.png){kind=icon}
.png){kind=icon}
Return
Multiple Morphology Perception Mamba for Hyperspectral Image Classification
.png){kind=icon}
OA
AI
DOI：10.1049/ipr2.70291.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The core challenge in hyperspectral image (HSI) classification lies in how to collaboratively model long-range dependencies and local structural features. This paper introduces the Multiple Morphology Perception Mamba Model (MMP-Mamba), which achieves a unified framework of global context awareness and local feature enhancement by deeply integrating dynamic morphological operations with a selective state space model. The model innovatively incorporates morphological priors by adaptively generating morphological kernels that enhance local details and suppress noise, while an attention mechanism dynamically fuses the original features with morphological gradients to focus on key regions. Additionally, a learnable gating network injects the morphologically enhanced features into the Mamba sequential modelling process, effectively compensating for local information loss caused by data serialisation. Experimental results on four benchmark datasets (Pavia University, Houston, HanChuan, and HongHu) demonstrate that MMP-Mamba significantly outperforms existing mainstream methods. Specifically, in the Pavia University scenario, the overall accuracy, average accuracy, and Kappa coefficient are improved by 2.97%, 2.91%, and 3.12%, respectively, compared to the runner-up model; in the HongHu crop sub-classification task, the model markedly enhances the ability to differentiate morphologically similar crops. While maintaining linear computational complexity, this model provides a solution for HSI classification that combines high precision with high efficiency.
Keywords:
dynamic feature fusion
hyperspectral image classification
Mamba
morphology
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
IF:
2.2
Papers:
118
Citations:
4.7K