Scholarlink

<p>To tackle spectral distortion and spatial detail loss in remote sensing image fusion, this paper proposes a fusion method for panchromatic and multispectral images based on IHS transform and NSST decomposition. However, when the grayscale distribution of the panchromatic image is highly concentrated and has low contrast, the fusion quality declines. To overcome this limitation, an image enhancement technique combining the guided filter (GF) and sigmoid function is introduced in the preprocessing stage of multispectral images. The guided filter effectively preserves edge details, while the sigmoid function enhances contrast by adjusting grayscale values near the center of the distribution. For low-frequency components, a fusion strategy integrating regional energy, regional gradient, and guided filtering (RE–RG–GF) is proposed, ensuring that both local energy and gradient information are retained while maintaining edge details. For high-frequency components, an adaptive pulse-coupled neural network (PA-PCNN) fusion method is applied. Experimental results on two public datasets validate the effectiveness of the proposed approach, showing an increase in standard deviation by 83.49 and 52.69% over the suboptimal results, along with an average gradient increasing by 78.49 and 60.64%, respectively, across seven evaluation metrics.</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

Fusion of Panchromatic and Multispectral Images Using Guided Filtering and Sigmoid Function Enhancement

  • 期刊论文

  • 发表日期：

    2025年9月12日

查看全文
  • Jianfei Gao,
  • Yun Fu,
  • Jiangnan Cui,
  • Ming Li,
  • Chunxiao Han,
  • Lun Jiang,
  • Yongliang Li

摘要

To tackle spectral distortion and spatial detail loss in remote sensing image fusion, this paper proposes a fusion method for panchromatic and multispectral images based on IHS transform and NSST decomposition. However, when the grayscale distribution of the panchromatic image is highly concentrated and has low contrast, the fusion quality declines. To overcome this limitation, an image enhancement technique combining the guided filter (GF) and sigmoid function is introduced in the preprocessing stage of multispectral images. The guided filter effectively preserves edge details, while the sigmoid function enhances contrast by adjusting grayscale values near the center of the distribution. For low-frequency components, a fusion strategy integrating regional energy, regional gradient, and guided filtering (RE–RG–GF) is proposed, ensuring that both local energy and gradient information are retained while maintaining edge details. For high-frequency components, an adaptive pulse-coupled neural network (PA-PCNN) fusion method is applied. Experimental results on two public datasets validate the effectiveness of the proposed approach, showing an increase in standard deviation by 83.49 and 52.69% over the suboptimal results, along with an average gradient increasing by 78.49 and 60.64%, respectively, across seven evaluation metrics.