融合高度信息的遥感图像语义分割网络  

Semantic segmentation network for remote sensing images with fused height information

作　　者：高梁 钱育蓉[1,3,4] 刘慧[2,3,4] GAO Liang;QIAN Yu-rong;LIU Hui(School of Software,Xinjiang University,Urumqi 830046,China;College of Information Science and Engineering,Xinjiang University,Urumqi 830046,China;Key Laboratory of Signal Detection and Processing,Xinjiang University,Urumqi 830046,China;Key Laboratory of Software Engineering,Xinjiang University,Urumqi 830046,China)

机构地区：[1]新疆大学软件学院,新疆乌鲁木齐830046 [2]新疆大学信息科学与工程学院,新疆乌鲁木齐830046 [3]新疆大学信号检测与处理重点实验室,新疆乌鲁木齐830046 [4]新疆大学软件工程重点实验室,新疆乌鲁木齐830046

出　　处：《计算机工程与设计》2023年第8期2417-2424,共8页Computer Engineering and Design

基　　金：国家自然科学基金项目(61966035、61562086、U1803261);新疆维吾尔自治区研究生创新基金项目(XJ2021G062)。

摘　　要：针对遥感图像分割精度低的问题,提出一种融合高度信息的遥感图像语义分割网络。将光谱信息和高度信息进行融合,从不同的数据中提取地物的特征信息。针对卷积网络受固定感受野的限制,无法有效获取特征图上下文信息的问题,将Transformer应用到网络中建模特征图的全局相关性,增强网络的特征提取能力。为有效利用融合特征图中的信息,提出基于自注意力的解码器。通过在韦兴根和波茨坦数据集上进行实验,验证了所提网络与其它先进网络相比具有竞争力。Aiming at the problem of low accuracy of remote sensing image segmentation,a semantic segmentation network of remote sensing image fused with high information was proposed.The network fused spectral and height information to extract feature information of the features from the different data.To address the problem that traditional convolutional networks are limited by fixed perceptual fields and cannot effectively obtain feature map contextual information,Transformer was applied to the network to model the global relevance of feature maps and enhance the feature extraction capability of the network.A self-attention-based decoder was proposed to effectively utilize the information in the fused feature map.By conducting experiments on the Vaihingen and Potsdam datasets,the proposed network is shown to be competitive with other state-of-the-art networks.

关 键 词：遥感图像 语义分割 自注意力 上下文信息 TRANSFORMER 高度信息 特征提取 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]