Scholarlink

<p>Deep learning (DL) has recently emerged as a transformative approach for atmospheric correction (AC) in satellite remote sensing, addressing the limitations of traditional image-based and physics-based models. This paper presents a comprehensive and structured systematic review of DL-based AC methods, aiming to bridge the existing knowledge gap by consolidating insights across 30 peer-reviewed studies. The reviewed methods are categorised into physics-aware and physics-agnostic frameworks based on whether they require auxiliary information such as atmospheric and geometric parameters and top of atmospheric reflectance. In addition, the paper compiles commonly used datasets, including satellite, simulated, and field-measured data and evaluation metrics. It also identifies critical DL model assumptions, such as fixed atmospheric profiles, clear-sky conditions, and Lambertian surface approximations. The paper identifies key challenges and proposes future directions by synthesising current research. To our knowledge, this is the first systematic review of DL-based AC. This work is a foundational resource for researchers and practitioners, guiding the development of robust, interpretable, and scalable DL-based AC models.</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

A Systematic Review on Deep Learning for Atmospheric Correction of Satellite Images

  • 期刊论文

  • 发表日期：

    2025年10月9日

查看全文
  • Maitrik Shah,
  • Mehul S. Raval,
  • Srikrishnan Divakaran

摘要

Deep learning (DL) has recently emerged as a transformative approach for atmospheric correction (AC) in satellite remote sensing, addressing the limitations of traditional image-based and physics-based models. This paper presents a comprehensive and structured systematic review of DL-based AC methods, aiming to bridge the existing knowledge gap by consolidating insights across 30 peer-reviewed studies. The reviewed methods are categorised into physics-aware and physics-agnostic frameworks based on whether they require auxiliary information such as atmospheric and geometric parameters and top of atmospheric reflectance. In addition, the paper compiles commonly used datasets, including satellite, simulated, and field-measured data and evaluation metrics. It also identifies critical DL model assumptions, such as fixed atmospheric profiles, clear-sky conditions, and Lambertian surface approximations. The paper identifies key challenges and proposes future directions by synthesising current research. To our knowledge, this is the first systematic review of DL-based AC. This work is a foundational resource for researchers and practitioners, guiding the development of robust, interpretable, and scalable DL-based AC models.