Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

HSIMamba: Hyperpsectral Imaging Efficient Feature Learning with Bidirectional State Space for Classification

Judy X Yang, Jun Zhou, Jing Wang, Hui Tian, Alan Wee Chung Liew

TL;DR

The paper tackles the challenge of hyperspectral image classification under high dimensionality and computational constraints. It introduces HSIMamba, a bidirectional state-space backbone built around HyperBiRNet, integrated with a spatial processing block to fuse spectral and spatial cues while avoiding the heavy self-attention of Transformers. Through extensive experiments on Houston 2013, Indian Pines, and Pavia University, HSIMamba outperforms state-of-the-art CNNs and Vision Transformers like SpectralFormer, while exhibiting substantially lower memory usage and computation. The results demonstrate HSIMamba as a practical, efficient solution for real-time remote sensing analysis on resource-limited platforms, potentially broadening access to advanced hyperspectral analytics.

Abstract

Classifying hyperspectral images is a difficult task in remote sensing, due to their complex high-dimensional data. To address this challenge, we propose HSIMamba, a novel framework that uses bidirectional reversed convolutional neural network pathways to extract spectral features more efficiently. Additionally, it incorporates a specialized block for spatial analysis. Our approach combines the operational efficiency of CNNs with the dynamic feature extraction capability of attention mechanisms found in Transformers. However, it avoids the associated high computational demands. HSIMamba is designed to process data bidirectionally, significantly enhancing the extraction of spectral features and integrating them with spatial information for comprehensive analysis. This approach improves classification accuracy beyond current benchmarks and addresses computational inefficiencies encountered with advanced models like Transformers. HSIMamba were tested against three widely recognized datasets Houston 2013, Indian Pines, and Pavia University and demonstrated exceptional performance, surpassing existing state-of-the-art models in HSI classification. This method highlights the methodological innovation of HSIMamba and its practical implications, which are particularly valuable in contexts where computational resources are limited. HSIMamba redefines the standards of efficiency and accuracy in HSI classification, thereby enhancing the capabilities of remote sensing applications. Hyperspectral imaging has become a crucial tool for environmental surveillance, agriculture, and other critical areas that require detailed analysis of the Earth surface. Please see our code in HSIMamba for more details.

HSIMamba: Hyperpsectral Imaging Efficient Feature Learning with Bidirectional State Space for Classification

TL;DR

The paper tackles the challenge of hyperspectral image classification under high dimensionality and computational constraints. It introduces HSIMamba, a bidirectional state-space backbone built around HyperBiRNet, integrated with a spatial processing block to fuse spectral and spatial cues while avoiding the heavy self-attention of Transformers. Through extensive experiments on Houston 2013, Indian Pines, and Pavia University, HSIMamba outperforms state-of-the-art CNNs and Vision Transformers like SpectralFormer, while exhibiting substantially lower memory usage and computation. The results demonstrate HSIMamba as a practical, efficient solution for real-time remote sensing analysis on resource-limited platforms, potentially broadening access to advanced hyperspectral analytics.

Abstract

Classifying hyperspectral images is a difficult task in remote sensing, due to their complex high-dimensional data. To address this challenge, we propose HSIMamba, a novel framework that uses bidirectional reversed convolutional neural network pathways to extract spectral features more efficiently. Additionally, it incorporates a specialized block for spatial analysis. Our approach combines the operational efficiency of CNNs with the dynamic feature extraction capability of attention mechanisms found in Transformers. However, it avoids the associated high computational demands. HSIMamba is designed to process data bidirectionally, significantly enhancing the extraction of spectral features and integrating them with spatial information for comprehensive analysis. This approach improves classification accuracy beyond current benchmarks and addresses computational inefficiencies encountered with advanced models like Transformers. HSIMamba were tested against three widely recognized datasets Houston 2013, Indian Pines, and Pavia University and demonstrated exceptional performance, surpassing existing state-of-the-art models in HSI classification. This method highlights the methodological innovation of HSIMamba and its practical implications, which are particularly valuable in contexts where computational resources are limited. HSIMamba redefines the standards of efficiency and accuracy in HSI classification, thereby enhancing the capabilities of remote sensing applications. Hyperspectral imaging has become a crucial tool for environmental surveillance, agriculture, and other critical areas that require detailed analysis of the Earth surface. Please see our code in HSIMamba for more details.
Paper Structure (37 sections, 6 equations, 2 figures, 9 tables, 1 algorithm)

This paper contains 37 sections, 6 equations, 2 figures, 9 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. CNNs in Hyperspectral Image Classification
  4. Transformers and Mamba in Hyperspectral Image Classification
  5. The Emergence of Mamba in Computer Vision
  6. Methodology
  7. HSIMamba Model Preliminaries
  8. Forward and Backward Spectral Dependencies
  9. Discrete Transformations for Spectral Sequencing
  10. Proposed Method Overview and Architecture Description
  11. HSIMamba Block
  12. Efficiency Analysis
  13. Parameter Efficiency
  14. Floating Point Operations (FLOPs)
  15. Runtime Efficiency
  16. ...and 22 more sections

Figures (2)

  • Figure 1: The architectural overview of the Proposed HSIMamba model. The framework consists of four main components: (A) A hyperspectral image patch with dimensions $p \times p \times \text{CH}$; (B) The HSIMamba Block; (C) The Spatial Processing Block; (D) The classifier. The process begins by extracting patches that serve as input to the HSI-Vim block. This block includes a spatial processing stage that precedes the unique forward and backward operations, offering a tailored solution to the challenges inherent in hyperspectral data. The design of this model differs from traditional models used in text sequence modeling and RGB image token sequence modeling, enhancing the classification accuracy through its specialized approach.
  • Figure 2: OA Performance Comparison of Different patch size based on UH2013