摘要
以大数据为驱动的深度学习图像信息挖掘技术是影像解译的研究热点之一,也是自然资源调查监测的重要手段,但存在耕地非农化图斑识别率低、人工工作量大、模型泛化性弱等问题。该文提出引入双注意力机制的CD-DLinkNet变化检测算法,并以地理国情普查成果为基础数据构建耕地非农化变化检测样本数据集,较好地解决了复杂应用场景下耕地非农化变化检测模型泛化性弱、误识别率高的问题。实验证明:针对耕地的非农化变化检测,引入双注意力机制的CD-DLinkNet变化检测算法的对象查全率可达82.52%,对象查准率可达77.02%,很好地满足了遥感监测项目的生产需要,能够有效辅助非农化图斑的变化发现和提取。
Image information mining technology based on deep learning is one of the research hotspots of imagery interpretation and an important means of natural resource investigation and monitoring.Nevertheless,there are some problems,such as low rate of identification,large amount of manual workload and low generalization performance.In this paper,a change detection algorithm based on dual attention mechanism CD-DLinkNet is proposed.The sample data set of non-agriculturalization of cultivated land was constructed based on the results of geographical conditions monitoring.The problems of weak generalization and high misidentification performance in complex application scenarios were solved betterly.Experimental results represent that:for the detection of non-agriculturalization of cultivated land,the object recall of the dual attention mechanism CD-DLinkNet algorithm can reach 82.52%and the object precision can reach 77.02%,meeting the production needs of remote sensing monitoring projects and effectively assisting the detection and extraction of non-agriculturalization of cultivated land.
作者
张永洪
韩红涛
杨晓锋
贺帅
Zhang Yonghong;Han Hongtao;Yang Xiaofeng;He Shuai(The First Institute of Photogrammetry and Remote Sensing,Ministry of Natural Resources,Xi’an 710054,Shaanxi,China;School of Information and Communications Engineering,Xi’an Jiaotong University,Xi’an 710049,Shaanxi,China;Xi’an Huawei Technology Co.,Ltd.,Xi’an 710300,Shaanxi,China)
出处
《计算机应用与软件》
北大核心
2025年第11期229-236,249,共9页
Computer Applications and Software
基金
陕西测绘地理信息局科技创新项目(SCK2021-09,SCK2019-3)。
关键词
深度学习
影像解译
非农化
调查监测
注意力机制
Deep learning
Imagery interpretation
Non-agriculturalization
Investigation and monitoring
Attention mechanism
