变更检测
计算机科学
人工智能
小波
二进制数
质量(理念)
数据挖掘
模式识别(心理学)
对偶(语法数字)
噪音(视频)
小波变换
计算机视觉
机器学习
可靠性(半导体)
特征提取
相变
语义学(计算机科学)
作者
Xiaopeng Chen,Quan Zhou,Baoyu Ge,Maosheng Wei,Yiming Yan
摘要
Precise modeling of spatio-temporal surface evolution is pivotal to high-resolution remote sensing change detection. However, existing methods are largely confined to binary change localization, where discrimination is frequently overlooked and high-quality annotation is prohibitive. To address these limitations, a semi-supervised framework integrating hybrid dual attention and reliable pseudo-labeling is proposed within a Teacher–Student architecture. Multi-scale features are enhanced by a wavelet interaction module, while pseudo-label quality is refined via uncertainty-guided optimization. Superior performance on semantic change detection datasets is demonstrated by experimental results.
科研通智能强力驱动
Strongly Powered by AbleSci AI