面向地表覆盖智能标报的双注意力深度交互式分割模型  

Deep Interactive Segmentation Model with Dual Attention for Land Cover Intelligent Marking

作　　者：王文科 侯东阳 周晓光[1] WANG Wenke;HOU Dongyang;ZHOU Xiaoguang(School of Geosciences and Info⁃physics,Central South University,Changsha 410083,China)

机构地区：[1]中南大学地球科学与信息物理学院,湖南长沙410083

出　　处：《测绘地理信息》2024年第4期37-41,共5页Journal of Geomatics

基　　金：国家自然科学基金(42171457,41971360)

摘　　要：针对当前人工标报存在的质量差、效率低的问题,本文提出一种面向地表覆盖智能标报的双注意力深度交互式分割模型。首先,采用磁盘编码方法模拟用户的点单击交互信息,并利用HRNet_18s网络提取地物的语义信息;然后,顺序引入通道注意力和空间注意力,强化局部重要地物的特征信息,抑制干扰背景的特征信息;最后,将特征输入OCRNet网络进一步细化,并得到分割结果。为验证本文方法的有效性,在2个公开数据集上开展了对比实验。实验结果表明,该方法在保证分割准确性的前提下,能够有效减少用户交互次数,有助于提高地表覆盖智能标报的效率。In order to solve the problems of poor quality and low efficiency of current manual marking,this paper proposes a deep interactive segmentation model with dual attention for land cover intelligent marking.Firstly,the disk coding method is used to simulate the user’s point click interaction information,and the HRNet_18s network is used to extract the semantic information of the ground objects.Then,channel attention and spatial attention are introduced to strengthen the feature information of local important objects and suppress the feature information of interference background.Finally,the features are input into the OCRNet to further refine,and the segmentation results are obtained.In order to verify the effectiveness of this method,comparative experiments were conducted on two public datasets.The experimental results show that this method can effectively reduce the number of user interactions on the same segmentation accuracy,which is helpful to improve the efficiency of intelligent land cover marking.

关 键 词：深度交互式分割模型 标报 地表覆盖更新 空间注意力 通道注意力 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置] TP391[自动化与计算机技术—控制科学与工程]