Graph-Guided Change Detection and State Space Model Captioning with Semantic Uncertainty Estimation
Siraj T M, Reji J, Raji Susan Mathew
Indian Institute of Science Education and Research Thiruvananthapuram
阅读操作
确认中在文库中上传 PDF 后可生成中文音频讲解。
摘要与影响
Automated change detection and captioning in satellite imagery play an important role in urban monitoring, infrastructure assessment, and land-use analysis. However, many existing approaches do not explicitly quantify uncertainty, which limits their practical use in decision-critical scenarios. In this work, we present a unified framework that integrates SemanticGraphCD for change detection, a state space model (SSM)-based approach for caption generation, and a Semantic-Weighted Sentence Entropy (SWSE) mechanism for uncertainty quantification. The SemanticGraphCD module integrates graphbased representations with semantic feature learning, enabling better identification of meaningful infrastructure and land-use changes. For captioning, we use a spatial difference-aware SSM (SD-SSM) to capture spatial variations and a temporal traversing SSM (TT-SSM) to model interactions across bi-temporal features. SWSE further emphasizes domain-relevant terms and provides interpretable uncertainty estimates. Experimental results on the LEVIR-CC and LEVIR-MCI datasets demonstrate improved captioning performance compared to existing SSM-based approaches, along with meaningful uncertainty quantification for urban monitoring applications.
逐年被引趋势
暂无年度引用数据
关键指标
同类平均 = 1
同领域 · 同年份 · 同类型
Google Scholar 与 OpenAlex 的被引统计范围不同,数值存在差异属正常。
AI 辅助阅读
依据:摘要
可就本文提问;依据不足时会说明。
学术脉络
学科主题
计算机 / AIRemote-Sensing Image Classification
Remote Sensing and Land Use · Automated Road and Building Extraction