高光谱成像
人工智能
遥感
上下文图像分类
计算机科学
模式识别(心理学)
计算机视觉
统计分类
特征提取
图像处理
图像分割
像素
地质学
作者
Junpei Zhang,Lingling Li,Licheng Jiao,X M Liu,Fang Liu,Wenping Ma,Shuyuan Yang,Puhua Chen
标识
DOI:10.1109/tgrs.2026.3686126
摘要
Hyperspectral image (HSI) classification plays a crucial role in fields such as remote sensing, agriculture, and environmental monitoring. However, few-shot HSI classification remains a significant challenge. In this paper, we propose PIMamba, a physics-informed HSI classification framework that integrates diffusion-inspired physical modeling into the state-space-based Mamba architecture. The proposed method models hidden-state evolution with physically guided spectral–spatial dynamics and introduces spectral physical loss and spatial physical loss to enforce physically consistent feature representations during training. In addition, a dynamic parameter adjustment (DPA) mechanism is developed to adaptively balance spectral and spatial physical constraints. Extensive experiments on four benchmark datasets (Indian Pines, University of Pavia, Salinas Scene, and Houston 2013) demonstrate that PIMamba achieves state-of-the-art classification performance under few-shot settings without relying on external annotated data or large-scale pretrained models.
科研通智能强力驱动
Strongly Powered by AbleSci AI