyramid Pooling，ASPP)和空间与通道注意力(Spatial and Channel Attention，SCA)模块融入到编码器-解码器结构中，通过ASPP模块来捕获和聚合多尺度上下文信息，采用SCA模块选择性增强特定位置和通道中更有用的信息，并将高低层特征信息输入解码网络完成建筑物信息的高效提取。在WHU建筑数据集(WHU Building Dataset)上进行实验，结果表明:文章提出的方法总体精度和F1 评分分别达到了 97。4%和 94。6%，相比其他模型能够获得更清晰的建筑物边界，尤其对图像边缘不完整建筑的提取效果较好，有效提升了建筑物提取的精度和完整性。
Building Extraction Model Based on ASPP and Dual Attention Mechanism
Accurate and efficient building information extraction from high-resolution remote sensing images is of great significance for land planning and mapping.In recent years,great progress has been made in building information extraction based on convolutional neural networks.However,there still exists the problems that the advanced semantic features of the images are not sufficiently utilized and it is difficult to obtain detailed and high-precision segmentation images when processing high-resolution remote sensing images.To solve the above problems,a deep learning network architecture,Atrous Space and Channel Perception Network(ASCP-Net),is proposed for automatic building extraction.The Atrous Spatial Pyramid Pooling(ASPP)and Spatial and channel-wise Attention(SCA)modules are integrated into the encoder-decoder structure.Multi-scale context information is captured and aggregated through the ASPP module.Meanwhile,the SCA module is used to selectively enhance the more useful information in specific locations and channels,and the high and low-layer feature information is input into the decoding network to achieve efficient building information extraction.Experiments on the WHU Building Dataset show that,for the overall accuracy and F1 score,the proposed method reachese 97.4%and 94.6%respectively,and can obtain clearer building boundaries compared with other models,especially for the extraction of incomplete buildings at image edges,and effectively improving the accuracy and integrity of building extraction.
NETL
NSTL
万方数据
