HyMamba: A Hybrid Mamba-CNN Network for Cross-Modality HSI-LiDAR Classification
Haowei Chen, Jian Feng Luo
China West Normal University
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
Cross-modality fusion of hyperspectral imagery (HSI) and LiDAR can jointly exploit rich spectral signatures and complementary geometric cues, yet existing methods often suffer from three practical trade-offs: CNN-based models favor local textures but lack global context modeling; Transformer-based models capture long-range dependencies at a high computational cost; and recent state-space-model (SSM) approaches (e.g., Mamba) offer linear complexity but may weaken high-frequency details and lead to suboptimal cross-modal fusion. To address these issues, we propose HyMamba, a hybrid Mamba–CNN network for HSI–LiDAR classification that preserves efficient global modeling while enhancing fine-grained structures and cross-modal consistency. Specifically, we develop: (1) HFS-Mamba, which couples Mamba-based global modeling with a lightweight CNN local branch and a frequency-enhancement module to emphasize high-frequency spatial structures; (2) GLS-Mamba, which integrates bidirectional spectral scanning with local 1D convolutions and efficient spectral attention to jointly model global spectral trends and local discriminative cues; and (3) CMMS-Fusion, which performs physics-inspired asymmetric gating and similarity-driven alignment for adaptive multi-scale cross-modal fusion. Experiments on public benchmarks including Houston2013, Trento, and MUUFL demonstrate that HyMamba achieves superior accuracy with favorable computational efficiency compared with recent CNN-, Transformer-, and Mamba-based methods.
逐年被引趋势
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
Advanced Image Fusion Techniques · Remote Sensing and LiDAR Applications