摘要
遥感图像变化检测在环境监测、灾害预警、城市规划以及土地管理等领域发挥着重要作用.为进一步提高遥感图像变化检测性能,提出了一种基于CT-SWBCE损失和可靠伪标记样本的半监督高分辨遥感图像变化检测方法.一方面,面向无标记样本提出了一种基于mIoU-OA联合度量的伪标记方法,选取高可靠的伪标记样本和有标记样本构建半监督变化检测模型,提高该模型的泛化能力.另一方面,为优化半监督变化检测模型精度,定义了CT-SWBCE损失函数,使其既能准确识别变化区域,又能处理样本不平衡问题.基于公开的遥感图像变化检测数据集LEVIR_CD和WHU_CD展开实验,验证了提出方法的有效性.实验结果表明,在LEVIR_CD数据集上的IoU和OA指标,提出方法比已有方法分别提高了1.3%~8.5%和0.09%~0.5%;在WHU_CD数据集上的IoU和OA指标,提出方法比已有方法分别提高了3.5%~24.5%和0.24%~1.11%.
Remote sensing image change detection is crucial for applications,such as environmental monitoring,disaster early warning,urban planning,and land management.To enhance the change detection performance for high-resolution remote sensing images,this paper proposs a deep semi-supervised method based on CT-SWBCE loss and reliable pseudo-labeled samples.On one hand,a joint mIoU-OA metric was designed for pseudo labeling of unlabeled samples.It was used to select highly reliable pseudo-labeled samples for constructing the semi-supervised change detection model with higher generalization ability.On the other hand,a loss function named CT-SWBCE was defined.The application of CT-SWBCE loss not only improves the accuracy of detecting changed areas,but also effectirely adresses.The experimental results on two widely used datasets show that the proposed method achieves better performance.Comparing to the state-of-art methods,the IOU and OA metrics of LEVIR_CD dataset is improved by 1.3%-8.5%and 0.09%-0.5%respectively.The IOU and OA metrics of WHU_CD dataset is improved by 3.5%-24.5%and 0.24%-1.11%respectively.
作者
杨燕
王艳宁
陈诺
刘译文
YANG Yan;WANG Yanning;CHEN Nuo;LIU Yiwen(School of Computer Science and Artificial intelligence,Liaoning Normal University,Dalian 116081,China)
出处
《辽宁师范大学学报(自然科学版)》
2025年第1期106-114,共9页
Journal of Liaoning Normal University:Natural Science Edition
基金
辽宁师范大学教师指导本科生科研项目(JSZDBKSXM2024034)。
关键词
遥感图像
变化检测
半监督学习
样本不平衡
remote sensing image
change detection
semi-supervised learning
sample imbalance
