Scholarlink

<p>The growing demand for real-time, low-power image analytics in applications such as remote sensing, environmental monitoring, and precision agriculture has significantly increased the importance of hyperspectral image classification (HSIC). Hyperspectral imaging captures rich spectral–spatial information; however, the high dimensionality and substantial computational requirements of conventional deep learning models limit their suitability for deployment on edge and resource-constrained devices. This review analyzes a range of efficient and lightweight architectures, including state-space Mamba models, lightweight transformers, and convolutional neural network (CNN)–based approaches, which represent notable advancements in lightweight deep learning for HSIC. Particular emphasis is placed on key methodologies, such as residual learning, spectral–spatial feature fusion, and attention-based mechanisms. Furthermore, this review explores current trends related to performance–efficiency trade-offs, evaluation metrics, model compactness, and real-world deployment challenges relevant to HSIC. By systematically analyzing and summarizing recent lightweight deep learning models and techniques, this paper provides a comprehensive overview of the challenges, available tools, and future research directions for implementing deep learning–based hyperspectral imaging solutions in resource-limited environments.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Efficient and Lightweight Deep Learning Approaches for Hyperspectral Image Classification: A Comprehensive Review

  • 期刊论文

  • 发表日期：

    2026年4月14日

查看全文
  • Veerpal Kaur,
  • M. K. Jindal,
  • Munish Kumar

摘要

The growing demand for real-time, low-power image analytics in applications such as remote sensing, environmental monitoring, and precision agriculture has significantly increased the importance of hyperspectral image classification (HSIC). Hyperspectral imaging captures rich spectral–spatial information; however, the high dimensionality and substantial computational requirements of conventional deep learning models limit their suitability for deployment on edge and resource-constrained devices. This review analyzes a range of efficient and lightweight architectures, including state-space Mamba models, lightweight transformers, and convolutional neural network (CNN)–based approaches, which represent notable advancements in lightweight deep learning for HSIC. Particular emphasis is placed on key methodologies, such as residual learning, spectral–spatial feature fusion, and attention-based mechanisms. Furthermore, this review explores current trends related to performance–efficiency trade-offs, evaluation metrics, model compactness, and real-world deployment challenges relevant to HSIC. By systematically analyzing and summarizing recent lightweight deep learning models and techniques, this paper provides a comprehensive overview of the challenges, available tools, and future research directions for implementing deep learning–based hyperspectral imaging solutions in resource-limited environments.