Zero-shot Mars scene classification (ZSMSC) methods aim to generalize knowledge from seen categories to unseen ones by leveraging semantic information, enabling robust recognition under the challenging and unpredictable Martian environment. Existing approaches typically employ convolutional neural networks or Transformers-based architectures for feature extraction, which either suffer from limited receptive fields or high computational costs. To address these limitations, we propose a Mamba-based ZSMSC framework (M2SC). M2SC integrates Vision Mamba and Mamba to extract visual and semantic features, respectively, and employs a cross-modal feature fusion module to align the two modalities. Experimental results on the ZSMars dataset demonstrate that M2SC achieves competitive classification performance across various semantic vectors and seen/unseen ratios, while substantially reducing model complexity and computational cost.
A MAMBA-BASED ZERO-SHOT MARS SCENE CLASSIFICATION FRAMEWORK
Gamba P.
2025-01-01
Abstract
Zero-shot Mars scene classification (ZSMSC) methods aim to generalize knowledge from seen categories to unseen ones by leveraging semantic information, enabling robust recognition under the challenging and unpredictable Martian environment. Existing approaches typically employ convolutional neural networks or Transformers-based architectures for feature extraction, which either suffer from limited receptive fields or high computational costs. To address these limitations, we propose a Mamba-based ZSMSC framework (M2SC). M2SC integrates Vision Mamba and Mamba to extract visual and semantic features, respectively, and employs a cross-modal feature fusion module to align the two modalities. Experimental results on the ZSMars dataset demonstrate that M2SC achieves competitive classification performance across various semantic vectors and seen/unseen ratios, while substantially reducing model complexity and computational cost.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
